# Prosper AI Consulting

> Outcome Driven AI

Full-content export for LLMs and agents. Generated live from the EmDash CMS. For the link-only summary see /llms.txt.

# Terms

Source: https://www.prosperconsulting.ai/terms

## Terms of use.

These terms govern your use of this website, our publications, and our public agent surface. By using the site you agree to them. Last updated 25 April 2026.

## What these terms cover

These terms apply to the public-facing parts of prosperconsulting.ai. They do not replace the engagement contract that governs paid consulting work; that contract is signed separately.

### The website and Insights

The pages, articles, navigation, and embedded media you see when browsing prosperconsulting.ai (and any subdomains we operate).

### Public agent surface

Our llms.txt and llms-full.txt endpoints, our public Model Context Protocol (MCP) server, and any other machine-readable index we publish for AI agents and automated tools.

### Publications and downloads

Briefings, reports, slide packs, and other documents made available through the site, including any gated downloads (publications you receive after submitting a request form).

## Acceptable use

Use the site as a reasonable visitor would. Most things are obviously fine; we call out the few that are not.

### Do

Read, share, link to, and quote our content with attribution. Submit enquiries and request publications via the forms. Use our public agent surface (llms.txt, public MCP) for legitimate research and automation.

### Don't

Don't try to circumvent rate limits, bypass authentication on private surfaces, scrape at a rate that degrades the service, attempt to extract personal data of other visitors, or use the site to send unsolicited bulk communications.

### Automated access

Bots and AI agents are welcome at our public surfaces. Identify yourself with a User-Agent string, respect robots.txt and the rate limits documented in the public MCP about endpoint, and prefer the structured surfaces (llms.txt, MCP) over scraping HTML.

## Intellectual property

Most content is ours. Some is licensed. The rules below tell you which is which.

### Our content

Articles, framework diagrams, methodology summaries, and downloadable publications are owned by Prosper AI Consulting unless explicitly attributed to a third party. You may share, quote, and link with attribution; you may not republish a substantial portion of a publication without written permission.

### Client work and case studies

Where we reference client work, the client retains rights to their data, processes, and any custom artefacts created during the engagement. Case studies are published with the named client's review and consent.

### Open-source components

Software components published under an open-source licence (for example skills published in our marketplace plugins) carry their own licence terms. Those terms govern use of those components.

## Disclaimer and liability

Standard limitation of liability. Read the engagement contract for the scope of liability that applies to a paid engagement; the cap below is the website-only cap.

### Provided as-is

The website, our publications, and our public agent surface are provided on an as-is basis without warranties of any kind, express or implied, including merchantability, fitness for a particular purpose, accuracy, or non-infringement.

### Not professional advice

Articles, briefings, and other public content are general information, not consulting advice for your specific situation. Don't rely on them as a substitute for an engagement scoped to your context.

### Limitation of liability

To the extent permitted by law, our total liability arising from your use of this website is limited to GBP 100. This cap does not apply where liability cannot be excluded by law (for example death, personal injury caused by negligence, or fraud). For paid consulting work, the engagement contract applies.

## Common questions

### Which law governs these terms?

These terms are governed by the laws of England and Wales. Any dispute that cannot be resolved between us is subject to the exclusive jurisdiction of the courts of England and Wales.

### Can you change these terms?

Yes. We update terms when our processes, providers, or regulations change. The "last updated" date at the top of the page tells you when the current version landed. For material changes we will note the change in our next Insights post and, where you have an active enquiry or engagement with us, by email.

### What if I see something that looks wrong on the site?

Email austin.anderson@prosperconsulting.ai. Site issues, broken links, factual errors, accessibility problems, and security concerns are all welcome. We acknowledge within one working day under normal conditions.

### Can my agent submit forms or use the public MCP?

Yes. The public MCP server (advertised on llms.txt and at /api/mcp) is the supported way for an agent to interact with us. Forms are also fair game; identify yourself in the User-Agent and don't pretend to be a human if you are not.

### Where is the registered company information?

Our company registration number, registered office address, and ICO registration number are listed in the footer of every page on this site.

## Questions about these terms

Email austin.anderson@prosperconsulting.ai. We respond within one working day under normal conditions.

Calls to action: [Contact us](/contact).

---

# Privacy

Source: https://www.prosperconsulting.ai/privacy

## Privacy.

How we collect, use, and protect your personal data. UK GDPR and Data Protection Act 2018. Last updated 25 April 2026.

## Who we are

Prosper AI Consulting is a UK consultancy. Our company registration, registered office, and ICO registration number are listed in the footer of every page on this site. For data protection enquiries email austin.anderson@prosperconsulting.ai with "privacy" in the subject line.

### Data controller

We are the data controller for personal data we collect about you through this site, when you contact us, or when we provide consulting services to your organisation.

### Contact for privacy

Email austin.anderson@prosperconsulting.ai with "privacy" in the subject line. We respond within one working day under normal conditions and within UK GDPR statutory windows for formal requests.

### Supervisory authority

If we cannot resolve your concern you have the right to complain to the Information Commissioner's Office (ICO) at ico.org.uk.

## What we collect

We collect the minimum personal data needed to do what you have asked us to do, plus standard technical data needed to run a secure web service.

### From you directly

Name, work email, organisation, role, and the message or notes you choose to include when you submit a contact form, request a publication, or book a discovery call.

### From your browser automatically

IP address, user-agent string, referrer, pages visited, and approximate geolocation derived from IP. Held as part of standard request logs and aggregated analytics.

### From our delivery systems

When we send you a publication or response by email, our email provider (currently Resend) records delivery, open, and click events. We use these to confirm successful delivery and to investigate failures.

## Why we collect it (lawful basis)

Different data is held under different lawful bases. We are explicit about which applies so you can exercise the rights that match.

### Legitimate interests

Operating and securing the site, responding to enquiries you initiate, and maintaining records of consulting work we have done for you. Balanced against your privacy rights.

### Consent

Marketing emails (you tick the opt-in box at the point of submission). You can withdraw consent at any time by clicking the unsubscribe link in any marketing email or emailing us.

### Contract

Where we hold personal data necessary to perform a consulting engagement under a contract with your organisation.

## Your rights

Under UK GDPR you have a defined set of rights over your personal data. We respect all of them and will respond to a valid request within the statutory window (one calendar month, extendable in limited circumstances).

### Access and portability

You can ask us for a copy of the personal data we hold about you (a Subject Access Request, or SAR). We provide it in a structured, commonly used, machine-readable format where reasonable.

### Correction and erasure

You can ask us to correct inaccurate data or to erase data we no longer have a lawful basis to hold. Erasure is not absolute (for example we may need to retain records of a consulting engagement for tax or audit reasons) but we will explain any limit.

### Objection and restriction

You can object to processing under legitimate interests, restrict processing while a dispute is resolved, or withdraw consent for processing that relies on it. You can also opt out of marketing at any time.

## Common privacy questions

### How long do you keep my data?

Enquiry and contact-form submissions: 24 months from last contact. Publication-request records: 24 months from delivery. Consulting engagement records: 7 years from end of engagement (HMRC and audit). Email delivery logs from our provider: rolling 30 days. We delete data sooner on a valid erasure request unless we are required to keep it.

### Do you share data with third parties?

Only the processors strictly needed to run the service. Currently Cloudflare (hosting and CDN), Resend (transactional and marketing email delivery), and Google Analytics (aggregated analytics with IP anonymisation). Each processor has a written data processing agreement with us. We do not sell personal data and we do not pass it to advertising networks.

### Where is my data processed?

Primarily in the UK and EEA. Some processors operate in the United States under approved transfer mechanisms (UK International Data Transfer Addendum and EU Standard Contractual Clauses). Email delivery via Resend may route through US infrastructure. Cloudflare's edge network may serve cached static assets from the geographically nearest region.

### Do you use cookies?

We use a small number of essential cookies for security and form integrity. We use Google Analytics with IP anonymisation enabled and Google Signals disabled. We do not use advertising cookies, third-party trackers, or cross-site profiling. A cookie banner offering an analytics opt-out will land before public launch.

### How do I make a Subject Access Request?

Email austin.anderson@prosperconsulting.ai with "SAR" in the subject line. Tell us your name, the email address you used to contact us, and (if relevant) the engagement or publication you are asking about. We will acknowledge within 5 working days and respond in full within one calendar month.

### What about AI agents reading this site?

We publish a public agent-readable surface (llms.txt and a public MCP server) so AI agents can discover what we do without needing to scrape. Agent traffic is logged the same way as human traffic. If an agent submits a contact request on behalf of a user we treat the request as the user's, and the same retention and rights rules apply.

## Questions or a formal request

Email austin.anderson@prosperconsulting.ai with "privacy" in the subject line. We respond within one working day under normal conditions.

Calls to action: [Contact us](/contact).

---

# Home

Source: https://www.prosperconsulting.ai/

## Turn AI hype into outcomes

Most businesses can see the potential of AI. Far fewer know where it will create real value, what to prioritise, or how to make it stick. We help mid-market leadership teams turn AI interest into practical, measurable business outcomes.

Calls to action: [Book a discovery call](/contact).

### From strategy to adoption

Move from AI ambition to implementation with clear support from initial strategy through to rollout, adoption and value realisation.

### Built for mid-market businesses

Get hands-on AI strategy, implementation and adoption support without big consultancy complexity or unnecessary overhead.

### Vendor neutral, delivery focused

Choose, shape and implement the right solution for your business, without being pushed towards a particular platform or vendor.

### Focused on value creation

Improve the workflows, data, skills and governance needed to turn AI adoption into measurable business outcomes.

## From AI uncertainty to business outcomes

We help leadership teams turn broad AI interest into a clear, practical and commercially grounded programme of action.

### Find where AI can create value

Identify where AI can improve productivity, quality, service, speed, cost, margin or growth, and where it is likely to be a distraction.

### Prioritise what matters

Assess opportunities by value, feasibility, risk and adoption effort, so you focus on the work most likely to deliver a return.

### Shape the right solution

Decide whether to buy, build, configure or improve existing tools, based on what your business actually needs.

### Test before you commit

Run focused experiments with clear success measures, so you know what to adopt, improve or stop.

### Implement and embed

Turn proven ideas into working changes across workflows, systems, roles, behaviours and management routines.

### Build the capability to keep moving

Develop the skills, governance and review rhythms needed to keep improving as AI and your market evolve.

## Practical support at every stage of AI adoption

Whether you are starting from uncertainty, running early experiments, or ready to embed AI more deeply into the organisation, we help you move forward with structure and pace.

### AI Discovery & Strategy

For leadership teams that need to understand where AI can create value, what to prioritise and how to turn interest into a practical roadmap.

### Outcome-Driven Technology Adoption

Our structured framework for turning AI opportunities into business cases, experiments, adoption plans and measurable outcomes.

### AI Navigator

Independent support to help you identify, assess, select and implement the right AI technologies, vendors and solutions.

### Fractional AI Leadership

Senior AI, technology and adoption expertise to provide direction, momentum and governance without the cost or complexity of a full in-house team.

### AI Capability Academy

Practical training and capability-building to help leaders, managers and teams understand AI, use it responsibly and apply it to the work that matters.

### Tech-Centric Transformation

Transformation support for organisations ready to redesign workflows, roles, data and operating models around technology and AI.

### Strategy through to implementation

You get joined-up support from initial direction and prioritisation through to technology selection, rollout, adoption and value realisation.

### Practical, common-sense delivery

We focus on what will work in your organisation, with your people, your constraints and your commercial priorities.

### Built around outcomes

Every recommendation, experiment and implementation decision is linked to business value, adoption effort and measurable improvement.

### Capability that stays with you

We work with your teams, not around them, so the knowledge, confidence and operating rhythm remain in the business.

---

# Contact

Source: https://www.prosperconsulting.ai/contact


---

# About

Source: https://www.prosperconsulting.ai/about

## We bridge AI's potential and mid-market reality.

Operators, technologists, and change experts. We do AI consulting the way it should have been done from the start: evidence-led, vendor-neutral, grounded in your operations.

## AI is not like other technologies.

It opens new possibilities for automation, efficiency, and innovation. It also comes with real constraints, especially inside the tight budgets and delivery pressure you already work under.

### Evidence over hype

Every recommendation is driven by a clear business case, focused on return on investment, and rooted in a deep understanding of your business and its needs. We won't propose an AI solution unless it's feasible, affordable, and genuinely useful.

### Bridge, not lecture

We bridge the gap between the vast potential of AI and the practical realities of mid-market operations. Open-minded about what's possible, grounded in making it work for you.

### Vendor neutral

We don't sell software. We evaluate the market on your behalf and help you pick the right tool, including the option of building nothing.

## Thirty minutes, no slides, no sales pitch.

Calls to action: [Book a discovery call](/contact).

---

# Team

Source: https://www.prosperconsulting.ai/about

- Adrian Tripp, Partner: Experienced entrepreneur and management consultant with over two decades in business strategy, innovation, and technology-driven growth. Adrian founded and exited three media-tech businesses, the last to Informa Plc. He has spent the past seven years consulting with the London Stock Exchange Group and Umbra Capital Partners, and leads the business strategy practice at Prosper AI Consulting.
- Paul Bratcher, Partner: Senior technology and transformation leader with over twenty years in board-level roles at Howdens Joinery, Rexel UK, Travis Perkins, and Screwfix. Paul has led AI initiatives at board level since 2019, currently shaping business cases, developing AI-enabled solutions, and keeping people and culture at the centre of change.
- Ben Ferns, Partner: Agentic AI specialist, digital consultant, and technology entrepreneur. Ben designs and implements agentic AI systems that autonomously manage complex business processes. He also serves as fractional CTO to clients building secure and scalable technology foundations, and founded and exited a digital agency in 2024.
- Amira Kohler, Head of Change: Change and people management expert with over 30 years leading organisational transformation, HR strategy, and IT implementation for FT100 companies, SMEs, and agile start-ups. Amira heads the change management practice, covering stakeholder engagement, organisational design, AI literacy, and skills development.
- Nick Russell, Principal Technology Architect: Seasoned technology leader with over two decades in IT architecture, digital transformation, and enterprise systems design. Nick has held senior roles at Accenture, KPMG, Willis Towers Watson, and Cynergy Bank, and leads the technology and architecture practice, aligning infrastructure with strategic business goals.

---

# The Hidden Risk in PE Portfolios

Source: https://www.prosperconsulting.ai/insights/the-hidden-risk-in-pe-portfolios

> The core thesis: most PE portfolios were built for a world of integrated, deterministic software. They are now misaligned with the modular, probabilistic architecture that AI is creating, and the implications for value creation are significant. --- The Foundational Mismatch Most PE portfolio…

*The core thesis: most PE portfolios were built for a world of integrated, deterministic software. They are now misaligned with the modular, probabilistic architecture that AI is creating, and the implications for value creation are significant.*

---

## The Foundational Mismatch

Most PE portfolio companies were architected for a world of large, integrated, deterministic software systems, built on traditional if-then logic, multi-year implementation cycles, and top-down technology decisions. But AI is creating a new reality of modular, probabilistic systems where end users can build sophisticated solutions in weeks using interconnected AI agents and tools. Companies built even four years ago are fundamentally misaligned with this new architecture of value creation.

## The Fast Follower Imperative

Survival now depends on "Fast Follower" capability. The organisational muscle to continuously spot emerging AI tools, evaluate ROI in weeks not months, implement at speed, and capture value before competitors catch up. This requires treating technology adoption as a continuous, rapid cycle (3 to 6 month waves) rather than periodic, major transformations (18 to 24 month projects). The winners will perpetually surf the wave of AI innovation, always one step ahead.

## The Three-Front Competitive Assault

Portfolio companies face unprecedented threats from:

- **Their own vendors becoming competitors.** Thomson Reuters now sells AI legal services directly to corporate counsel, bypassing law firms.
- **Large players moving down-market.** Big 4 using AI to make small company audits economically viable.
- **AI-native disruptors.** DeepL and OpenAI making traditional translation agency business models obsolete.

## The Leadership and Organisational Blind Spot

Current management teams have two critical gaps:

- **Awareness gap:** they do not understand AI's current capabilities or trajectory.
- **Capability gap:** they have never built organisations designed for rapid, distributed technology adoption.

The new requirement: every team member thinking daily about AI integration, not as a support tool but as a fundamental capability multiplier. This demands a complete inversion from centralised, IT-led technology adoption to distributed, business-led experimentation. A move from tech-supported to tech-centric.

## The New Investment Criteria

PE firms must now evaluate:

- Does leadership have a well-developed 5-year view of AI's impact on their sector?
- Is AI strategy integrated into (not alongside) the core business strategy?
- Is every department actively experimenting with AI tools?
- Can the organisation move from pilot to scale in months, not years?
- Is there infrastructure and scaffolding for rapid business case evaluation and ROI measurement?

---

## The Implications for Private Equity: A Fundamental Shift in Value Creation

### The PE Value Creation Model is in Question

The traditional PE playbook, operational improvements, geographic expansion, bolt-on acquisitions, and multiple arbitrage, assumes relatively stable competitive dynamics over a three to seven year hold period. But when AI can obsolete a market leader in as little as 24 months, these strategies become insufficient.

PE firms must now underwrite not just current performance but transformation velocity. A company with 30% EBITDA margins but no Fast Follower capability may prove less valuable than one with 20% margins and demonstrated AI agility.

### Due Diligence Must Evaluate New Dimensions

Current due diligence focuses on financial performance, market position, and management track record. This misses the critical question: *can this organisation transform fast enough to prosper?*

PE firms need new assessment frameworks to evaluate:

- Management's technology awareness and learning velocity
- Organisational readiness for distributed innovation
- Speed of decision-making and implementation
- Cultural resistance to continuous change
- Infrastructure and organisational capability for rapid technology experimentation and scaling

### Portfolio Company Support Must Transform

PE operating partners traditionally focused on cost reduction, sales excellence, and M&A integration. Now they must become transformation accelerators, providing:

- AI opportunity scanning across the portfolio
- Shared learning and best practices from AI implementations
- Access to transformation expertise and capability assessments
- Board-level pressure for continuous AI and technology innovation
- Cross-portfolio AI talent and resource sharing

The leading PE firms are already moving in exactly this direction, and their approaches offer a blueprint for what "transformation accelerator" means in practice.

There are already clear signs of this shift. BCG argues that PE firms need new AI playbooks that can be piloted in one portfolio company and then translated across others, with the emphasis moving from simply deploying tools to reshaping operating models and selectively reinventing value propositions.

Vista Equity Partners has launched what it calls its "Agentic AI Factory", a systematic programme to move every portfolio company from AI experimentation to AI-native operations. Vista anticipates deploying 5 to 10 AI agents per user across its holdings, a scale that would represent a complete architectural shift in how its portfolio organisations operate.

### Exit Strategy Implications

The buyer universe is shifting. Strategic buyers are increasingly likely to pay premiums for demonstrated AI transformation capability rather than just market position. The evidence is already in the market. PitchBook's Q4 2025 analysis shows AI-enabled companies commanding a 22% valuation premium over non-AI peers at the pre-money stage.

Portfolio companies that have successfully transitioned from tech-supported to tech-centric operations, with metrics proving Fast Follower velocity, are positioned to capture these premiums. Conversely, those unable to demonstrate this transformation may face valuation discounts as buyers recognise the transformation debt.

---

## The Strategic Imperative for PE

The old valuation logic is starting to break down. In an AI-driven market, buyers will increasingly distinguish between businesses that have simply adopted tools and those that have built the capability to keep adapting. The premium will not come from superficial AI activity, but from credible evidence that a company can move faster, reshape workflows, strengthen customer value, and respond to market change without relying on slow, top-down transformation programmes. Companies that can adopt, exploit, and evolve beyond new technologies more quickly should become more productive and more profitable over time, because they can turn technological change into advantage faster than their competitors.

That shifts the valuation lens. The question is no longer just how well a business performs today, but how confident a buyer can be in its ability to stay competitive tomorrow. Companies with strong Fast Follower capability should command more confidence, while those carrying transformation debt may increasingly face a discount.

For PE firms, this means changing both how they invest and how they create value. At investment stage, the assessment must go beyond traditional metrics to test whether leadership understands AI's trajectory in the sector, whether the organisation has muscle memory for rapid change, and whether the culture can support distributed innovation. During ownership, value creation must focus less on one-off implementations and more on building the organisational capability for continuous AI adoption. Firms that fail to apply this lens risk both overpaying for businesses that look strong today but lack adaptability, and failing to build the capabilities future buyers will expect.

This is not about adding AI features to existing businesses. It is about rewiring organisations so they can keep responding to continuous technological change, and recognising which companies have the potential for that rewiring before writing the cheque.

---

**Sources:** PitchBook Q4 2025 AI Valuation Analysis; Bain and Company Global Private Equity Report 2025.

---

*Adrian Tripp is a Partner at Prosper AI Consulting, leading the business strategy practice and advising PE firms on AI-era value creation, due diligence, and portfolio transformation.*

---

*Article | Prosper AI Consulting, UK*

---

# The Comfort of Being Equally Behind

Source: https://www.prosperconsulting.ai/insights/the-comfort-of-being-equally-behind

> Why the winners of the AI era will not be the organisations that picked the best tools. --- On 31 March, Jack Dorsey and Sequoia's Roelof Botha published an essay, From Hierarchy to Intelligence, arguing that corporate hierarchy is an obsolete information routing system. Their claim is that AI can…

*Why the winners of the AI era will not be the organisations that picked the best tools.*

---

On 31 March, Jack Dorsey and Sequoia's Roelof Botha published an essay, *From Hierarchy to Intelligence*, arguing that corporate hierarchy is an obsolete information routing system. Their claim is that AI can now perform the coordination functions that management layers have always existed to provide, making the entire structure redundant. The essay is compelling, the direction is real, and it is already being cited in boardrooms as evidence that the future of organisational design has arrived.

It has not arrived. Not for most organisations.

Dorsey is describing what a finished building looks like. The architecture is coherent, the logic is sound. What the essay does not contain is anything about the ground it needs to be built on. For most organisations, that ground is not prepared. The foundations are not laid. And nobody is talking about what it takes to prepare them.

Block, with millions of daily transactions flowing through Cash App and Square, has spent years building those foundations. Clean, machine-readable data across its entire operation. A culture that defaults to technology when it faces any problem. A leadership team that has already made peace with a fundamentally different conception of what the business is.

Most businesses are not starting from that place. And their leadership teams, if they are honest, have not yet looked hard at what that distance actually means for them.

---

## The Comfort of Being Equally Behind

Most leadership teams look around their sector, see that everyone is roughly as far along on AI as they are, and take quiet comfort in that. It is the most expensive mistake they will make this decade.

The organisations that will beat them are not in that comparison group. In our experience, they never are. They stopped asking, "How do we use technology to support the way we work?" and started asking a different question entirely: "How do we build an organisation in which technology is central to how people think, decide and operate?"

That is not a technology upgrade. It is a different conception of what the organisation is. Different in its structure, its culture, its economics and its speed. We see it in the work we do with our clients every day. And the leaders building it are not announcing it.

The gap between a tech-supported organisation and a tech-centric one does not close with a new tool or a productivity initiative. It closes with a leadership team that has genuinely reimagined what their organisation needs to become and then built the operating system to get there.

Most exec teams have not had that conversation yet. I fear many never will.

---

## The Mistake Hiding in Plain Sight

Walk into most businesses today and you will find some version of the same pattern. A few AI tools in use, mostly productivity focused. A handful of enthusiastic individuals who have gone deep on the technology. A leadership team that is broadly supportive but has not fundamentally changed how it thinks about the business. And a growing anxiety that something important is happening that the organisation is not quite keeping pace with.

The typical response is to accelerate implementation. Find more tools. Launch a customer-facing AI initiative. Hire someone with AI in their job title. Move faster.

This is where most organisations compound the original mistake.

Customer-facing AI, deployed before the internal operations are understood and optimised, tends to produce visible failures at exactly the moment the business needs to be building confidence. Complex solutions, built before the organisation has developed the judgement to evaluate them, consume resource and generate noise without generating value. And all of it happens without addressing the actual problem, which is not a shortage of technology. It is a shortage of organisational capability to adopt technology well.

The businesses getting traction are not the ones moving fastest on implementation. They are the ones that started somewhere less glamorous: their own internal operations, their own processes, their own people. They banked the easy wins first, because easy wins prove the business case, build internal confidence, and teach the organisation something about what AI can and cannot do in their specific context. That knowledge compounds. The businesses skipping that phase are not moving faster. They are accumulating a different kind of debt.

---

## The Debt Nobody is Measuring

Technology debt is a concept most leaders understand in its conventional form: you are behind on implementation, your systems are outdated, you need to catch up. That is a real problem, but it is the visible one. It shows up in IT audits and vendor conversations.

The debt that is silently compounding in most organisations right now is harder to see and significantly more dangerous. It is the absence of an operating model for adoption itself: a repeatable way to identify opportunities, test them quickly, measure value, build capability and scale what works.

No clear mechanism for identifying where technology creates genuine value. No cultural norm that makes technology the first response to a problem rather than an afterthought. No shared leadership conviction about what the organisation needs to become or why.

That debt does not appear on a balance sheet. It does not show up in a board report. It shows up when a competitor who has been quietly building that capability starts moving faster than you across multiple fronts simultaneously, and you discover that your response time is structural rather than circumstantial. You are not slow because you made a bad decision. You are slow because you never built the machine that makes fast decisions and rapid execution possible.

By the time that gap is visible, it has usually been widening for two or three years.

---

## What the Leadership Conversation Actually Needs to Be

In our experience working with leadership teams across professional services, financial services, and PE-backed businesses, the organisations that make genuine progress share one characteristic that has nothing to do with technology. Their leadership teams have sat together and done something uncomfortable: they have looked honestly at what their competitive environment is becoming, imagined the organisations that will thrive in it, and then looked back at their own business with clear eyes.

That exercise tends to produce a specific kind of discomfort. Not panic, but recognition. The realisation that the question they have been asking, "How do we adopt AI?", is not the right question. The right question is: "What kind of organisation do we need to become, and what will it take to build it?"

Those are different conversations with different implications. The first is a technology project. The second is a leadership mandate. And until a board is having the second conversation, the first one will continue to produce patchy results regardless of how much is spent on tools.

What follows from that conversation is not a technology strategy. It is an operating model for how the organisation identifies, tests, measures and scales technology adoption as a continuous capability. The people structures that enable distributed thinking about technology. The governance that makes fast decisions without creating chaos. The data foundations that make AI genuinely useful rather than theoretically interesting. The cultural norms that make every person in the business a participant in the adoption process rather than a recipient of it.

---

## Why No Technology Position Lasts, and Why That is the Point

Here is the claim that most boards are not yet willing to sit with: no technology position is likely to hold for long. In many cases, the advantage from a specific tool, workflow or implementation may last only 12 to 18 months before it is matched, commoditised or overtaken. The organisations that are currently ahead on a specific AI implementation will not stay ahead by virtue of that implementation. The advantage will erode.

This is not a counsel of despair. It is a clarification of what the actual prize is.

The organisations that will win the AI era are not the ones that picked the right tools in 2025 or 2026. They are the ones that built an organisation capable of adopting, extracting value from, and moving on from technologies faster than their competitors. Continuously. As a core competency. Every wave of new technology becomes an opportunity rather than a disruption, because the machine built to ride it is already running.

That is a fundamentally different objective than the one most leadership teams are currently pursuing. It shifts the question from "what should we implement?" to "how do we build an organisation that is permanently better at this than anyone else in our market?"

The businesses building that capability quietly right now are not talking about it. They are not publishing essays or briefing journalists. They are doing the work. And the advantage they are compounding will be very difficult to close once it becomes visible.

---

## The Conversation Most Boards are Not Having Yet

Twelve months ago, it was reasonable for a leadership team to treat AI as an important initiative requiring attention and resource. That framing is now insufficient.

The organisations that will define the competitive landscape in your sector over the next five years are not the ones that treated AI as an initiative. They are the ones whose boards and executive teams understood that the shift from tech-supported to tech-centric was an organisational transformation requiring a different kind of leadership commitment, a different set of structural foundations, and a different conception of what the business is for.

That conversation starts with leadership genuinely imagining the destination. Dorsey and Botha are pointing in a direction that is worth taking seriously. But the path there runs through a shift most organisations have not yet made: from tech-supported to tech-centric. And that shift does not begin with technology. It begins with foundations. It is a clearer picture of their own business: faster, leaner, more adaptive and more capable than the one they are running today. What does that look like? What does it require? What stands between here and there?

Most boards have not had that conversation in any serious depth. The ones that have are already building.

The question worth asking in your next leadership meeting is not which AI tools to invest in. It is whether you are building the kind of organisation that can ride every wave of disruption that follows, faster and more effectively than your competitors.

That is a different question. It requires a different conversation. And the longer it is deferred, the more expensive the gap becomes.

---

*Footnote: This argument is most visible in mid-market businesses, where the capability gap tends to be sharpest and the consequences most immediate. But the pattern holds across organisations of every size and sector. The question of whether a business is building the operating system for sustained technology adoption is as relevant in a FTSE 100 boardroom as it is in a 200-person professional services firm.*

---

*Adrian Tripp is a Partner at Prosper AI Consulting, leading the business strategy practice and advising boards on the shift from tech-supported to tech-centric organisations.*

---

*Article | Prosper AI Consulting, UK*

---

# The Web Is Rebuilding Itself for AI Agents - And Your Organisation Needs to Pay Attention

Source: https://www.prosperconsulting.ai/insights/the-web-is-rebuilding-itself-for-ai-agents

> The web was built for humans clicking links. In Q1 2026, three of the biggest infrastructure companies on the planet - Cloudflare, Shopify, and Google - quietly started rebuilding it for AI agents. Not as an experiment. Not as a beta feature buried in developer documentation. As production…

The web was built for humans clicking links. In Q1 2026, three of the biggest infrastructure companies on the planet - Cloudflare, Shopify, and Google - quietly started rebuilding it for AI agents. Not as an experiment. Not as a beta feature buried in developer documentation. As production infrastructure, available now, reshaping how content is delivered, commerce is conducted, and information is discovered.

This article connects five developments that landed in a six-week window. Individually, each is interesting. Together, they tell a story that every business leader needs to understand: the web now has two front doors, and most organisations have only built one of them.

---

## The Language Is Changing: From SEO to AEO to GEO

For two decades, digital strategy meant one thing: SEO. Get your pages ranking. Drive clicks. Measure traffic.

That vocabulary is expanding. Two new terms have entered the conversation in 2026:

- **AEO (Answer Engine Optimisation)** - optimising content so AI answer engines (ChatGPT, Perplexity, Google AI Overviews) surface it directly in their responses
- **GEO (Generative Engine Optimisation)** - the broader discipline of ensuring your content is cited, referenced, and used by AI systems across platforms

The data behind this shift is striking. AI Overviews now reduce traditional clicks by 58% (Ahrefs, February 2026). But here is the counterpoint that matters: AI-driven sessions convert at 14.2% compared to 2.8% for traditional organic search (Conductor, Q1 2026). Fewer clicks, but dramatically better ones.

This creates what the industry is calling the "Crocodile Mouth Effect" - impressions are up 31%, but organic clicks are down 18% year-over-year. The mouth is opening. If you are still measuring success by clicks alone, the picture looks dire. If you measure by citations and conversions, the picture looks different.

The practical implication: success in 2026 is measured by whether AI systems cite your content, not whether humans click on it. SEO is not dead, but it is no longer sufficient on its own.

---

## Cloudflare: Serving Two Versions of the Web

On 12 February 2026, Cloudflare launched Markdown for Agents. The concept is elegant. When an AI agent requests a web page, it sends an `Accept: text/markdown` header. Cloudflare intercepts the request at the CDN layer and converts the HTML to clean markdown before returning it. Human visitors see normal HTML. AI agents get content in their native format.

The economics are significant. A blog post that requires 16,180 tokens in HTML shrinks to approximately 3,150 tokens in markdown - a reduction of over 80%. When AI agents are processing thousands of pages to answer a query, that difference translates directly into speed, cost, and quality.

But the more important feature is Content Signals. Alongside Markdown for Agents, Cloudflare introduced a mechanism that lets website owners declare how AI systems may use their content. Through the `Content-Signal` HTTP header, publishers can signal whether their content is available for AI training, search indexing, agentic use, or any combination. This is not robots.txt (which blocks crawlers). This is content negotiation - serving the same content in different ways to different consumers, with explicit publisher control over what AI can do with it.

The feature is available at no additional cost for Pro, Business, and Enterprise Cloudflare plans. Claude Code and several other AI development tools already send the `Accept: text/markdown` header when fetching web content.

**The debate:** Google's John Mueller called converting pages to markdown "such a stupid idea" in January 2026. The counter-argument from Cloudflare and supporters: this is standard HTTP content negotiation, the same mechanism that has served images in different formats for years. It delivers the same content in a different format. It is not cloaking. Both perspectives have merit, and the debate is far from settled.

---

## llms.txt: A Curated Front Door for AI

While Cloudflare handles content delivery, a complementary standard is addressing content discovery. Jeremy Howard's llms.txt specification provides a markdown-formatted map of a site's most important resources, placed at the root directory - like robots.txt, but designed to guide AI rather than block it.

The format is simple: a markdown file with a title, a summary blockquote, and structured links to the pages the site owner considers most important. Where robots.txt says "do not go here", llms.txt says "start here, and here is what matters."

Adoption is growing. Anthropic, Cloudflare, Docker, and HubSpot have published llms.txt files. Google referenced the standard in their Agent-to-Agent (A2A) protocol. Implementation takes 1-4 hours and carries no demonstrated downside.

The honest assessment: there is no proof yet that having an llms.txt file increases AI citation frequency. The impact is unproven at scale. But the implementation cost is near zero and the downside risk is negligible. This is a classic low-risk, potential-reward calculation that appeals to early adopters - and that calculation is why adoption is accelerating.

---

## Shopify and Google: AI Agents That Buy Things

In January 2026 at NRF, Shopify and Google jointly announced the Universal Commerce Protocol (UCP) - an open standard that defines how AI agents discover products, negotiate terms, and complete purchases across merchants.

The ambition is significant. UCP is not a Shopify feature or a Google product. It is an open protocol, publicly available at ucp.dev, backed by over 20 companies including Walmart, Target, Etsy, Wayfair, Mastercard, Visa, and Stripe. The architecture is layered - shopping service, capabilities, and extensions - each independently versioned, deliberately reminiscent of how TCP/IP structured the internet itself.

The practical mechanism is a `ucp.json` manifest - a machine-readable "passport" for your store that broadcasts its capabilities to AI agents. What products are available. What payment methods are accepted. What checkout flows are supported. An AI agent reading this manifest can discover, compare, and purchase products without ever rendering a web page.

The early numbers are compelling. Shopping-related searches on AI platforms grew 4,700% between 2024 and 2025. Shopify merchants are already selling via ChatGPT, with Google AI Mode and Microsoft Copilot integrations rolling out. The shift from "search and click" to "ask and buy" is not theoretical. It is happening.

**The reality check:** only 17% of consumers are currently comfortable letting AI complete a purchase on their behalf (ChannelEngine, 2026). The infrastructure is being built for 100% adoption, but consumer trust has not caught up. This gap is a timing opportunity, not a reason to wait. The merchants who are machine-readable when consumer comfort arrives will have a structural advantage over those who start building then.

---

## Google AI Mode: The End of Ten Blue Links

Google's AI Mode, powered by Gemini, has rolled out globally. It replaces the traditional search results page with an AI-synthesised answer that includes citations to source material.

The distinction from AI Overviews matters. AI Overviews appear alongside traditional search results - you might lose the featured snippet, but your organic listing still exists. AI Mode has no organic results alongside. You are cited in the AI answer, or you are invisible.

Research from Ahrefs (December 2025) found only 13.7% citation overlap between AI Mode and AI Overviews. They cite different sources. This means optimising for one does not automatically cover the other.

The traffic impact is real. Some sites report losing 20-60% of organic traffic. AI Overviews alone reduce clicks by 58%. But the quality story is more nuanced: when AI Mode does send traffic, those visitors have been pre-qualified by the AI. They know what they are looking for. They convert better.

Google reports that AI Mode queries are 2-3 times longer than traditional searches. Users are asking complex, multi-part questions that would have required multiple searches before. The AI synthesises across sources. If your content is authoritative, well-structured, and citable, you benefit. If it is thin, duplicative, or poorly structured, you disappear.

---

## The Pattern: A Parallel Web for Agents

These five developments share a common thread. The web is building dedicated infrastructure for AI agents - not as an afterthought, but as a parallel system.

| Layer | Human Web | Agent Web |
|:------|:----------|:----------|
| **Content delivery** | HTML via browser | Markdown via `Accept: text/markdown` (Cloudflare) |
| **Site map** | `sitemap.xml` | `llms.txt` |
| **Commerce** | Shopping cart + checkout page | UCP manifest + agent negotiation (Shopify/Google) |
| **Search** | Ten blue links | AI-synthesised answer with citations (Google AI Mode) |
| **Success metric** | Clicks, rankings | Citations, AI visibility |
| **Optimisation discipline** | SEO | AEO / GEO |

This is not six separate stories. It is one architectural shift happening simultaneously across content delivery, discovery, commerce, and search. The infrastructure companies are not waiting for standards bodies or industry consensus. They are building the agent web now.

---

## What This Means for Your Organisation

The response depends on what you do.

### If You Own a Website

Your content now serves two audiences. For AI agents:

1. **Publish an llms.txt file** (1-4 hours, no downside). Point AI to your most important content.
2. **Enable Cloudflare Markdown for Agents** if you use Cloudflare. Free. Opt-in. Immediate token savings for agents consuming your content.
3. **Review your Content Signals** - decide explicitly how AI may use your content rather than leaving it to default.
4. **Structure your content for citation** - clear headings, factual claims with evidence, authoritative sourcing. This is what AI systems cite.
5. **Measure AI visibility** alongside traditional SEO metrics. Tools are emerging. Start tracking now.

### If You Sell Online

The "ask and buy" channel is real and growing.

1. **Understand UCP** and evaluate whether your commerce platform supports it. Shopify merchants have a head start.
2. **Make your product data machine-readable** - structured data, clear specifications, competitive pricing in parseable formats.
3. **Monitor AI-driven commerce metrics** - Shopify's AI commerce dashboard is a model for what to track.
4. **Prepare for the consumer trust inflection** - when the 17% becomes 50%, the merchants who are already machine-readable will capture disproportionate share.

### If You Deploy AI Agents

The infrastructure your agents interact with is changing.

1. **Send `Accept: text/markdown` headers** when fetching web content. The token savings are immediate and significant.
2. **Check for llms.txt** when researching organisations. It tells you what the site owner considers important.
3. **Explore UCP** for agentic commerce use cases. The protocol is open and the ecosystem is growing.
4. **Design for content negotiation** - your agents should gracefully handle both HTML and markdown responses.

---

## The Window

The web did not ask permission to split in two. The infrastructure companies built it. Cloudflare, Shopify, and Google made their decisions in the space of six weeks, and the parallel web for agents is now live.

Organisations that understand this shift now have a window. Not to panic, not to overhaul everything overnight, but to make deliberate, low-cost moves - an llms.txt file, a Content Signals header, structured product data - that position them for the agent-first world that is arriving.

Those that wait will find their content invisible and their products undiscoverable to the fastest-growing channel on the internet. Not because they were blocked. Because they never opened the second door.

---

**Sources and further reading:**
- [Cloudflare: Introducing Markdown for Agents](https://blog.cloudflare.com/markdown-for-agents/) (12 February 2026)
- [llms.txt specification](https://llmstxt.org/)
- [Shopify: Building the Universal Commerce Protocol](https://shopify.engineering/ucp)
- [UCP specification](https://ucp.dev/)
- [Google Developers: Under the Hood - UCP](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/)
- [Conductor: AEO/GEO Benchmarks Report Q1 2026](https://www.conductor.com/academy/aeo-geo-benchmarks-report/)
- [Ahrefs: AI Search Visibility Research](https://searchengineland.com/ai-search-visibility-seo-predictions-2026-468042)

---

*Paul Bratcher is a Partner at Prosper AI Consulting, specialising in outcome-driven technology adoption for organisations navigating AI transformation.*

---

*Article | Prosper AI Consulting, UK*

---

# Production Architectures for Agentic AI: Microsoft, Google, and AWS Compared

Source: https://www.prosperconsulting.ai/insights/production-architectures-agentic-ai

> The hyperscalers have spoken. Microsoft, Google, and AWS have each published production reference architectures for agentic AI. The naming differs, the services differ, but the underlying patterns are remarkably similar. This matters because it signals where enterprise AI is heading and what…

The hyperscalers have spoken. Microsoft, Google, and AWS have each published production reference architectures for agentic AI. The naming differs, the services differ, but the underlying patterns are remarkably similar. This matters because it signals where enterprise AI is heading and what infrastructure decisions you need to make now.

Microsoft calls theirs the "Golden Path." Google publishes detailed architecture guides through their Cloud Architecture Center. AWS offers Bedrock AgentCore. All three address the same fundamental challenge: moving AI agents from pilot to production at enterprise scale.

## The Common Pattern

Strip away the branding and all three architectures solve for the same requirements:

**Multi-agent orchestration.** Single agents solving isolated problems was the 2023 model. Production systems in 2025 require networks of specialised agents working together under unified coordination.

**State and memory management.** Agents need to maintain context across conversations and learn from interactions. All three platforms provide managed memory services for both short-term session state and long-term knowledge retention.

**Tool integration.** Agents must connect to enterprise systems, APIs, and data sources. Each platform offers gateways that transform existing services into agent-callable tools.

**Security and identity.** Enterprise-grade authentication, authorisation, and audit trails. Non-negotiable for production workloads.

**Observability.** Tracing, logging, and monitoring of agent decisions. You cannot govern what you cannot see.

## Platform Comparison

The table below maps equivalent components across all three platforms:

| Capability | Microsoft Azure | Google Cloud | AWS |
|------------|-----------------|--------------|-----|
| **Unified Platform** | Azure AI Foundry | Vertex AI Agent Builder | Amazon Bedrock |
| **Agent Runtime** | Foundry Agent Service | Agent Engine | AgentCore Runtime |
| **Orchestration Framework** | Semantic Kernel / Agent Framework | Agent Development Kit (ADK) | Strands Agents / Bedrock Agents |
| **Memory Management** | Cosmos DB (threads) | Agent Engine Sessions / Firestore | AgentCore Memory |
| **Tool Gateway** | Logic Apps, Functions, MCP | Cloud Run, MCP servers | AgentCore Gateway |
| **Knowledge/RAG** | Azure AI Search | Vertex AI Search | Bedrock Knowledge Bases |
| **Identity** | Microsoft Entra ID | IAM / Agent Identity | AgentCore Identity |
| **Observability** | Azure Monitor, Application Insights | Cloud Logging, Cloud Trace | AgentCore Observability (CloudWatch) |
| **Code Execution** | Code Interpreter | Code Interpreter | AgentCore Code Interpreter |
| **Container Deployment** | Container Apps | Cloud Run / GKE | Lambda / ECS |

The architectural patterns are nearly identical. The choice between platforms typically comes down to existing cloud investments, not technical capability gaps.

## What Each Platform Does Well

**Microsoft Azure** offers the tightest integration with enterprise productivity tools. If your organisation runs on Microsoft 365, SharePoint, and Dynamics, the Golden Path provides native connectors. The Semantic Kernel framework has strong enterprise adoption, with organisations like KPMG and Fujitsu using it for multi-agent orchestration in production.

**Google Cloud** leads on the open-source developer experience. The Agent Development Kit (ADK) has been downloaded over 7 million times and powers agents across Google's own products. The Agent Garden provides pre-built samples and tools that accelerate development. Google also pioneered the Agent-to-Agent (A2A) protocol for cross-platform agent communication.

**AWS** offers the broadest model selection and framework flexibility. AgentCore is explicitly framework-agnostic, working with CrewAI, LangGraph, LlamaIndex, or custom implementations. If you need to mix models from different providers or want maximum portability, AWS provides that optionality.

## The MCP Convergence

All three platforms now support Model Context Protocol (MCP), the open standard for connecting agents to tools and data sources. This is significant. It means agents built on one platform can potentially use tools exposed by another. The walls between ecosystems are becoming more permeable.

For enterprises, this reduces lock-in risk. Invest in MCP-compatible tool development now, and those tools remain usable regardless of which platform you standardise on later.

## Where the Platforms Diverge

**Managed vs. flexible.** Microsoft and Google offer more opinionated, managed services. AWS AgentCore provides more infrastructure primitives that you assemble yourself. The trade-off is speed-to-production versus customisation depth.

**Framework preference.** If your team knows LangChain or LangGraph, Google's integration is smoother. If you prefer .NET and C#, Microsoft's Semantic Kernel is the natural choice. AWS supports everything but optimises for nothing specific.

**Pricing models.** All three charge for compute, memory, and model inference. The specifics vary significantly. Model your expected workload before committing.

## For the C-Suite: Platform Selection Criteria

The platform choice is less about technical features and more about organisational fit:

1. **Where do you already have cloud investment?** Migrating to a new cloud for agentic AI alone rarely makes sense. Build on what you have.

2. **What is your developer ecosystem?** The framework your team knows will be deployed faster than the framework with better features.

3. **What are your compliance requirements?** All three meet major compliance standards, but the specific certifications differ. Check before assuming.

4. **How important is model flexibility?** If you need to swap models frequently or use specialised fine-tuned models, evaluate each platform's model catalogue.

## Getting Started

**Audit your current state.** Map every active AI initiative against production readiness. Be honest about what is progressing versus what is stalled.

**Identify your first multi-agent candidate.** Look for workflows requiring multiple handoffs, approvals, or system integrations. These are natural candidates for agent orchestration.

**Run a proof-of-architecture.** Before committing to a platform, build the same simple agent on two platforms. Compare developer experience, deployment complexity, and observability quality.

**Define success criteria before building.** Latency thresholds, accuracy requirements, compliance obligations, cost limits. Write them down. Measure against them.

## The Bottom Line

The three major cloud providers have converged on a common architectural pattern for production agentic AI. Multi-agent orchestration, managed memory, tool gateways, security, and observability. The implementations differ in detail but not in substance.

This convergence tells us something important: the production patterns for agentic AI are stabilising. The experimental phase is ending. The question is no longer whether to build multi-agent systems, but on which foundation.

For most organisations, the answer is straightforward: build on the cloud where you already operate. The architectural patterns are similar enough that platform switching later, while not trivial, remains possible. What matters more is getting production experience now, learning what works for your specific use cases, and building organisational capability while competitors are still running pilots.

---

*Which platform is your organisation evaluating for agentic AI? What factors are driving your decision? Share in the comments.*

---

**Paul Bratcher** is a Partner at Prosper AI Consulting, specialising in AI transformation for mid-market and enterprise organisations. He developed the ODTA (Outcome-Driven Technology Adoption) framework to help leadership teams navigate technology decisions with clarity.

---

# Building Claude Code Skills: What I Have Learned So Far

Source: https://www.prosperconsulting.ai/insights/building-claude-code-skills

> In my last newsletter, I shared Career Helper, a Claude skill for job seekers. I mentioned I'd follow up with what I've learned from building 50+ skills. This is that follow-up. I've been building with Claude Code since its launch in February 2025. When Anthropic formalised the skills feature in…

In my last newsletter, I shared Career Helper, a Claude skill for job seekers. I mentioned I'd follow up with what I've learned from building 50+ skills. This is that follow-up.

I've been building with Claude Code since its launch in February 2025. When Anthropic formalised the skills feature in October, we already had dozens of working skills in production. Career Helper, which I gave away free on GitHub, was one of them. It has fifteen capability modules, handles everything from CV optimisation to salary negotiation, and demonstrates what's actually possible when you move beyond simple prompts.

This isn't a tutorial. Anthropic's documentation covers the basics well enough. This is what I've learned from doing it at scale: the patterns that work, the mistakes that cost time, and the mental models that make everything easier.

## What Skills Actually Are

A skill is a way to give Claude structured expertise in a specific task or activity. But that framing undersells what's possible.

Skills aren't just prompts. They're a combination of prompts, context, templates, and workflow logic that together create something closer to a specialist tool than a general assistant.

For example, my Career-Helper skill has fifteen capability modules. Each one has its own approach, its own templates, its own quality standards. When you ask it to prepare you for an interview, it researches the company first, generates questions likely for that specific role, and builds answers from your actual experience. Same approach every time.

That's the gap between a prompt and a skill. A prompt is a one-shot instruction. A skill is a persistent capability with memory, structure, and standards. It has a level of context and completeness that provides a repeatable result.

## The Three Levels of Automation

I've found it useful to think about three distinct levels when building with Claude Code:

**Slash commands** are simple, repeatable actions that users invoke explicitly. Type `/commit` and get a formatted commit message. Type `/review` and get a code review. They're good for things you do frequently that don't require judgement about when to do them.

**Skills** are domain expertise encoded as structured workflows. The key difference from commands: Claude decides when to use them based on what you're asking for. Ask about job interviews, and Career Helper activates without you typing anything. Skills are semantic, commands are explicit. Right now this is where I am finding the most success in implementing working solutions.

**Sub-agents** run in isolated context windows with their own tool access. A planning agent might coordinate research agents, analysis agents, and writing agents, each with their own specialisations, working in parallel on different aspects of a larger task. The sub-agents and hooks system has great promise but right now feels like it needs a little more control and orchestration options.

The automation hierarchy matters. Commands require user invocation. Skills require semantic matching. Sub-agents require orchestration logic. Most people stop at commands. The real opportunity is in skills and sub-agents.

## Who This Is For

Building skills isn't excessively complex, but it's not purely no-code either. It sits somewhere between low-code and some-code.

You'll benefit from some experience with coding, even if it's just enough to read and extend existing examples. More importantly, you need to be able to write well-structured prompts that are repeatable and testable. And you need to think in systems: how components connect, where state lives, what happens when things fail.

If you're comfortable with structured problem-solving and can write clear instructions, you have the foundation. The rest is learnable.

## Progressive Disclosure

One pattern that took me a while to appreciate: you don't load everything upfront.

Before skills were introduced, the @ symbol in Claude Code was primarily a way to include subfiles. When Anthropic formalised skills, this changed how progressive disclosure worked. What I had previously built as sub-agents, I moved to skills. The context loading became more intentional.

The better approach is progressive disclosure. Start with minimal context. Load more as needed.

The mechanism is straightforward: your main skill file links to supporting documents using standard Markdown links. Claude reads these linked files only when the current task requires that information. Your core instructions might be 300 lines, but the full skill could have thousands of lines of templates, examples, and reference material that only get loaded when relevant.

Structure your skill as layers: a core instruction set that's always loaded, supporting documents that are linked and read on-demand, and detailed templates that are pulled in for specific outputs.

## The Date Problem

Here's something that will bite you if you don't know about it: Claude's awareness of the current date isn't always reliable.

The issue isn't that Claude can't access dates. It's that different execution environments handle this differently, and the documentation doesn't make guarantees. Claude Code provides environment context that typically includes the current date. But I've seen enough edge cases to adopt a simple rule: always explicitly provide the current date in your skill's context if your outputs depend on it.

This matters more than you'd think. Research tasks need current information. Document generation needs accurate dates. Scheduling logic breaks if the model gets confused about temporal context.

The fix is defensive: inject the date explicitly rather than assuming it's available. Belt and braces.

## Standalone vs Interdependent Design

Early on, I built skills that depended heavily on each other. Research skill feeds into analysis skill feeds into writing skill. Very elegant on paper.

In practice, dependencies create fragility. If one skill changes, others break. Testing becomes complex. Users can't use individual skills in isolation.

Now I design skills to be standalone by default. Each skill should be useful on its own. Dependencies are optional enhancements, not requirements.

The exception is sub-agent orchestration, where one skill explicitly coordinates others. But even then, the individual skills remain independently functional.

## Templates and Persistence

Skills need to produce consistent outputs. The way to achieve this is templates.

But templates aren't just formatting. Good templates encode methodology. They force comprehensive coverage of a topic. They ensure outputs include everything they should, sections, quality checks, citations, structured analysis.

Persistence is the other half. Skills need to remember what they've done in a conversation. What research has been gathered, what decisions have been made, what outputs have been produced.

Claude Code gives you file system access. Use it. Write intermediate results to disk. Create working documents that accumulate context. Don't rely on conversation memory for anything important.

One of the key things to learn about to help with templates, and prompt structuring, is the use of Markdown and basic html formatting. These both provide excellent ways to define templates and document structure to LLMs.

## Environment Differences

This one catches people: there are multiple ways to run Claude, and they have different capabilities.

**Claude.ai** (desktop and web) does support loading capabilities and skills. The mobile app, at the time of writing, does not. No file system access or code execution in any of these, so they're useful for quick interactions but not for building agentic workflows.

**Claude Code** comes in two forms: the local CLI that runs on your machine, and a cloud-based version that runs in a browser with GitHub integration. Both support skills, file system access, and bash execution. The cloud version runs in an isolated VM with pre-installed toolchains.

**Agent SDK** (Python and TypeScript) gives you programmatic control for building skills into larger applications. Same capabilities, different interface.

Know your target environment before you start building. And if you're building for teams, assume they'll be running in different environments than you.

## Testing Skills

You need a way to test skills systematically. Not unit tests in the traditional sense. Claude is probabilistic. The same input can produce different outputs. But you still need quality assurance.

The approach that works: build validators. Create test scenarios with expected patterns. Run the skill through each scenario. Check that outputs meet structural requirements, content thresholds, and quality standards.

I have a long history with evals and rubrics, so this felt natural. The challenge is automation. Running validators as part of a CI/CD pipeline is still tricky. Running them in bulk causes runtime issues and failures. I haven't fully got this working to my satisfaction, but I'm building towards it. The validators themselves are valuable even when run manually.

## Distribution Across Teams

If you're building skills for yourself, this doesn't matter. If you're building for teams, it matters a lot.

Skills need to be installable without explanation. They need documentation that answers the questions people actually ask. They need versioning so updates don't break existing users.

The challenge is that Claude Code skills are still maturing. You can bundle skills as a private plugin for the marketplace, but that requires users to be running Claude Code. Anthropic has just announced distribution tools for enterprise and teams users, which includes a distribution framework. I haven't tried it yet, so I can't comment on how well it works.

For now, plan for some support overhead. This will improve.

## Four Weeks to Competence

If you're starting from zero, here's a realistic learning path:

**Week One:** Get comfortable with Claude Code basics. Understand the environment, file access, and command execution. Don't build anything complex yet.

**Week Two:** Build your first simple skill. Something you actually need, probably a command or simple workflow. Focus on making it work reliably.

**Week Three:** Add complexity. Templates, progressive disclosure, persistence. Convert your simple skill into something more robust.

**Week Four:** Build for others. Document your skill, test it on someone who wasn't involved in building it, iterate based on their confusion.

Four weeks won't make you an expert. But it will make you competent enough to build useful things.

## The Deeper Pattern

Here's what I've come to believe: skills aren't really about AI. They're about capturing and scaling expertise.

When I build a skill, I'm not teaching Claude something new. I'm encoding knowledge that already exists, methodologies, templates, quality standards, into a form that can be applied consistently at scale.

This is valuable regardless of what happens with AI. The process of building a skill forces you to articulate what you actually know. It reveals gaps in your methodology. It makes implicit expertise explicit.

Even if you never deploy the skill, the exercise of building it clarifies your own thinking.

## What I'm Not Covering

This newsletter is already long, and I've deliberately left things out.

I haven't covered the specific techniques for complex workflows. I haven't explained how to build sub-agent systems that coordinate multiple skills. I haven't discussed quality assurance at scale or team governance around skill development.

Those are deeper topics. They're also the difference between skills that work and skills that work reliably at enterprise scale.

If you're exploring this space and want to go deeper, that's what we do at Prosper. Building skills, training teams, implementing AI capabilities that actually work in production.

---

*This is part of my ongoing series on building and using agentic AI. Previous edition: [Giving Back: A Claude Skill for Job Seekers](https://www.linkedin.com/pulse/giving-back-claude-skill-job-seekers-paul-bratcher-qezfe/)*

*If you're building skills for your organisation and want expert help, reach out. It's what we do.*

---

# Giving Back: A Claude Skill for Job Seekers

Source: https://www.prosperconsulting.ai/insights/giving-back-claude-skill-job-seekers

> I don't know why, but January is often a forced career transition moment, so it seems now is a good time to subvert the newsletter to talk about this Claude skill. It's a pretty useful demonstration of what you can do with a skill, making use of all sorts of techniques. Next newsletter, I will…

I don't know why, but January is often a forced career transition moment, so it seems now is a good time to subvert the newsletter to talk about this Claude skill. It's a pretty useful demonstration of what you can do with a skill, making use of all sorts of techniques. Next newsletter, I will take a dive into what I have learnt having written 50+ Claude skills.

## Sometimes Everyone Needs a Little Help

I've been fortunate in my career. Good timing, good mentors, and more than a few people who went out of their way to help when they didn't have to.

One piece of advice I was given years ago has stayed with me: give help forwards. When someone helps you, you can't always repay them directly. But you can help the next person who needs it.

This is my attempt at that.

## What I Built

Career Helper is a skill for Claude. If you use Claude (the AI assistant from Anthropic), you can install this skill and it extends what Claude can do for you. Specifically, it helps with job searching.

I'm not a recruiter. I'm not a career coach. I'm a consultant who has been through career transitions myself, helped colleagues through theirs, and spent enough time with Claude to realise that AI is genuinely useful for the heavy lifting that job searching requires.

The insight isn't complicated: job searching involves a lot of repetitive, time-consuming work that most people do badly because they're doing it alone, often under stress, and without the bandwidth to do proper research or preparation. AI is good at exactly this kind of structured, research-heavy, detail-oriented work.

So I built a tool and made it free.

## What AI Assistance Actually Looks Like

There's a lot of noise about AI replacing jobs. Less discussion about AI helping people find jobs.

Here's what I've learned from building this: AI assistance isn't about replacing human judgement. It's about doing the legwork so humans can make better decisions.

Career Helper doesn't tell you what job to apply for. It doesn't make decisions for you. What it does is the preparatory work that most candidates skip because they don't have time, don't know how, or are too overwhelmed to do properly.

It will research a company before you interview there, surfacing information about their strategic priorities, recent news, leadership changes, and financial health. You still have to decide what to do with that intelligence. But at least you have it.

It will rewrite your CV to pass through applicant tracking systems, identifying the keywords you're missing and restructuring your experience to match what the algorithms are looking for. You still have to verify the result is accurate. But the heavy lifting is done.

It will prepare you for interviews by generating likely questions based on the specific role, creating structured answers that draw on your actual experience, and explaining what interviewers are really assessing behind each question. You still have to practise and perform. But you're not starting from scratch.

The pattern is consistent: AI handles the volume, the research, the structural work. You provide the judgement, the authenticity, the final decisions.

## Non-Judgemental Support

One thing I've noticed about working with AI on career topics: it doesn't judge.

Job searching is emotionally difficult. There's rejection, uncertainty, and often a sense that you should already know how to do this. Asking for help can feel like admitting failure.

AI doesn't care. It will show up for you at 11pm when you're anxious about tomorrow's interview. It will help you rewrite your CV for the fifteenth time without sighing. It will research the same company again because you forgot what it told you yesterday.

It's not a replacement for human connection. But it's a form of support that's available when you need it, doesn't tire of your questions, and treats every request as worth taking seriously.

For people going through career transitions, especially difficult ones, that consistency matters.

## What Career Helper Actually Does

The skill extends Claude with fifteen capabilities designed for different stages of job searching. You don't need to use all of them. Start with whatever you need most.

### Getting Your Materials Ready

- **LinkedIn Profile Optimisation** - Rewrites your profile so recruiters can actually find you when they search. Covers headline, about section, skills ordering, and discoverability.
- **CV Optimisation for ATS** - Restructures your CV to pass through applicant tracking systems. Identifies keywords from job descriptions and ensures your experience is presented in formats that algorithms can parse.
- **LinkedIn Content Strategy** - For longer searches, helps you maintain visibility with a sustainable posting approach. Authentic content based on your actual expertise, not viral hooks.

### Researching Opportunities

- **Company Research** - Comprehensive intelligence gathering before you apply or interview. Financial health, strategic direction, leadership, culture, recent news. Every claim cited with sources.
- **Networking Intelligence** - Identifies who to connect with at target companies. Maps hiring managers, internal recruiters, potential advocates, and warm introduction paths.

### Preparing for Interviews

- **Interview Preparation** - Generates likely questions for the specific role and creates structured answers using your actual experience. Explains what interviewers are really assessing.
- **Interviewer's Perspective Reports** - Shows you the view from the other side of the table. What makes a strong answer. What red flags interviewers watch for.
- **Mock Interviews** - Practice sessions with different interviewer personas. Detailed feedback on your responses.

### When Things Don't Go to Plan

- **Post-Interview Coaching** - Diagnoses what went wrong after a rejection. Identifies whether you faced a skill gap, a signal gap, or a fit issue. Helps you learn from it rather than just move on.

### Planning and Strategy

- **3-Month Job Search Plan** - Structured activity planning with goals, weekly breakdowns, and daily rhythms. Includes wellbeing practices because job searching is hard.
- **Application Strategy** - Timeline planning, stakeholder mapping, and follow-up protocols for high-priority opportunities.

### Offer Stage

- **Salary Negotiation Coaching** - Market research, counter-offer scripts, and total compensation analysis. Adapted for UK, US, EU, and APAC markets.
- **Offer Evaluation** - Framework for comparing multiple offers or evaluating a single one. Weighted decision matrix based on your priorities.

### Non-Traditional Paths

- **Portfolio and Fractional Careers** - For those building fractional executive, advisory, or portfolio careers. Rate setting, legal structures, client acquisition.
- **AI Readiness Assessment** - Skills gap analysis and upskilling roadmap for demonstrating AI competency in the 2025+ job market.

## Not Looking for a Job?

Career Helper isn't only for active job seekers.

If you're happily employed but want to keep your LinkedIn profile current, the profile optimisation tools work just as well for maintenance as for job hunting. Profiles go stale. Skills sections become outdated. Headlines that made sense three years ago no longer reflect what you do. Periodic refresh matters, even when you're not looking.

If you're building a portfolio career, juggling fractional roles, advisory work, consulting, or board positions, there's dedicated support for that. Rate setting guidance, legal structure options (IR35 considerations for the UK, LLC versus S-Corp for the US), client acquisition strategies, and LinkedIn positioning specifically for fractional executives.

If you're focused on thought leadership, the content strategy coaching helps you develop a sustainable posting rhythm based on your actual expertise. Not viral hooks. Not LinkedIn clichés. Authentic content that positions you as a thinking professional in your domain.

The tool adapts to where you are. Not everyone is job searching. But almost everyone could use help keeping their professional presence sharp.

## Career Stage Matters

The skill also adapts its guidance based on where you are in your career. The challenges are genuinely different.

Early career candidates need to demonstrate potential without much track record. Mid-career professionals often face pivots and transitions. Experienced executives encounter age bias, both quiet and explicit. Late-career professionals may be exploring fractional, advisory, or board positions rather than traditional employment.

Career Helper doesn't require you to navigate this complexity. Tell it where you are, and the guidance adapts accordingly.

## How to Get It

The skill is available on GitHub. You'll need a GitHub account, which requires nothing more than an email address. If you don't have one, now's a good time. GitHub is increasingly where useful tools live, and having a login opens up a lot of resources.

**Download link:** github.com/Zal4DW/career-helper/releases/tag/1.0.7

Installation instructions are in the repository. If you use Claude Pro or Claude for Work (you do need a subscription, and this is for the desktop/web app), you can add skills to extend its capabilities. Career Helper is one of those skills.

It's free. No catch. No upsell. Licensed under CC BY-NC 4.0, which means you can use it, share it, and adapt it for non-commercial purposes. If you think it would be useful to someone, share it forwards, if you end up as a CEO at a company needing AI help, you know how to find us.

## What This Demonstrates About Claude Skills

If you're curious about what Claude skills can actually do, Career Helper is a reasonable example.

Skills are essentially structured instructions that extend Claude's capabilities in specific domains. They can include templates, workflows, research methodologies, and quality standards. They can make Claude behave more like a specialist tool than a general assistant.

Career Helper includes fifteen capability modules, each with its own approach and output templates. It uses web search for research, generates structured documents, maintains context across a conversation, and adapts its guidance based on user inputs.

This isn't magic. It's careful prompt engineering, good templates, and thoughtful workflow design. Anyone can build skills like this for their own domains.

If you're interested in building skills for your own use cases, Career Helper's repository shows one approach. The methodology is transferable.

## The Bottom Line

Job searching is hard. It's emotionally taxing, procedurally complex, and most people have to do it without much support.

AI won't make it easy. But it can make it less lonely, less overwhelming, and more structured. It can do the research you don't have time for, prepare you for interviews you're anxious about, and help you learn from rejections rather than just endure them.

Career Helper is my contribution to that. Built because someone once helped me when they didn't have to. Given away because that's what giving help forwards means.

If it's useful to you, use it. If it's useful to someone you know, share it.

And if you land a role, perhaps consider what giving help forwards might look like for you.

---

*Career Helper is a Claude skill developed by Paul Bratcher at Prosper AI Consulting. Free for non-commercial use.*

*Download: github.com/Zal4DW/career-helper/releases/tag/1.0.7*

---

# The Horizon Trap: Why Your Strategy Only Looks Forward in One Direction

Source: https://www.prosperconsulting.ai/insights/the-horizon-trap

> Let us begin with a familiar scene. The annual strategy offsite. A three-year roadmap on the screen. Year one is detailed, year two is vague, year three is aspirational. Everyone nods. The meeting ends. Twelve months later, the same people gather to create another three-year plan that looks…

Let us begin with a familiar scene. The annual strategy offsite. A three-year roadmap on the screen. Year one is detailed, year two is vague, year three is aspirational. Everyone nods. The meeting ends. Twelve months later, the same people gather to create another three-year plan that looks remarkably similar to the last one.

This isn't strategic planning. It's operational planning with a longer timeline and better slides.

I've watched this pattern repeat across UK businesses of all sizes, but it hits medium enterprises hardest. Large corporations have dedicated strategy teams scanning horizons. Small businesses survive on agility and instinct. The companies in between, the £10m to £250m revenue bracket, often have neither the resources for dedicated foresight nor the flexibility to pivot quickly when disruption arrives.

The result is predictable. These businesses optimise what they already do whilst remaining blind to what's coming. They're excellent at efficiency. They're poor at anticipation.

This article provides a framework for thinking about strategy across multiple horizons. It's designed to be practical for resource-constrained leadership teams who can't afford dedicated futures capability but can't afford to ignore the future either.

## The Core Problem: One-Dimensional Thinking

Most strategy discussions focus on a single question: when can we do this? Near term, medium term, long term. That's a reasonable question, but it's not the only one that matters.

Opportunities vary along two dimensions, not one:

**Time Horizon:** How far away is practical implementation?

**Adjacency Horizon:** How close is this to what we already do?

An automation opportunity in your core operations is fundamentally different from applying a technique you've seen in another industry, even if both could be implemented in the same timeframe. The first requires execution discipline. The second requires translation, experimentation and probably some failures before it works.

Treating them the same way is a mistake. And it's a mistake that leads to strategic blind spots.

## Part 1: The Two-Dimensional Model

### Time Horizons

| Horizon | Timeframe | Focus |
|---------|-----------|-------|
| H0 | Now | Execute current operations |
| H1 | Near-term (0-18 months) | Extend and optimise |
| H2 | Medium-term (18-36 months) | Build new capabilities |
| H3 | Long-term (3+ years) | Explore possibilities |

### Adjacency Horizons

| Adjacency | Description | Risk Level |
|-----------|-------------|------------|
| A0 | Core business | Low |
| A1 | Adjacent to current operations | Medium |
| A2 | Techniques from other sectors | Medium-High |
| A3 | Emerging/unproven approaches | High |

### The Combined Matrix

When the two dimensions combine, they create 16 distinct zones. Each zone suggests a different strategic response.

The discipline is matching your response to the zone. An H0/A0 opportunity needs a business case and execution plan. An H2/A2 opportunity needs a learning agenda and connections to people who understand that space. Treating them the same way wastes resources and creates frustration.

## Part 2: Applying the Framework

### Phase One: Map Your Current Reality

Before changing anything, understand where your attention actually goes.

Take your active initiatives, projects and areas of focus. Classify each one by time horizon and adjacency. Be honest. Most UK mid-market businesses discover their portfolio is heavily skewed toward H0-H1 and A0-A1. That's not necessarily wrong, but it should be deliberate rather than default.

Questions to ask:

- Which zones have zero attention? Is that a conscious choice?
- Where is your innovation budget actually going?
- What would happen if a significant disruption emerged from a zone you're ignoring?

### Phase Two: Identify the Gaps That Matter

Not every gap needs filling. Resource constraints are real, especially for medium enterprises competing against both larger players with deeper pockets and smaller ones with lower overheads.

The question isn't "are we covering all sixteen zones?" It's "are we covering the zones that matter for our specific situation?"

For most UK mid-market businesses, the critical gaps tend to be:

**H2/A0 (Medium-term, Core):** Preparing for how your core business will need to operate in three years. Many businesses assume the future will look like the present with minor adjustments. It often doesn't.

**H1/A2 (Near-term, Alt Sector):** Learning from what other industries have already figured out. UK businesses are often too insular, missing techniques that are mature elsewhere but novel in their sector.

**H2/A1 (Medium-term, Adjacent):** Building capabilities for adjacent opportunities before they become urgent. By the time an adjacency is obvious, competitors have already moved.

### Phase Three: Build the Scanning Discipline

Horizon thinking isn't a one-time exercise. It's an ongoing discipline that requires structured information sources.

For a medium enterprise, this doesn't require a dedicated team. It requires discipline. One person tracking each circle, thirty minutes per week, shared monthly. The investment is modest. The value compounds over time.

### Phase Four: Establish Signposts and Kill Criteria

Signposts are specific indicators that would suggest an opportunity is maturing or a threat is approaching. "If X happens, we should accelerate our work on Y."

Examples for UK mid-market businesses:

- If a competitor launches a direct-to-consumer channel, we accelerate our own D2C pilot
- If regulatory consultation on X concludes, we revisit our compliance roadmap
- If adoption of technology Y reaches 20% in our customer base, we move from H2 to H1 posture
- If our key supplier announces vertical integration, we accelerate alternative sourcing
- If customer acquisition cost rises above threshold Z, we prioritise retention technology

The discipline is writing these down and reviewing them regularly.

Kill criteria are equally important. What evidence would suggest you should stop pursuing an opportunity? Sunk cost thinking destroys strategic discipline. Define the exit conditions before you need them.

For each H1 or H2 initiative, I recommend defining:

- **Time gate:** If we haven't achieved X by date Y, we stop or pivot
- **Resource gate:** If investment exceeds £Z without milestone achievement, we reassess
- **Evidence gate:** If assumption A proves false, we acknowledge the learning and move on

This feels uncomfortable. Nobody wants to pre-commit to potentially abandoning something they're advocating for. But the alternative is initiatives that drift indefinitely, consuming resources whilst delivering nothing, the pilot purgatory I've written about previously.

### Phase Five: Match Evaluation Criteria to Horizon

Different horizons need different success metrics. I've seen promising initiatives killed because someone applied the wrong evaluation criteria.

| Horizon | Appropriate Metrics |
|---------|---------------------|
| H0 | ROI, efficiency, cost savings |
| H1 | Pilot results, adoption rates |
| H2 | Learning, capability building |
| H3 | Options created, insights gained |

This connects directly to pilot purgatory, which I've written about previously. Many pilots fail not because the underlying opportunity is poor, but because someone demanded H0 metrics from an H1 experiment, or expected H1 timelines from H2 capability building.

### Phase Six: Integrate with Annual Planning

Horizon thinking should inform your annual strategy cycle, not replace it.

Before the strategy offsite, map the opportunity landscape across all zones. Bring the matrix to the discussion. Use it to structure conversation about where attention and resources should go.

During budget allocation, explicitly discuss horizon distribution. "We're proposing 70% to H0-H1, 20% to H2, 10% to H3. Is that right given our competitive position and disruption risk?"

In quarterly reviews, revisit signposts. What has changed? Do any zones need rebalancing? Has anything moved from one horizon to another?

## Part 3: The AI Question

Here's where this framework becomes particularly relevant right now.

Every UK business is being asked about AI. Vendors are pitching. Competitors are announcing initiatives. The pressure to "do something with AI" is intense.

The horizon framework provides a structured way to think about this:

**Where does AI sit on your matrix?**

For some applications, AI is H0/A0: proven technology, applicable to your core operations, ready for implementation. Process automation, document handling, customer service augmentation. If the business case works, execute.

For others, AI is H1/A1: early adopters showing results in adjacent applications, worth piloting but not yet proven in your specific context. Predictive analytics, personalisation, workflow optimisation.

For others still, AI is H2/A2 or even H3/A3: promising capabilities being developed in other sectors that might eventually apply to your business, but requiring significant translation and experimentation before relevance becomes clear.

The mistake is treating all AI opportunities as if they sit in the same zone. Some need execution. Some need experimentation. Some need watching. The appropriate response depends on where the specific application sits on your matrix, not on the general hype cycle.

Ask yourself: for each AI opportunity being discussed in your business, which zone does it actually occupy? And are you responding appropriately for that zone?

## Part 4: Common Failure Patterns

The framework helps, but cognitive biases work against it:

**Recency bias:** Overweighting recently encountered information. The opportunity someone mentioned last week gets more attention than something identified six months ago, regardless of strategic importance.

**Confirmation bias:** Seeking evidence that supports existing beliefs. If leadership has already decided AI is transformational (or overhyped), new information gets interpreted to confirm that view.

**Proximity bias:** Favouring opportunities close to current operations. Core opportunities are easier to evaluate, so they get more attention. Adjacent and alternative sector opportunities require more imagination.

**Urgency bias:** Prioritising immediate concerns over strategic importance. Quarterly pressure always feels more pressing than five-year positioning.

The framework doesn't eliminate these biases. It makes them visible. When you can see that your portfolio is 80% H0-H1 and A0-A1, you can at least ask whether that's deliberate.

## Part 5: Getting Started

For a UK medium enterprise, here's a practical starting point:

**Week One: Map Your Current Portfolio**

Take every active initiative and classify it by horizon and adjacency. Be rigorous. Include the projects everyone knows about and the informal experiments happening in corners of the business.

Create a simple visual: initiatives plotted on the 4x4 matrix. Most businesses discover their portfolio is heavily clustered in one or two zones. That clustering should be deliberate, not accidental.

**Week Two: Identify the Gaps That Matter**

Not all gaps need filling. Look specifically at:

- H2/A0: Are you preparing for how your core business will operate in three years?
- H1/A2: Are you learning from what other industries have already solved?
- H2/A1: Are you building capabilities for adjacent opportunities before they become urgent?

Pick three gaps maximum. Resource constraints are real.

**Week Three: Assign Scanning Responsibility**

Distribute the circles of listening across your leadership team. One person per circle, thirty minutes per week, monthly share-out. This isn't a full-time job. It's a discipline.

**Week Four: Define Your Signposts**

Write down five specific indicators you'll track. For each one, define what response it would trigger. Put this somewhere visible and review it quarterly.

**Ongoing: Quarterly Matrix Review**

Every quarter, revisit the framework:

- What's moved between horizons? (H2 opportunities maturing to H1, H1 initiatives ready for H0 execution)
- What's emerged that wasn't on the matrix before?
- Does allocation need adjusting given competitive developments?
- Have any signposts triggered?

This cadence matters. Annual review isn't frequent enough, the landscape changes faster than that. Monthly is probably excessive for most mid-market businesses. Quarterly hits the balance.

## The Bottom Line

Most strategic planning is one-dimensional. It asks when opportunities might be ready, but not how close they are to current capabilities. It focuses on time horizons whilst ignoring adjacency horizons.

The result is predictable. Businesses optimise what they already do whilst remaining blind to what's coming from unexpected directions. They execute well on H0/A0 whilst being surprised by H2/A2.

The framework in this article won't predict the future. Nothing does. But it will structure your attention so you're more likely to notice relevant signals before they become obvious. For resource-constrained UK businesses competing in uncertain markets, that peripheral vision matters.

The question isn't whether you can afford to think across multiple horizons. It's whether you can afford not to.

---

# Better ChatGPT 5.1 Prompt Writing

Source: https://www.prosperconsulting.ai/insights/better-chatgpt-51-prompt-writing

> All of the Major providers have cook books, or prompt writing guides. With the release of chatGPT 5.1 there are a few changes you can make to when you are creating prompts to get better results out of it. I know this looks a little alien, and is a bit more effort. Do I use these approaches all of…

All of the Major providers have cook books, or prompt writing guides. With the release of chatGPT 5.1 there are a few changes you can make to when you are creating prompts to get better results out of it.

I know this looks a little alien, and is a bit more effort. Do I use these approaches all of the time, no. But once I have got to a better more complex need, then this is the approach I personally lean into.

## Tips for better chatgpt5.1 prompts

### Give it a clear role, not just a task

Open with who you want it to be: "You are my operations consultant for our UK business" or "Act as our Head of Customer Service". Specify tone (plain, executive, friendly) and how decisive it should be (offer options vs choose a recommendation).

### Be explicit about length and format

Tell it exactly what you want: "Two short paragraphs", "Maximum 7 bullet points", "Briefing note with headings: Context, Analysis, Recommendation". GPT-5.1 is very good at obeying concrete rules on length, sections and bullet use if you spell them out.

### Ask for persistence and end-to-end answers

Add something like: "Do not stop at high level ideas. Work through to a concrete, implementable recommendation and only ask questions if something is genuinely blocking." This reduces half-finished answers and pushes it to finish the job in one go.

### Use progress updates for bigger work

For heavier tasks, say: "First give me a short plan, then brief progress updates as you work, then a final summary at the end." This is helpful for policy drafting, process redesign, or multi-part analysis, and makes the work easier to review and adjust.

### Make it plan complex work as a checklist and execute it

For anything non-trivial, ask: "Create a 3-6 step checklist of outcomes, then work through each step one by one. Mark each as 'Planned', 'In progress' or 'Done' and only finish when all are Done or clearly cancelled." This turns GPT-5.1 into a lightweight project helper rather than a single-shot answer generator.

If you are writing agents, or want to look further the full cookbook is worth a read here.

## Complete Example Prompt

I thought it might be useful to pull all of the above into a complete prompt, in this case lets imagine we are a HR Director planning a Senior Leadership away day. As always when we provide prompts we are trying to make it obvious to you where you need to add the data (edit). In this case `[ .... ]`, and the list of attendees. Note this prompt is more to show the how to use the tips, than probably a fully practical workshop building tool.

```
<role>
You are a senior leadership and organisation development consultant for UK organisations.
You speak in clear, plain UK business English and focus on practical, people-centred outcomes.
</role>

<variables>
Company name: [COMPANY_NAME]
Company type: [COMPANY_TYPE]
Company location: [COMPANY_LOCATION]
Away day location: [AWAY_DAY_LOCATION]
Away day Date and time: [AWAY_DAY_DATE]
Time horizon in months: [TIME_HORIZON]
Number of attendees: [NUMBER_OF_ATTENDEES]

Attendee list for invites:
{ATTENDEE_LIST}

Format {ATTENDEE_LIST} as a list of lines:
- Name - Role - Email

ATTENDEE_LIST {
# Include each of these as attendees
- Jane Smith - Finance Director - jane.smith@example.com
- Ahmed Khan - Operations Director - ahmed.khan@example.com
}
</variables>

<context>
Company: [COMPANY_NAME], a [COMPANY_TYPE] based in [COMPANY_LOCATION].
Event: 1 day in person leadership away day in [AWAY_DAY_LOCATION].
Attendees: [NUMBER_OF_ATTENDEES] person senior leadership team.
Timeframe: Planning leadership and team development priorities for the next [TIME_HORIZON] months.
User role: HR Director who will sponsor and chair the day.
</context>

<goals>
- Agree 3 to 5 clear leadership and team development priorities for the coming year.
- Identify critical capability gaps, succession risks and cultural themes.
- Define practical development approaches (for example learning, coaching, projects) for the leadership team.
- Agree owners, first actions and review points for each priority.
</goals>

<task>
Design the away day so it can be run with minimal extra work.
Focus on purpose, flow, exercises and decision points, not logistics such as venues or travel.
</task>

<output_format>
Keep the whole answer under 1,000 words.
Use exactly these headings in the final answer:

1. Objectives and success criteria
2. One day agenda with timings
3. Exercises and materials
4. Pre work for attendees
5. Risks and facilitation tips
6. Draft personalised invite emails

In section 6:

- First provide a short, generic invite template.
- Then generate one personalised email per person in [ATTENDEE_LIST].
- Use their name, role and email address. Show the email address clearly at the top of each draft so it can be pasted into the To field.
- Tailor the message slightly using their role or perspective where relevant.
- Keep each email concise and ready to paste into an email client. Include location, date, and a summary of any pre-work to be done, include directions to the location if possible
- Use UK spelling, a professional but warm tone, and leave clear placeholders for date, location and any pre work.

Use short paragraphs and bullet points where helpful.
No tables. No emojis.
</output_format>

<solution_persistence>
Do not stop at high level ideas.
Produce a concrete, implementable design that a competent facilitator could run.
Only ask clarifying questions if something is genuinely blocking.
If details are missing, make sensible assumptions and state them briefly.
</solution_persistence>

<plan_tool_usage>
At the start of your reply:

- Create a checklist of 3 to 6 outcome focused steps.
- Show each step with a status: Planned, In progress, Done or Cancelled.

Within this single reply:

- Work through the steps in order and update the status line as you go.
- The checklist should cover understanding the context, shaping objectives,
designing the agenda, defining exercises and pre work, capturing risks,
and drafting the invite emails.
</plan_tool_usage>

<user_updates_spec>
At the top, give a 2 to 3 sentence plan for how you will approach the work.
When you complete each checklist step, briefly note it before moving on.
End with a short recap of key decisions and 3 to 5 next actions for the HR Director.
</user_updates_spec>

<start>
Now begin.
</start>
```

---

# When Max Met Aisha

Source: https://www.prosperconsulting.ai/insights/when-max-met-aisha

> In October 2025, Channel 4 aired a documentary about AI and employment. The presenter, Aisha Gaban, delivered her lines with the slightly stilted cadence you expect from documentary narration. She looked professional. She sounded credible. And at the end of the programme, she revealed something…

In October 2025, Channel 4 aired a documentary about AI and employment. The presenter, Aisha Gaban, delivered her lines with the slightly stilted cadence you expect from documentary narration. She looked professional. She sounded credible. And at the end of the programme, she revealed something viewers hadn't spotted: "I'm not real. In a British TV first, I'm an AI presenter."

Most people watching didn't notice. That's the point we've reached.

A few weeks earlier, Hollywood unveiled Tilly Norwood, an entirely AI-generated actress. The backlash was immediate and fierce. SAG-AFTRA released a statement emphasising that creativity "is, and should remain, human-centred." Actors condemned the concept. Film directors called it a threat to their craft. Yet beneath the outrage sits an uncomfortable reality: the technology works well enough to provoke this response.

We've crossed a threshold. Digital humans are no longer confined to expensive visual effects studios or experimental research labs. They're reading the news in India, Kuwait, Taiwan and Greece. They're influencing purchasing decisions on Instagram. They're appearing in corporate training videos and customer service chat windows. And most significantly, they're increasingly difficult to distinguish from real people.

This isn't science fiction. It's current practice. The question isn't whether synthetic humans are coming, it's what happens now they're here.

## From Max Headroom to Photorealism: A Forty-Year Journey

The concept of digital presenters isn't new. In 1985, Max Headroom became television's first supposed computer-generated host. Except he wasn't. Max was an actor in prosthetics, marketed as an AI creation. The technology didn't exist yet. The satire did.

By 2000, someone actually tried it. Ananova launched as the world's first virtual newsreader, recognised by Guinness Records. She was a 3D-animated character reading news via text-to-speech on a website. Primitive by modern standards, but her creators predicted something prescient: a "population boom in virtual people" for roles like agents, receptionists and sales representatives.

They were right about the future. Wrong about the timeline.

Throughout the 2000s and 2010s, realistic digital humans remained expensive, time-consuming and firmly in the uncanny valley. Hollywood used motion capture for films like The Polar Express and created CGI replicas of actors for particular scenes, but these were bespoke projects requiring months of work and massive budgets. The technology improved gradually. The business case remained weak.

Then, around 2018, something shifted. China's Xinhua news agency debuted an AI news anchor whose face and voice were synthesised from footage of a real presenter. Not perfect, but functional. Not indistinguishable, but close enough for a news bulletin.

By 2023, AI newsreaders had spread globally. India introduced Sana and Lisa. Greece unveiled Hermes. Kuwait launched Fedha. Taiwan deployed Ni Zhen. These weren't experimental curiosities. They were production systems delivering actual news content to actual audiences.

The Guardian observed in late 2023 that "country after country debuted their first AI news anchor" that year. What changed wasn't just the technology. The economics shifted. The practicality improved. The acceptance grew.

## The Current State: Better Than You Think, Not as Good as They Claim

Let's be precise about capabilities. Today's AI presenters exist on a spectrum.

At one end, you have systems like South Korea's Zae-In, who reads live news on SBS. Technically impressive, but there's a detail worth noting: a real human actor drives her body and voice in real time. Zae-In's flawlessly AI-generated face is overlaid via deepfake technology. It's not autonomous. It's augmented.

At the other end, you have fully synthetic avatars like Channel 4's Aisha Gaban, reading from pre-written scripts with no human performance underneath. The visual fidelity has improved to the point where, on first viewing, she looks human. Close inspection reveals tells. Some viewers noticed a "deadness" in the eyes. The mouth sync wasn't perfect on certain sounds. But these are subtle failures, not obvious ones.

The trajectory matters more than the current state. When NVIDIA revealed in 2021 that part of CEO Jensen Huang's keynote was secretly delivered by a CGI replica, attendees had no idea until NVIDIA disclosed it afterwards. They'd done a full body and face scan of Huang, used AI to mimic his gestures and expressions, and produced a 14-second segment where the digital version was virtually indistinguishable from the real thing.

Fourteen seconds required a truck full of DSLR cameras and custom AI animation tools. Today, creating a decent-looking talking avatar takes minutes using platforms like HeyGen or Synthesia. You type a script, select an avatar from a library, and receive a video of a lifelike person speaking it with appropriate lip-sync and expressions.

The gap between "obviously fake" and "difficult to detect" is closing faster than most people realise.

## The Technology Stack: What Makes This Possible Now

Two technical advances explain the recent acceleration.

First, generative AI and deepfake techniques have matured. Modern GANs and diffusion models can synthesise high-resolution, photorealistic human faces. These systems learn the complex patterns of real human faces moving, then generate new, realistic motion. The Korean company Pulse9 uses deepfake-style face generation to power Zae-In, ensuring her AI face precisely follows the real actor's micro-expressions in real time.

Second, real-time animation and game engines have become accessible. Epic Games' Unreal Engine and its MetaHuman system allow creators to build highly detailed 3D human models with realistic skin, hair and facial rigs, then animate them with motion capture or AI. What once required a Hollywood visual effects studio can now be done by a competent developer with cloud compute access.

The integration point matters. Tools like NVIDIA's Audio2Face automatically generate facial animation from an audio track. Feed it speech, and it produces the corresponding mouth movements and micro-expressions. This removes one of the traditional bottlenecks in avatar creation: manual animation of facial performance.

For static images, we're essentially already at undetectable. AI-generated photos of people regularly fool audiences at scale. For video and interactive avatars, the timeline is shorter than most expect. Some developers claim we could see real-time AI avatars that fool most viewers within just a few years. Others are more cautious, pointing to lingering imperfections and the enormous complexity of human behaviour.

The technical challenge isn't just visual realism. It's contextually aware behaviour. An AI that looks exactly like a human news anchor but speaks with odd phrasing or lacks true understanding will still feel wrong. Progress in large language models is addressing this on the content side, enabling avatars that can hold natural conversations or generate ad-lib responses.

NVIDIA's ACE (Avatar Cloud Engine) combines realistic graphics with AI brains for interactive characters. It provides cloud-based AI models for speech recognition, natural language understanding and facial animation. Early demos show NPC avatars that no longer speak pre-scripted lines but generate dialogue on the fly, with matching facial expressions.

As these systems improve, the line between a video game character, a virtual assistant and a "real" human interaction will blur. For many use cases, it already has.

## The Hollywood Problem: When Technology Meets Culture

The case of Tilly Norwood illustrates both the technical achievement and the cultural collision.

Norwood was created by London-based AI studio Xicoia and introduced via a short parody film. On screen, she appears as a photoreal young woman, described as resembling a fusion of real actresses like Gal Gadot and Ana de Armas. She "acted" in the skit alongside other AI-generated characters, with all her movements and dialogue synthesised by AI tools.

The creator, Eline van der Velden, touted Norwood as a potential "next Scarlett Johansson" and claimed talent agents were interested in representing this virtual actress. The cost argument was explicit: using an AI actor like Tilly Norwood could cut production costs by 90% for certain roles.

SAG-AFTRA's response was unequivocal: "Tilly Norwood is not an actor. It's a character generated by a computer programme. It has no life experience to draw from, no emotion." The union stressed that replacing human performers with AI violates the core of creativity and copyright.

Technically, the demo had problems. Reviewers noted exaggerated mouth movements and occasional visual artifacts like blurring teeth that created an "uncanny valley effect." Fully digital actors aren't quite ready for prime time in serious acting roles. A University of Southern California media tech expert was blunt: much of Hollywood has "zero interest" in purely synthetic stars at present.

Real actors carry human depth, spontaneity and audience connection that an AI creation cannot authentically replicate. Not yet, anyway.

But film production already uses AI-enhanced CGI routinely. De-ageing actors is standard. Recreating deceased actors for brief appearances happens. One 2024 film, The Brutalist, used AI to generate some actors' spoken dialogue in a foreign language. Studios are eyeing AI for digital stunt doubles and background extras at scale.

The economic pressure is real. An AI character doesn't require a trailer, salary or time off. It never ages or gets injured. It can be perfectly controlled. Some producers will continue pushing AI performers, especially for marketing or experimental projects, regardless of creative resistance.

The ability to convincingly replicate a feature-length human film performance with AI stand-ins is still seen as far off. Short clips or single scenes can fool the eye. Maintaining the illusion over hours of content, with all the subtle emotional range and spontaneity of a human, remains extremely challenging.

The technology will improve. The cultural resistance won't disappear.

## Who's Building This: The Vendor Landscape

Multiple companies are driving development and deployment. Understanding the ecosystem matters because these tools are becoming increasingly accessible.

NVIDIA provides core technology through its Omniverse platform and ACE toolkit. The company's GPUs enable the compute needed for real-time rendering. NVIDIA has partnered with game studios including NetEase, Tencent and Ubisoft to integrate AI characters. Their research in AI-generated facial animation and voice synthesis is widely used in avatar systems.

Epic Games provides Unreal Engine and MetaHuman Creator, dramatically reducing the time to get a convincing digital human on screen. The technology has been used in everything from game cutscenes to live virtual concerts featuring digital pop stars.

Specialised startups focus specifically on AI-driven avatars. Synthesia (London) offers an AI video generation platform with a roster of virtual presenters. Soul Machines (New Zealand) creates digital people with animated faces for customer service and education. UneeQ provides a platform for corporate virtual assistants used by banks and other enterprises. DeepBrain AI (South Korea) produces AI news anchors using deepfake technology.

Big tech isn't absent. Meta invests in ultra-realistic codec avatars for VR telepresence. Microsoft's Azure includes avatar SDKs. Apple's Vision Pro introduced the concept of a realistic personal avatar (the "Persona") for video calls, using machine learning to create a lifelike representation.

Importantly, many of these technologies are accessible via cloud APIs or software-as-a-service. You don't need a Hollywood budget to leverage an AI presenter anymore. This democratisation fuels rapid uptake.

## Disrupted Sectors: Where This Actually Matters

The march of photoreal AI humans impacts several industries differently.

Broadcasting and media can run overnight or niche bulletins with AI anchors, reducing staffing costs for certain formats. As seen with Channel 4's experiment and various international examples, AI newsreaders can deliver information around the clock. Most major outlets still value human credibility and will likely limit AI to supplementary roles, but the economic pressure exists.

Gaming and interactive entertainment sits on the cusp of an AI-NPC revolution. Traditionally, game characters follow scripted dialogue trees and pre-animated gestures. With tools like NVIDIA ACE and platforms from Inworld or Convai, developers are giving game characters dynamic AI brains and voices. In future open-world games, every NPC could engage in unscripted conversation with realistic facial animation and speech. This greatly enhances immersion. The gaming industry will likely adopt these AI avatars at scale because they reduce the labour of scripting thousands of lines and recording voiceovers.

Customer service and commerce increasingly use digital humans to staff virtual storefronts, websites and call centres. A bank might have a friendly avatar on its app to answer questions. Retail could deploy virtual shopping assistants. Healthcare is testing AI nurse avatars for patient intake. Education experiments with AI tutors available 24/7 who can personalise instruction and make e-learning more interactive.

Film and television production faces the biggest creative resistance but also the strongest economic incentives. AI could replace extras, stand in for stunt performers, or generate actors for minor roles. Advertising might use AI stand-ins for celebrity endorsers (with permission and licensing). Animation studios might create "virtual actors" for movies or interactive stories. The sector will evolve, but slowly and with significant pushback from creative unions.

## The Personal Avatar Question: When Digital You Becomes Practical

Here's where this gets genuinely interesting for individuals rather than corporations.

We're approaching a point where creating a digital version of yourself becomes practical. Not as a novelty or experiment, but as a functional interface tool.

Consider the implications. Someone with social anxiety could let their digital avatar handle initial video calls. A person with ADHD might find that a digital representation helps them communicate more clearly because the avatar can be programmed to maintain eye contact and measured speech patterns they struggle with naturally.

For professionals who do repetitive presentations or training, a digital avatar could handle the routine deliveries while they focus on strategic work. Customer-facing roles could use avatars for initial interactions, with humans stepping in for complex situations.

The technology for personal avatars already exists. Apple's Persona on Vision Pro creates a realistic digital representation for video calls. The quality isn't perfect yet, but it demonstrates the concept. As the technology improves, we'll see personal avatars become more common in:

Virtual meetings (your avatar attends while you're genuinely engaged but not on camera)
Educational content (record once, your avatar delivers it hundreds of times)
Customer service (your avatar handles tier-one queries, you handle escalations)
Social media (content creation becomes less about filming yourself and more about directing your digital representative)

The cultural acceptance is the bigger barrier than the technology. We're not there yet. But the same trajectory that made AI newsreaders unremarkable suggests personal avatars will eventually become normalised.

Think about how video calls went from awkward novelty to daily routine within a decade. The same adoption curve could apply to digital representatives, especially if they solve real problems for people who struggle with traditional communication methods.

## What This Means for You: Practical Implications

If you're a CTO, engineering leader or business executive, several strategic questions emerge.

First, evaluate where synthetic humans might actually add value in your organisation. Customer service and training are the obvious candidates. If you're spending significant resources on repetitive video content, avatar systems could deliver genuine efficiency gains. But don't deploy them just because the technology exists. Deploy them where they solve actual problems.

Second, understand the detection and authenticity challenge. If you use AI presenters or avatars in customer-facing roles, be transparent. The Channel 4 documentary worked because it disclosed the deception at the end. Using undisclosed AI in contexts where trust matters (financial advice, healthcare, etc.) creates significant risk. Your customers will eventually notice, and the trust damage may exceed the efficiency gain.

Third, watch the talent and IP implications. If you create digital versions of employees, who owns those representations? What happens when the employee leaves? These legal and ethical questions don't have settled answers yet. Get ahead of them rather than discovering the problems after implementation.

Fourth, consider accessibility applications. The technology that creates photorealistic avatars can also help people who struggle with traditional communication. If you're building products or services, think about how digital representatives might improve accessibility for users with social anxiety, communication disorders or other challenges.

Finally, prepare for the normalisation. In five years, seeing AI presenters in training videos or customer service contexts will be unremarkable. The companies that experiment now, learn the limitations and build appropriate guardrails will be better positioned than those who wait for perfection.

## The Bottom Line

We've moved from Max Headroom's satirical fake AI presenter to Aisha Gaban's genuinely convincing one in forty years. The technology has crossed from expensive special effects to accessible cloud services. The quality has progressed from obviously synthetic to difficult to detect.

This isn't a distant future concern. Digital humans are reading news, influencing purchases, staffing customer service and appearing in entertainment today. They're not perfect. They're not indistinguishable in all contexts. But they're good enough for an increasing number of practical applications.

The pattern is familiar. Technologies that seem remarkable become routine through gradual improvement and cultural acceptance. GPS seemed like magic twenty years ago. Now it's background infrastructure. AI-generated voices sounded robotic five years ago. Now they pass for human on customer service calls.

Synthetic humans are following the same trajectory. The question isn't whether they'll become normal, it's how quickly and in which contexts.

For organisations, the strategic move isn't to ban them or rush to adopt them everywhere. It's to identify where they genuinely add value, implement them thoughtfully, maintain transparency, and prepare for a world where digital and human representatives coexist routinely.

For individuals, the question of personal digital avatars remains open. The technology exists. The use cases are emerging. The cultural acceptance is developing. Whether you want a digital version of yourself handling certain interactions is a choice you'll likely face sooner than you expect.

The technology has stopped being special effects. What it becomes next depends on how we choose to use it.

What's your view? Can you imagine contexts where a personal digital avatar would genuinely help you? I'd be curious to hear your perspective in the comments.

---

*Want to explore how emerging technologies like AI avatars might impact your organisation? Connect with me here on LinkedIn for practical insights on navigating technology change without the hype.*

---

# The Pilot Trap: Why Your Next Innovation Will Be a Nightmare

Source: https://www.prosperconsulting.ai/insights/the-pilot-trap

> And How to Save It Let us begin with a difficult truth. That new pilot you are about to kick off, the one being championed internally as the next major leap in transformation, is more likely to stall than succeed. This will not be because of bad code, immature algorithms or even external market…

## And How to Save It

Let us begin with a difficult truth. That new pilot you are about to kick off, the one being championed internally as the next major leap in transformation, is more likely to stall than succeed. This will not be because of bad code, immature algorithms or even external market forces. It will happen because, organisationally, we have not learned the lessons of the last thirty years.

I have seen this story repeat across sectors and industries. Smart people create well-intentioned ideas, backed by credible insight. They gain internal traction, produce promising results and then disappear. The pilot is not cancelled. It is not even declared a failure. It simply stops progressing. There is no budget request for scale-up, no handover into operations and no official decision either way. Instead it slides quietly into organisational limbo. The team disperses. Stakeholders turn their attention elsewhere. What was once described as high impact becomes a historical footnote, recycled into someone's CV as an experiment that showed potential.

This is pilot purgatory. It is a state in which innovation is neither alive nor officially dead, sustained only by inertia and the reluctance to confront what went wrong.

## The Data Has Been Trying to Warn Us

According to the Standish Group, which has tracked IT project outcomes since 1994, the overall success rate for technology initiatives remains low. In 1994, only 16 per cent of projects were delivered on time, on budget and with acceptable scope. Three decades and multiple waves of transformation later, that number has risen only marginally, to 31 per cent. Despite agile frameworks, digital tooling and executive buy-in, most innovation initiatives still stall or fail.

**31% of change projects are successful, after 30 years of doing them...**

Worse still, 19 per cent of projects fail completely, and over 50 per cent are classified as challenged. These are not edge cases. This is the industry norm. The lesson is simple. Pilots do not fail because of technical limitations. They fail because we do not design them to succeed beyond their own narrow scope.

## What Pilots Are Supposed to Do

When executed well, pilots serve a clear and essential function. They allow organisations to manage risk, validate assumptions and assess feasibility before committing large-scale resources. A good pilot should explore technical viability, process fit, user adoption and commercial value. It should be constructed as the first step of a repeatable and supportable system, not as a proof-of-concept for internal theatre.

**Pilots and trials should assess outcome need, reduce risk, be disposable**

Too often, however, this purpose becomes distorted. Pilots are used to impress stakeholders, to build momentum or to demonstrate capability. They are built quickly, often in isolation, using curated data and shortcuts that will never survive in production. Once pressure mounts to show results, a dangerous phrase enters the conversation: "Let us just put the pilot live."

At this point, all the things that were deliberately skipped are suddenly required. Monitoring, logging, error handling, security, performance optimisation and user support are now urgent problems. The data pipeline is not production-ready, and the integration points have not been tested. Business ownership is unclear, and there is no budget provision for scale. The result is predictable. The system stalls, trust declines and the project joins the long list of initiatives that almost made it.

## How to Spot a Doomed Pilot Before It Begins

There are usually signs. These patterns repeat so consistently that they are worth identifying up front. If your pilot shows more than a few of the following characteristics, it is almost certainly heading for purgatory.

**No unit economics.** If nobody can tell you the cost per transaction, breakeven point or expected efficiency gain, you are not building a business case. You are building a toy.

**A focus on demonstration over foundation.** If the pilot is designed to look impressive, but cannot handle real data, it is unlikely to succeed beyond the first presentation.

**Lack of ownership.** If there is no individual in the business who is accountable for making this live, operating it and supporting its users, then there is no future state. Departments do not scale software. People do.

**Clean, curated data.** If your pilot depends on perfect datasets, it will fail the moment it meets production reality. Messy, incomplete, contradictory data is the norm. Systems must be designed accordingly.

**Uncontrolled scope.** If the pilot started with one objective and now includes five, the probability of success has dropped significantly. Simplicity is not just helpful. It is essential.

**Technology driving the conversation.** If the discussion begins with what the platform can do, rather than what the business needs, the pilot will become a showcase, not a solution.

**Change management as an afterthought.** If most of the effort is spent on building the system, and little is invested in supporting people through the transition, adoption will fail. Culture eats code.

## Even in Flight, the Signs Are Clear

Sometimes it is not obvious until later. There are a few reliable indicators that a pilot is already drifting.

**Endless extensions.** If timelines keep slipping with no hard decision points, the project has lost momentum. Pilots should have fixed gates, not rolling deadlines.

**Shifting success criteria.** If the metrics for success change every few weeks, nobody really knows what good looks like. Without clarity, you are chasing approval rather than impact.

**Stakeholder disengagement.** If your original champions are no longer responding, the project has lost internal credibility. Attention is a proxy for belief.

**No integration plan.** If integration is treated as a future task, it will become a future problem. If it is not designed in from the start, it is almost always too expensive to retrofit.

**User avoidance.** If end users are not adopting the system, or are finding workarounds, the pilot is solving the wrong problem. Resistance is a form of feedback.

**If stakeholders lose interest, stop showing up, stop the pilot.**

## A Framework for Making It to Production

It is not all bad news. There are organisations that succeed, and the ones that do tend to follow a very different approach. They treat the pilot as the beginning of the production system, and they follow a deliberate sequence of decisions, design and delivery.

This is the framework I recommend, based on work with organisations that have succeeded in moving pilots to real-world deployment.

### Phase One: Define the Sandbox

Before anything is built, define your constraints. This is not about opportunity. It is about boundaries. Begin with the systems that cannot be touched, the processes that must remain intact, and the organisational realities that will not change. Then define your security obligations, your integration capabilities, your data availability and your regulatory context.

Clarity of constraint is what drives realism in design. A pilot built without boundaries will inevitably collapse the moment it meets one.

### Phase Two: Find the Owner

Ownership must be personal. Someone must be named. That person must have the authority to fund production, the power to allocate team time and the willingness to be held accountable if it fails.

There are three tests to apply here. Can they approve the budget without requiring five layers of committee? Can they mandate adoption across teams? And what happens to them if it does not work? If the answer to any of those is uncertain, you need a new owner.

### Phase Three: Define Success in Four Scenarios

Before the first line of code is written, define your decision points. These should include:

- **A clear failure threshold.** If the pilot cannot process live data by a certain point, it stops.
- **A minimum viable outcome.** For example, a 20 per cent improvement in speed with 80 per cent reliability.
- **A target success state.** What the business would consider a strong result, such as a 40 per cent reduction in effort, a measurable financial return or high user satisfaction.
- **A runaway success scenario.** If the pilot performs significantly better than expected, is the infrastructure ready to scale? Are resources available?

These scenarios must be written down, circulated and revisited, signed off. They become the reference point when decisions are required under pressure.

### Phase Four: Build the Coalition

There are three critical groups who must be engaged early.

- The operators, who will run the system.
- The integrators, whose systems it touches.
- The affected, whose workflows will change.

Each of these groups can block you if excluded. They can also become your champions if involved meaningfully.

This is not stakeholder management, it needs to be personal and relentless.

### Phase Five: Design for Day 100, Build for Day 1

Do not design the system for the demo. Design it for production. Then scale it back to fit the pilot. From the beginning, include logging, monitoring, error handling, basic documentation, and access control. Use production data. Expect failure and build paths through it.

This is more work. It is also significantly faster than rebuilding everything later.

### Phase Six: Run the Pilot with Hard Gates

Your pilot should have at least three structured checkpoints.

- At day thirty, confirm technical viability. Does it work with real data? Do the numbers still make sense?
- At day sixty, test integration. Can it connect to the required systems? Do the users understand it?
- At day ninety, make the go or no-go decision. Are the criteria met? Is the business case intact? Is the scale-up plan viable?

Do not extend these gates. Make the call.

### Phase Seven: Plan the Transition Before It Ends

The most common reason that good pilots fail to scale is that nobody plans for what happens next. As you approach the end of the pilot, document the transition.

- Who takes ownership of the system?
- What will the first thirty days look like in production?
- Where is the funding coming from?
- What will be scaled first, and what risks will that introduce?
- What is the rollback plan if something breaks?

If these questions are not answered before the pilot concludes, the likelihood of success drops sharply. You need to consider probably the worst case question and scenario from a risk perspective.

**The pilots are so successful, can't we just turn it on everywhere?**

A runaway success pilot brings the greatest risk of all, in addition to failure you should have a plan for success, and rapid scaling. Including the costs, risks and resource requirements, resource includes training, operational staff, releasing the innovation / pilot team from becoming the 'forever support team'.

## Change is Human, Not Technical

The final insight may upset your engineering team. The success of your pilot does not depend on technical excellence. It depends on change management.

**Most pilots are 70% people, not software engineering. Most pilots do not plan for that.**

The most successful programmes allocate 70 per cent of their resources to people and process, 20 per cent to systems and data and only 10 per cent to the algorithms themselves. This is not a fluke. It is because the majority of failure modes are organisational, not computational.

People do not resist technology. They resist systems that make their work harder, threaten their roles or change things without explanation. Your pilot is only valuable if it is adopted, and adoption is a human problem.

## The Bottom Line

We have had thirty years of innovation pilots. And we have spent much of that time getting it wrong. The tools have changed. The failure patterns have not.

We still confuse impressive demonstrations with usable foundations. We still avoid the hard questions about ownership, scale and risk. We still push pilots into production without re-architecting for reality.

Your next pilot does not have to fail. But if you want it to succeed, you will need to do something that most organisations still avoid. You must treat the pilot not as a test, or a demonstration, or a hope. You must treat it as the deliberate first step of your production system.

It is less exciting. It is more difficult. It is also far more likely to work.

The decision is yours, need some help... you know how to find us.

*p.s, if you are part of a leadership team we are running two of our well received AI for C-Suites day events in the next quarter, if you are interested in attending and the details and getting your AI pilot off the ground and headed to production... drop me a message*

---

# The Board's Dilemma: Navigating 5 Forces Reshaping the Future of Work

Source: https://www.prosperconsulting.ai/insights/the-boards-dilemma

> Smart, experienced leaders are sitting around polished tables, surrounded by reports that would make a data analyst weep with joy, yet they're more uncertain than I've seen them in years. Why? Because we're navigating something genuinely unprecedented. Five generations working together. Digital…

Smart, experienced leaders are sitting around polished tables, surrounded by reports that would make a data analyst weep with joy, yet they're more uncertain than I've seen them in years.

Why? Because we're navigating something genuinely unprecedented. Five generations working together. Digital transformation programmes that promised the world but delivered mixed results. AI disruption accelerating faster than anyone planned. Economic uncertainty rippling from Brexit impacts to global tensions.

After analysing many reports on the future of work across Europe and the UK, one statistic keeps haunting me: only 35% of companies are successfully navigating this complexity. The rest aren't failing through lack of effort or investment. They're struggling because the old playbooks simply don't work when five forces converge simultaneously.

Let me take you through what I've learned, what most boards are missing, and more importantly, what the successful ones are actually doing differently.

## The Reality Check: We're Genuinely in Uncharted Territory

Here's what the data tells us about where we stand right now. Economic uncertainty sits at its highest levels since Covid, with the Economic Policy Uncertainty Index showing volatility that would have been unthinkable just five years ago. Brexit's impact runs deeper than most anticipated, with the UK economy sitting 4% smaller than pre-referendum projections suggested it would be by now.

But here's what makes this moment genuinely unique: we have five generations working together for the first time in history. Not four. Five. Each bringing fundamentally different expectations about work, technology, leadership, and life balance.

Meanwhile, our digital transformation efforts are struggling. Despite massive investment, only 35% of digital transformation programmes actually succeed. The EU's ambitious Digital Decade targets look increasingly optimistic when you consider that 55.6% of EU citizens still lack basic digital skills, and current trajectories suggest only 17% of EU companies will use AI by 2030. The target? 75%. There's a bit of a gap there.

What makes this particularly challenging for boards is that these aren't separate problems requiring separate solutions. They're interconnected forces that amplify each other. The generational divide affects digital adoption. Economic uncertainty impacts investment in transformation. Skills gaps slow AI implementation. And round it goes.

## The Generational Reality: Your Workforce Has Different Operating Systems

Let me share something that illustrates the difference.

**Only 6% of Gen Z workers actually want leadership roles.**
**Boomers want work-life balance, Gen Z want life-work balance.**

Here's the fuller picture. Gen Z, which will represent 23% of the UK workforce by 2025, operates with fundamentally different assumptions about work than every generation before them.

- Seventy-seven percent would quit rather than work full-time in an office.
- Eighty percent have been forced to change jobs due to cost of living pressures.

They're not being difficult; they're responding rationally to economic realities that previous generations didn't face.

Compare this to Boomers, who hold 81% favourable views of managers, with 38% considering delaying retirement and only 22% actively job searching. The contrast isn't just generational preference; it's operational incompatibility.

I watched one company struggle with this for months. Their Boomer managers couldn't understand why their Gen Z engineers seemed disengaged during traditional team meetings but were brilliant in async collaboration tools. The breakthrough came when they realised they weren't dealing with attitude problems, but communication protocol mismatches.

The successful approach? Stop trying to force everyone into the same mould. Start designing systems that work across generational preferences. This isn't about being nice to different age groups; it's about operational effectiveness in a multi-generational workplace.

## The Digital Transformation Paradox: Big Investment, Mixed Results

Here's where things get interesting from a strategic perspective. The EU has allocated €288.6 billion across member states for digital initiatives. That's serious money. Yet fibre reaches just 64% of EU households, and businesses report that their top barriers to digital transformation are data quality issues (49%) and skills shortages (44%).

The numbers tell a story that most boards find uncomfortable: we're spending massive amounts on digital transformation while the fundamental infrastructure and capabilities remain patchy.

Look at your digital transformation dashboard. Green lights everywhere. Investments on track, milestones achieved, stakeholder satisfaction high. What about actual business outcomes? Any measurable changes? Or is it hidden in jargon and excuses?

The disconnect isn't deliberate; it's systemic. We've become brilliant at measuring activity rather than impact. Project completion rates rather than operational improvements. Technology deployment rather than business transformation.

The State of Digital Decade report reveals this pattern across Europe. Countries are hitting infrastructure targets while missing adoption goals. Companies are implementing systems while struggling with user engagement. The technology works; the transformation often doesn't.

## The AI Revolution We Didn't Plan For: It's Moving Faster Than Anyone Expected

While boards were debating AI strategies, their employees got on with it. Seventy-five percent of global knowledge workers are already using AI tools. Microsoft's data shows 40% using AI daily. That's not a future trend; that's current reality.

But here's what caught everyone off guard: the speed of adoption varies dramatically by generation. Seventy-four percent of Gen Z believe AI will transform work this year. They're not waiting for corporate AI strategies; they're integrating AI into their workflow and expecting their employers to catch up.

The twist that's causing real tension? Sixty-two percent of Gen Z also fear AI replacement. They're simultaneously embracing the technology and worrying about job security. This isn't contradictory; it's entirely rational. They understand AI's capabilities better than most and recognise both the opportunities and threats.

The jobs most affected so far are writers, translators, and customer service roles. But the impact isn't what most predicted. Instead of wholesale replacement, we're seeing augmentation and role evolution. The writers using AI effectively are becoming more productive. Those who resist are becoming less competitive.

One client's content team provides a perfect example of this dynamic. Initially resistant to AI tools, they're now producing 40% more content while reporting higher job satisfaction. Why? Because AI handles the routine work, freeing them for strategic and creative tasks. Their roles evolved rather than disappeared.

## The Economic Pressure Points: External Forces Amplifying Internal Complexity

The economic backdrop makes everything more challenging. Brexit reduced UK productivity by 4% according to recent analysis. Business investment is down 11% due to ongoing uncertainty. Over 50% of Gen Z workers are living paycheck to paycheck, making job security a primary concern rather than career development.

Across the EU, growth forecasts have been slashed to 1.1% for 2025. Trade tensions are escalating globally. Inflation concerns persist despite policy interventions. For boards trying to plan multi-year transformation programmes, this economic volatility creates a planning nightmare.

But here's what I've observed in successful organisations: they treat economic uncertainty as a design constraint rather than an excuse for inaction. Instead of waiting for stability, they build adaptive capacity into their strategies.

Rather than a five-year plan with fixed milestones, could you pivot to create a series of six-month experiments with clear go/no-go decision points? When market conditions change this allows you to adapt more rapidly.

## The Strategy Framework That Actually Works: Future-Ready Board Thinking

After working with dozens of organisations navigating this complexity, I've identified a framework that consistently delivers results. It's built around three principles that successful boards have adopted. It's a journey from viewing tech as support to central. From a cost to a value part of the organisation. Technology as new product innovations, PLM for systems.

### Think Outcomes, Not Jobs

Stop thinking about replacing roles. Start thinking about better outcomes. This shift in perspective changes everything.

Instead of asking "Which jobs will AI replace?" ask "Which outcomes can AI genuinely enhance?" The difference is profound. Role replacement thinking leads to defensive strategies and employee resistance. Outcome enhancement thinking leads to collaborative innovation and employee engagement.

Rather than questioning which roles to replace. The breakthrough question is: "How can we help our analysts deliver better insights faster?" Suddenly, AI became a tool for professional enhancement rather than job threat.

The practical approach involves mapping outcomes to capabilities rather than roles to tasks. Which outcomes absolutely require human creativity and judgement? Which can benefit from AI augmentation? Which processes could be genuinely automated? Structure work around results, not just efficiency improvements.

### Build Generational Bridges

The statistic that changes everything:

**94% of employees stay when companies invest in their development.**

But development means different things to different generations.

For Gen Z, development often means flexibility and purpose alignment. Seventy-seven percent consider flexible work arrangements non-negotiable. For them, rigid office requirements signal that the company doesn't trust or value them.

For older generations, development might mean leadership opportunities and knowledge transfer. They often want to mentor and share experience but need modern frameworks for doing so effectively.

The successful approach isn't choosing one generational preference over another; it's creating systems that bridge different working styles and expectations. What about "reverse mentoring" programmes where Gen Z employees teach AI tools to senior colleagues while learning business context and decision-making frameworks?

### Create Ruthless Priorities

This is perhaps the most critical element. Make space for AI-based change by stopping things that don't matter.

Most transformation efforts fail not because the new initiatives are wrong, but because organisations try to add them on top of everything else. People are already working at capacity. Adding AI tools and new processes without removing existing work simply creates overwhelm and resistance.

The question every board should ask: "What can we stop doing to focus on transformation?" Which legacy processes consume energy without adding value? Which meetings could be eliminated? Which reports are created out of habit rather than necessity?

Conduct a "corporate archaeology" exercise, identifying processes that existed solely because they'd always existed. Typically you will eliminate 30% of recurring meetings and consolidate reporting systems. Make the bandwidth to implement AI tools effectively.

## Your Leadership-First Action Plan: All Successful Changes Start with Understanding

The pattern I see in successful transformations is consistent: they start with leadership understanding, not delegation. Boards that successfully navigate this complexity engage directly with the challenge rather than assuming others will handle it.

### Learn It: Month 1

Board members must understand the changing technology landscape personally. This isn't about becoming technical experts; it's about developing sufficient familiarity to make informed strategic decisions.

Hands-on experience with AI tools relevant to your sector is essential. If you're in manufacturing, understand how AI impacts supply chain management. If you're in financial services, experience AI-powered risk assessment tools. Don't just read reports about these technologies; use them.

Map what AI can actually do for your specific business outcomes. Generic AI strategies fail because they don't connect to real operational challenges. What are your organisation's top three business outcomes? How specifically could AI enhance your ability to achieve them?

Stop delegating technology understanding to others. CTOs and IT directors provide crucial expertise, but strategic technology decisions require board-level understanding. You wouldn't delegate financial decisions solely to the CFO; don't delegate technology strategy solely to technical teams.

### Lead It: Month 2

Identify clear, measurable goals and outcomes for AI integration. Vague aspirations like "become more data-driven" or "embrace digital transformation" don't create actionable strategies.

Define success metrics that matter to your business. Revenue impact, cost reduction, customer satisfaction improvements, employee productivity gains. Choose metrics you already track and understand rather than creating new measurement systems.

Communicate the vision across all generational groups using language and channels that resonate with each. Gen Z employees might engage better with interactive workshops and digital platforms. Gen X managers might prefer detailed written strategies and structured implementation plans. Boomers might want face-to-face discussions and historical context.

Make the hard choices about what stops to make space for change. This is where leadership courage matters most. Eliminating familiar processes and comfortable routines requires executive sponsorship and clear communication about why change is necessary.

### Train the Change: Month 3

Not everyone can learn in isolation. Build structured support systems that recognise different learning preferences and generational approaches.

Create cross-generational learning partnerships. Pair tech-savvy younger employees with experienced older colleagues. The knowledge transfer goes both ways: technical skills flowing up, business wisdom flowing down.

Establish continuous learning pathways, not one-off training events. Technology changes too quickly for annual training programmes. Create systems for ongoing skill development that integrate with daily work rather than competing for time and attention.

Scale what works, stop what doesn't quickly. Successful organisations are ruthless about discontinuing ineffective programmes. Create clear metrics for training effectiveness and be prepared to pivot when approaches aren't delivering results.

**Write it down! If you aren't recording your results you are not experimenting you are messing about**

## The Companies Getting It Right: Success Stories Worth Studying

Microsoft's "Frontier Firms" research reveals fascinating patterns among companies successfully navigating this complexity. Seventy-one percent report their organisation is thriving, compared to just 37% average across all companies. Ninety percent report meaningful work opportunities for employees. Fifty-five percent say their employees can take on more challenging work.

What makes them different? They treat AI as a capable colleague rather than a replacement threat. They focus on continuous learning rather than one-off training events. They actively bridge generational divides instead of ignoring them.

Real examples provide concrete inspiration. Wells Fargo used AI to cut customer service response times by 95% while improving satisfaction scores. Unilever uses AI to augment creative teams, generating more ideas and testing concepts faster without replacing human creativity. ASML continues leading semiconductor innovation despite economic uncertainty by treating AI as an enhancement tool for their already exceptional engineering teams. Cisco and zapier both have published additional thoughts and results which are worth looking into.

The pattern across successful organisations is clear: they view current challenges as design constraints rather than obstacles. Instead of waiting for stability, they build adaptive capacity. Instead of choosing between generations, they create bridges. Instead of fearing AI, they integrate it thoughtfully.

## Your Decision Point: The Choice Is Quite Simple, Actually

We're at an inflection point that demands fresh thinking from every board. The convergence of generational shifts, digital transformation challenges, AI acceleration, and economic uncertainty isn't approaching, it's here.

The question isn't whether your organisation will face these challenges. You already are. The question is whether your board will be among the 35% that successfully navigate this complexity or the 65% that get overwhelmed by it.

The difference isn't resources, intelligence, or market position. It's approach.

Successful boards engage directly with change rather than delegating it. They build bridges across generational divides rather than choosing sides. They treat AI as an augmentation tool rather than a replacement threat. They create space for transformation by eliminating what doesn't matter.

Your three immediate actions are straightforward but not easy:

First, assess your Digital Decade readiness honestly. Where do you actually stand on digital capabilities, not where your reports suggest you stand? What are the real barriers to adoption in your organisation? Which assumptions about digital transformation need updating based on current evidence?

Second, create a genuinely multi-generational AI strategy. This means understanding how different generations interact with technology and designing implementation approaches that work across age groups. It means treating generational diversity as a strategic advantage rather than a management challenge.

Third, build your resilience plan today, not tomorrow. Economic uncertainty isn't temporary; it's the new operating environment. Design strategies that thrive in volatility rather than waiting for stability that may not return.

## The Bottom Line

The future of work isn't a distant concept requiring long-term planning. It's the current reality requiring immediate attention. The boards that recognise this and act decisively will thrive. Those that continue treating these challenges as separate problems requiring eventual solutions will struggle.

The convergence of five generations, digital transformation realities, AI acceleration, economic uncertainty, and changing work expectations creates complexity that traditional management approaches can't handle. But it also creates opportunities for organisations willing to think differently.

Your employees, across all generations, are already adapting to this new reality. The question is whether your board will lead that adaptation or get left behind by it.

The choice, as I said, is quite simple. The execution, of course, is where the work begins.

What's your take? How is your board navigating these converging forces? I'd love to hear your experiences and insights in the comments.

*Want more practical insights on leadership in complex times? Connect with me here on LinkedIn for regular analysis on the intersection of strategy, technology, and generational dynamics in modern organisations. Perhaps we can help you on this exciting strategic change we all face*

---

# Future-Proofing a Legacy Finance Firm with AI

Source: https://www.prosperconsulting.ai/insights/future-proofing-a-legacy-finance-firm

> Customer Profile A well-established financial services firm with a 50-year legacy, over 350 employees, and consistently strong margins. Renowned for its trust, expertise, and people-driven workflows, the firm had long been a leader in its space. But as AI-enabled competitors emerged, pressure…

## Customer Profile

A well-established financial services firm with a 50-year legacy, over 350 employees, and consistently strong margins. Renowned for its trust, expertise, and people-driven workflows, the firm had long been a leader in its space. But as AI-enabled competitors emerged, pressure mounted to adapt or risk being left behind.

## The Challenge

The CFO saw the writing on the wall: AI-driven disruptors were delivering services faster, cheaper, and at scale. While some in the leadership team were eager to act, others dismissed AI as overhyped. Without a clear strategy, the firm risked wasting time and resources. Or worst case: losing its competitive edge.

## The Turning Point

Prosper AI Consulting was engaged to provide clarity and a roadmap. Through our AI Readiness and Adoption Programme, we guided the leadership team from uncertainty to alignment, ensuring they understood both the risks of inaction and the opportunities AI offered. We helped them shift from reactive anxiety to proactive strategy.

## The Impact

Within six weeks:

- Leadership aligned on a strategic AI vision
- Six senior leaders achieved practical AI literacy
- A company-wide AI Manifesto and Adoption Plan was developed, including training, budgeting, and risk frameworks

...and within four months:

- A 7% productivity boost was achieved, without increasing headcount
- Only 15% of the AI budget was spent on technology; 85% focused on people, process, and capability
- The business transformed its mindset. AI is no longer an abstract concept but a central driver of future value

The CFO summed it up:

> "Without Prosper, we'd have been tinkering with AI tools without seeing the bigger picture. Now, we're not waiting to be disrupted, we're leading the change in our sector."

## Future Steps

- Expand AI adoption across workflows and departments
- Continue building internal AI capability through targeted training
- Leverage AI to grow market share and sustain competitive margins

## Ready to Future-Proof Your Business?

Prosper AI Consulting equips leadership teams to navigate AI with clarity, confidence, and a roadmap tailored to your business.

---

# Understanding Competence, and why it matters

Source: https://www.prosperconsulting.ai/insights/understanding-competence

> Introduction: Why Competence Matters More Than Curiosity The growing presence of generative AI tools like ChatGPT in the workplace has sparked a wave of experimentation. Business users are drafting emails, summarising documents, generating product ideas, and even scripting client communications,…

## Introduction: Why Competence Matters More Than Curiosity

The growing presence of generative AI tools like ChatGPT in the workplace has sparked a wave of experimentation. Business users are drafting emails, summarising documents, generating product ideas, and even scripting client communications, all with the help of AI. In some teams, it feels like an innovation breakthrough. In others, it already looks like chaos.

But what separates productive AI use from risky overreliance? The answer, more often than not, lies in the user's level of competence, not just their enthusiasm.

To make sense of how people develop skill with generative AI, it helps to revisit a classic model from educational psychology: the Four Stages of Competence. Originally developed to describe how people learn new skills, this framework remains one of the most useful tools for understanding user behaviour, risk, and potential. Before diving into AI-specific examples, it's worth talking about the model itself.

## The Four Stages of Competence: A Learning Model That Explains More Than You Think

The Four Stages of Competence describe how individuals move from ignorance to mastery in any skill-based area. The framework is simple, but its explanatory power is significant.

1. **Unconscious Incompetence:** The individual does not understand or recognise their lack of skill. They are unaware of the risks or gaps in their knowledge and often overestimate their competence.

2. **Conscious Incompetence:** The individual becomes aware of their limitations. They recognise what they do not know and begin seeking improvement.

3. **Conscious Competence:** The individual has developed a level of skill and can apply it effectively, though they still need to think deliberately about how to do it.

4. **Unconscious Competence:** The skill becomes second nature. The individual can perform tasks automatically, relying on experience and intuition without conscious effort.

This model applies to everything from driving a car to delivering a keynote. It's especially useful in rapidly evolving areas, where new tools create a perception of immediate ability, even when real competence lags far behind.

**Where you are in your journey, not important... KNOWING where you are in your journey IMPORTANT**

And this is precisely what we're seeing now with generative AI.

## Applying the Model: Competence as the Missing Piece in AI Adoption

Generative AI is easy to access and use. It doesn't require code. There's no installation. You type something in, and something remarkably fluent comes back. That fluency gives many users the false sense that they are using the tool well. In reality, the risks, from factual inaccuracy to reputational damage, often remain invisible until something goes wrong.

This is where the Four Stages of Competence become critical. By understanding where users sit on this progression, businesses can anticipate the challenges, guide skill development, and avoid some of the more predictable failures we've already seen in the early days of AI adoption.

Let's now examine each stage in turn, starting with the most dangerous of all: unconscious incompetence.

## Stage One: Unconscious Incompetence - Confident Without Cause

The first stage is defined by ignorance, but not in a pejorative sense. The problem is not that users are unintelligent or unmotivated. It is simply that they do not yet understand what they do not know. They have jumped into generative AI with enthusiasm, but without the knowledge required to use it responsibly.

### Typical Behaviour

At this level, users are heavily influenced by the hype surrounding AI. They might describe tools like ChatGPT as "amazing" or "game-changing" and apply them across a wide range of tasks without question. Common behaviours include:

- Trusting AI outputs without verification.
- Sharing sensitive or proprietary data in public AI platforms.
- Using AI-generated text verbatim in client documents or public-facing material.

What drives this behaviour is the illusion of competence. Because the AI's responses are fluent, users assume they are accurate. The language is persuasive. The interface is simple. There is no immediate feedback loop to suggest anything is amiss, until someone notices an error, or worse, a breach.

*This is like a learning musician.. you might think you are in time, and sound great, until...*

### The Real-World Consequences

The risks at this stage are both significant and well documented. For instance:

- **Data leaks:** Samsung engineers inadvertently submitted proprietary code to ChatGPT. The tool stored the data, prompting the company to ban internal use and reassess its approach to AI access.
- **Compliance violations:** Financial institutions including JPMorgan and Goldman Sachs have restricted ChatGPT usage due to concerns that employees might enter confidential information without understanding the implications.
- **Legal failures:** A US lawyer famously submitted a court brief generated by ChatGPT that cited fictional legal cases. The court sanctioned the lawyer and criticised the firm's lack of basic fact-checking.

These are not fringe cases. They are examples of exactly what happens when people operate with high confidence and low awareness.

### A Question of Culture, Not Just Knowledge

Stage One users are not reckless by nature. Often, they are simply uninformed. They have not received training, they are unaware of the risks, and they may not yet have been corrected or coached. In some cases, early enthusiasm is encouraged by leadership, particularly when the focus is on productivity or innovation.

This is why unconscious incompetence is as much a cultural issue as a technical one. If the environment rewards speed over scrutiny, or experimentation without support, Stage One behaviour can flourish unchecked.

### What's Worth Preserving

Despite the dangers, there is some value in this stage, specifically, the curiosity and momentum that users bring. These are the early adopters. They are not afraid to try new tools. Their experiments, while often flawed, can surface ideas worth refining. In safe, low-stakes contexts, they may even achieve quick wins, such as drafting emails faster or summarising meeting notes.

The goal is not to shut this behaviour down, but to channel it. Businesses can extract valuable insight from what users try to do, even when they don't yet know how to do it properly.

### The Shift That Must Happen

The key challenge at this stage is to move users from unconscious incompetence to conscious incompetence. In other words, they need to understand that they don't yet know enough to use AI safely. Once they reach that point, they become coachable. They begin asking questions, seeking guidance, and recognising where they need support.

This is where structured training, clear usage policies, and peer learning become essential. Many organisations have responded to Stage One incidents with reactive bans or restrictions. A better approach is to acknowledge the gap and help users close it.

## Stage Two: Conscious Incompetence - Aware, Cautious, and Teachable

The second stage is where progress begins. Users have had a wake-up call. Perhaps they made a mistake, saw an example of misuse, or received basic training that introduced the limits of generative AI. Whatever the trigger, they now understand that they lack full competence and act more cautiously as a result.

### Behavioural Traits

At this stage, users are still exploring, but their attitude has shifted. Confidence is now tempered by awareness. They ask more questions. They second-guess AI outputs. They might rerun a prompt to see if it produces different results or paste an answer into Google to cross-check it.

In practice, a Stage Two user might use ChatGPT to generate a first draft of a document, but they will thoroughly review and rewrite it before sharing. They know the tool can make mistakes. They are beginning to understand how to spot them. Common habits at this stage include:

- Testing the AI with known questions to assess reliability.
- Avoiding high-stakes use cases, such as client-facing content or financial reporting.
- Discussing AI use openly with colleagues, often by sharing both successes and cautionary tales.

### The Risks at This Stage

The bluntest risks from Stage One are now reduced. Users no longer trust AI blindly and are more likely to catch obvious errors. However, subtler risks remain. These include:

- Overcorrecting by becoming too risk-averse, and therefore underusing AI even in safe scenarios.
- Relying on laborious manual checks that cancel out time-saving benefits.
- Missing nuanced mistakes because they do not yet know how to prompt effectively or verify efficiently.

It is also common to see inconsistency in usage. A user might have a good experience one day and a frustrating one the next, depending on how well they've framed their request or understood the context of the AI's reply.

### Opportunities for Development

This is the most teachable stage. Users are ready to improve. They are seeking better prompts, asking for advice, and keen to apply lessons. Businesses should support this with structured learning: training workshops, AI usage guidelines, and safe environments to experiment without fear of consequence.

At this point, users begin identifying where AI is helpful and where it is not. They might decide it's ideal for brainstorming or summarising, but not for final wording or factual accuracy. That growing discernment is a sign of progress. It's no longer "AI can do anything" but rather "AI can help with this part, and I'll handle the rest."

In team settings, these users often share what they've learned. Peer-to-peer learning becomes a valuable channel. One person might show a colleague how to prompt for a more structured response, while another shares a workaround to avoid repetitive phrasing.

### Business Value Begins to Emerge

Although cautious, these users are beginning to generate value. They might use ChatGPT to speed up research or create outlines for reports. The gains are still moderate, but they are consistent and increasingly reliable. Importantly, most of the major risks are now mitigated through their own awareness and checks.

The challenge at this stage is avoiding stagnation. Some users may remain overly cautious or lose motivation if results are inconsistent. The role of leadership here is to encourage continued experimentation, provide accessible examples of effective use, and reward good judgment rather than just productivity.

## Stage Three: Conscious Competence - Effective, Reliable, and Responsible

This is the stage where real benefit starts to scale. Users now have practical skills, and they know how to use generative AI thoughtfully. They are confident, but not careless. They no longer just avoid mistakes, they actively design their process to get the best from the tool.

### Behavioural Traits

Stage Three users treat AI as a useful but imperfect assistant. They understand the system's strengths and limitations. They approach tasks with a method. For example, they might:

- Break tasks into components and assign appropriate parts to AI.
- Use structured prompts to control format and tone.
- Apply a checklist to review AI outputs before publishing or sharing.

They also move beyond general-purpose queries. Instead of asking ChatGPT to "write a product summary," they might say, "Based on the product sheet below, summarise the key features in 100 words, using UK English, and avoiding technical jargon." Their prompts are deliberate and their reviews are swift. They know what they are expecting, and they know when the output is plausible, as opposed to credible.

### Risks Become Subtle but Still Present

Errors are now rare, but not impossible. A Stage Three user might miss a nuanced factual inaccuracy or let a polished-sounding response through without deep verification. There is also the risk of overextending into areas they know less well, where their ability to spot issues may be weaker.

Another organisational risk is bottlenecking. When only a few people in a team are truly competent, others may defer too much. This can create pressure on the "AI-capable" staff and slow down wider adoption.

### Opportunities for Broader Use

With these users, AI becomes a dependable tool. Time savings are significant. Quality is often improved. For example, a policy draft might go from blank page to finished document in half the time, with better structure and tone.

Stage Three users are also well-positioned to teach others. They can support Stage Two colleagues by sharing prompt templates, use cases, or known pitfalls. Their methods may still be conscious and sometimes manual, but they are proven.

Businesses benefit most when Stage Three usage becomes normal across departments. This requires internal documentation, policy updates, and perhaps a shift in roles, recognising those who can lead on AI enablement or integrate it into team workflows.

## Stage Four: Unconscious Competence - Integrated, Intuitive, and Strategic

At this stage, users are no longer consciously managing their AI usage. They work fluidly, applying best practices almost automatically. Generative AI is embedded into their workflow with little friction, and they routinely deliver high-quality outputs with minimal risk.

### Behavioural Traits

These individuals use AI as naturally as they would a spreadsheet or search engine. They do not need to pause and think about how to prompt or verify, they just do it, instinctively. They:

- Combine AI with other tools or processes for greater efficiency.
- Anticipate AI errors before they happen and adjust prompts accordingly.
- Adapt to new features or models with minimal learning curve.

For example, an experienced manager might routinely use ChatGPT to prepare briefing notes, draft comms, generate ideas, or simulate customer queries. They do so without hesitation, knowing where to let AI assist and where to retain human control.

### Risks Are Low but Still Exist

The biggest risk at this stage is complacency. A highly competent user might not realise that a new AI model behaves differently or that a recent regulatory update affects how data should be handled. Their habits, while refined, may not always align with new conditions.

Another risk is knowledge being locked in individuals. When competence becomes intuitive, it can be harder to explain. If an expert leaves the organisation, their approach might go undocumented.

### Opportunities for Strategic Advantage

This is where AI moves beyond productivity and starts to reshape strategy. Expert users can help redefine processes, advise on tool selection, and mentor others. They can spot where AI could create new services or unlock previously unattainable insights.

Some organisations develop an internal AI centre of excellence at this point, led by Stage Four users. Others embed them into product, operations, or client-facing roles, using their expertise to scale responsible AI adoption across the business.

The goal now is to ensure that their skill spreads, rather than becomes a silo. Businesses should invest in knowledge transfer and tool development to codify these practices into broader systems and training.

## Final Reflections: AI Competence is a Journey, Not a Status

The Four Stages of Competence give us more than a way to categorise users. They offer a blueprint for how to move people from curiosity to capability, from risk to reliability.

Early adopters may act with confidence but without understanding. That's not a reason to shut them down, it's a reason to support them. Awareness leads to learning. Learning leads to control. And control opens the door to real productivity and innovation.

As with any new skill, AI competence grows over time. It requires exposure, practice, feedback, and reflection. But with the right structure and support, business users can evolve into thoughtful, capable practitioners who use AI not just effectively, but responsibly.

Organisations that recognise this journey, and design their AI adoption programmes around it, are far more likely to succeed. Not just by avoiding the obvious risks, but by unlocking the full creative and strategic potential of this emerging technology.

---

# The Fast Follower Advantage: Why Mid-Sized Firms Should Stop Chasing AI and Start Catching Up Smartly

Source: https://www.prosperconsulting.ai/insights/the-fast-follower-advantage

> Why Mid-Sized Firms Do Not Need to Lead the AI Race, but Do Need to Catch Up Over the last two years, we have worked closely with dozens of mid-sized organisations to understand how they are responding to the acceleration of artificial intelligence. While headlines focus on disruption and…

## Why Mid-Sized Firms Do Not Need to Lead the AI Race, but Do Need to Catch Up

Over the last two years, we have worked closely with dozens of mid-sized organisations to understand how they are responding to the acceleration of artificial intelligence. While headlines focus on disruption and innovation, most mid-sized firms are not trying to lead the AI revolution. They are focused on something far more practical: keeping up.

But keeping up is not easy in a market flooded with new AI tools, emerging platforms, and growing pressure to act quickly. In this environment, businesses can fall into reactive decision-making or get lost in a cycle of pilots that never scale.

There is a better way. It is called the fast follower approach.

## The Fast Follower Defined: Strategic Timing Over Early Adoption

Fast followers are not late adopters. They are organisations that monitor the AI landscape, observe what is working, and then act decisively once the value is proven. They do not rush to experiment with every tool. Instead, they wait for evidence, align adoption to strategy and business case, and scale only what fits.

The firms that succeed over time are not always the first to move. They are the ones who get the timing right.

This approach is particularly effective for mid-sized businesses that must balance innovation with focus and resource constraints.

## Why the Fast Follower Strategy Works for Mid-Sized Firms

Mid-sized companies often lack the AI labs, global tech budgets, and in-house machine learning expertise that large enterprises enjoy. However, they do possess other strengths. These include operational focus, domain depth, and the ability to move quickly once priorities are clear.

We help these firms turn those strengths into a structured AI capability by applying five core principles.

## Building Fast Follower Capability: Five Core Principles

**1. Track the Right Signals**

We help your team focus on what is actually working in your sector. This means identifying real use cases with strong ROI and avoiding distraction from vendor marketing or speculative trends.

**2. Build a Cross-Functional Evaluation Group**

AI affects more than just technology teams. We bring together leaders from operations, finance, IT, and strategy to evaluate AI through every lens that matters.

**3. Create a Lean Pilot Framework**

We design pilots that are small in scope, focused in purpose, and structured to deliver measurable insights. This helps avoid overengineering or wasting time on unclear experiments.

**4. Prepare the Organisation**

We support internal alignment by educating teams, clarifying roles, and helping your organisation understand what changes are coming and why they matter.

**5. Define Smart Thresholds for Action**

We work with you to set clear criteria for when to move. This includes defining what success looks like, identifying when data is ready, and ensuring your people are prepared. When the right conditions are in place, your team can act without hesitation.

## What Fast Followers Do Differently

Not every organisation that waits is a fast follower. The difference lies in preparation, structure, and execution.

Fast followers:

- Build foundational AI literacy across leadership
- Apply rigorous frameworks to evaluate tools and vendors
- Create cultures that are open to change and innovation
- Know when to scale, and when to stop
- Act decisively and swiftly when the opportunity arises

They avoid jumping on every trend. Instead, they act when it matters.

## Why Fast Does Not Mean Reckless

Speed alone is not a strategy. The goal is to move at the right time with the right foundation in place. Fast followers succeed because they invest in the structure, mindset, and governance needed to make confident decisions.

At Prosper, we work with leadership teams to build that readiness. The outcome is a company that can adopt AI without confusion, risk, or wasted effort and that starts to realise measurable value faster than its peers.

## Conclusion: Catching Up, Smartly

Success in the AI era is not about being first. It is about being ready. It is about understanding where AI fits, preparing the organisation to act, and moving with purpose when the opportunity is clear.

If your business is ready to stop chasing trends and start catching up smartly, we are here to help.

---

# Old Rules for New Tools: Engineering Principles That Make Agents Work

Source: https://www.prosperconsulting.ai/insights/old-rules-for-new-tools

> We are rushing to a development pattern where multi-agent systems, collections of autonomous yet collaborating AI entities, are becoming increasingly critical to business operations. As organisations race to implement these powerful new tools, the question facing CTOs and engineering leaders is…

We are rushing to a development pattern where multi-agent systems, collections of autonomous yet collaborating AI entities, are becoming increasingly critical to business operations.

As organisations race to implement these powerful new tools, the question facing CTOs and engineering leaders is not just whether to adopt agentic systems, but how to build them in ways that ensure long-term maintainability, reliability and scalability.

Have we been here before...

## Bringing Software Engineering Wisdom to a New Domain

The challenges we face in multi-agent system development are not entirely novel. For decades, software engineering has refined principles that promote clean, maintainable code and robust systems. As we move into this 'new territory' of interconnected AI agents, these time-tested principles offer valuable guidance that can help us avoid repeating the mistakes of the past.

## The Value of DRY in Multi-Agent Architectures

DRY (Don't Repeat Yourself) represents one of the most fundamental principles in software engineering. At its core, this principle asserts that every piece of knowledge or logic should exist in exactly one place within a system. When applied to traditional development, this prevents duplication of code. In multi-agent systems, however, the principle extends beyond mere code reuse.

Consider how many organisations currently implement AI agents: separate chatbots handling customer service, data analysis, content generation and other tasks, each built as isolated systems. This approach inevitably creates redundant components that handle similar concerns, such as authentication, logging, error recovery and context management.

By applying DRY to agent architecture, we can centralise these common functionalities into shared libraries or frameworks. This centralisation ensures that improvements to core components propagate throughout the entire system automatically. When we update our error handling approach or enhance security protocols, all agents benefit simultaneously rather than requiring individual updates.

For technical leaders, embracing DRY means investing in reusable infrastructure that accelerates development while reducing maintenance costs. It means creating common foundations that allow your teams to focus on creating value through unique agent capabilities rather than reinventing core components.

The same principle can be applied to its often forgotten counterpart, data.

## SOLID Foundations for Flexible Agent Systems

The SOLID principles have guided object-oriented programming for years, but their relevance extends remarkably well to agent-based systems. Each principle offers specific guidance on creating robust multi-agent architectures:

**Single Responsibility Principle (SRP):** This principle suggests that each agent should focus on doing one thing exceptionally well. Rather than building monolithic "super agents" that attempt to handle numerous tasks, SRP encourages us to develop specialised agents with clear, focused responsibilities.

For instance, rather than tasking a single agent with both retrieving information and formulating responses, we might separate these concerns: one agent specialises in efficient information retrieval while another excels at crafting responses based on that information. This specialisation makes each agent simpler to develop, test and maintain.

Of course we learnt that when we do this, we need to have discipline, and consider the orchestration and flow.

**Open-Closed Principle (OCP):** Perhaps the most powerful principle for evolving systems, OCP states that software entities should be open for extension but closed for modification. In practical terms, this means designing interfaces and protocols that allow new capabilities to be added without altering existing components.

In a multi-agent context, this might mean creating standardised ways for agents to communicate that can accommodate new types of messages or data without requiring updates to existing agents. Agent orchestrators designed with OCP in mind can incorporate new specialist agents without requiring rewrites of core coordination logic.

This leans well into the emerging MCP (Model Context Protocol).

**Liskov Substitution Principle (LSP):** This principle ensures that objects can be replaced with instances of their subtypes without altering program correctness. For agent systems, this translates to the ability to seamlessly upgrade or replace individual agents as long as they conform to the expected interfaces. It should be possible to change models, the underlying systems should be isolated from that.

LSP enables gradual evolution of agent capabilities. You might start with a simple retrieval agent using basic keyword matching, then later replace it with a more sophisticated semantic search agent, all without disrupting the overall system behaviour.

**Interface Segregation Principle (ISP):** ISP advises that clients should not be forced to depend on interfaces they don't use. For agent systems, this means designing minimal, purpose-specific interaction protocols rather than creating broad, catch-all interfaces.

Following ISP leads to cleaner separation between agent responsibilities and more focused communication patterns. Agents that handle document processing shouldn't need to understand interfaces designed for customer interaction, for example.

Interfaces to AI Agents, can be over specified. There is value in specific managed type, (pydantic) but there is also value in 'letting it be language', after all these are Language models.

**Dependency Inversion Principle (DIP):** This principle states that high-level modules should not depend on low-level modules; both should depend on abstractions. In agent systems, this means orchestration layers should communicate with agents through well-defined interfaces rather than relying on specific implementation details.

By embracing DIP, organisations can create flexible architectures where agents can be developed independently by different teams or even sourced from different vendors, all while maintaining system cohesion through agreed-upon abstractions.

## YAGNI: Restraint as a Design Virtue

You Aren't Gonna Need It (YAGNI) provides a valuable counterbalance to our natural tendency toward overengineering. As technologists, we often anticipate future requirements that may never arise, leading to unnecessary complexity and bloated systems.

In agent development, YAGNI reminds us to focus on current, proven needs rather than hypothetical future capabilities. Instead of building agents with extensive features "just in case," start with minimal viable functionality and expand based on actual usage patterns and feedback.

This principle is particularly important given the constant updates of foundation models and agent frameworks. Today's elaborate custom implementations may be rendered obsolete by tomorrow's out-of-the-box capabilities. By adhering to YAGNI, organisations can remain nimble, adapting to emerging technologies rather than being constrained by premature investments in complex custom solutions. Keep it cheap enough, and safe enough to change.

Vibe coding is not going to follow YAGNI, tread carefully.

## TDD and Evals: Ensuring Agent Reliability

Test-Driven Development (TDD) has changed how we build reliable software, and its principles apply equally to agent development. However, testing AI agents presents unique challenges that require adapted approaches.

In traditional TDD, developers write tests before implementing features, using tests to drive design decisions. For agent systems, we can follow a similar pattern using evaluation frameworks (evals) that assess agent performance across various dimensions.

Evals define expected behaviours and outputs for specific inputs or scenarios. They might test an agent's ability to correctly extract information from documents, generate appropriate responses to customer inquiries or collaborate effectively with other agents. By establishing these expectations upfront, teams can focus their development efforts on meeting well-defined criteria rather than chasing vague notions of "intelligence."

The tests need to consider not just abstract assertions, pass or fail, but acceptable margins. For example responses will be language. Language can have (and you want it to have nuance), this means that a clear approach to scoring and assertions needs to be defined.

For CTOs and engineering leaders, investing in robust eval frameworks pays dividends in reliability and predictability. These frameworks establish guardrails that ensure agents behave as expected even as their underlying models or components evolve. They also provide objective measures of improvement, allowing teams to demonstrate clear progress in agent capabilities over time.

## Standardised Interfaces: The Communication Layer

While internal design principles guide how we structure individual agents and systems, standardised interfaces determine how agents interact with each other and external systems. The right protocols and standards enable seamless integration whilst promoting consistency and reliability.

### RESTful Architectures and OpenAPI

RESTful architectures have become the default approach for web service design due to their simplicity, scalability and statelessness. These same qualities make REST an excellent foundation for agent interactions, particularly when agents need to communicate with existing enterprise systems.

By defining agent interfaces using OpenAPI specifications, organisations create clear contracts that simplify integration and enable automated testing. These specifications serve as living documentation, helping developers understand how to interact with agents without diving into implementation details.

For organisations with significant investments in REST-based infrastructure, leveraging these established patterns for agent communication minimises disruption and accelerates adoption. Existing monitoring, security and management tools can be extended to cover agent interactions with minimal modification.

### Model Context Protocol and JSON-RPC2

As agent systems grow more sophisticated, new protocols tailored to their unique needs are emerging. Anthropic's Model Context Protocol (MCP) represents one such advancement, designed specifically to address the contextual needs of large language models in agent systems.

MCP allows AI systems to maintain context as they interact across diverse external systems, eliminating the need to build custom connectors for each new integration. This contextual awareness is crucial for agents that need to synthesise information from multiple sources while maintaining a coherent understanding of user intentions and system state.

Similarly, JSON-RPC2 offers a lightweight alternative to REST for agent-to-agent communication. Its procedural approach often maps more naturally to agent interactions than REST's resource-oriented model, particularly for complex operations that don't fit neatly into CRUD paradigms.

Together, these emerging protocols enable:

- Context preservation across system boundaries
- Standardised error handling and response formatting
- Simpler integration patterns for agent-to-agent communication
- Low implementation foot print, cloud native

By embracing these protocols alongside traditional REST approaches, organisations can select the right communication pattern for each interaction type whilst maintaining overall system coherence.

## Practical Implementation: Applying Principles to Practice

Moving from theoretical principles to practical implementation requires careful consideration of how these concepts interact in real-world systems. Let's explore some concrete approaches to applying these principles in multi-agent architectures.

### Building a Common Agent Framework

To truly embrace DRY principles in agent development, organisations should invest in creating a common framework that handles cross-cutting concerns. This framework might include:

- Standardised logging and telemetry collection
- Unified authentication and authorisation mechanisms
- Common error handling and recovery strategies
- Shared context management and session persistence
- Centralised configuration and deployment tooling

By providing these capabilities as reusable components, the framework frees agent developers to focus on domain-specific logic rather than infrastructure concerns. This separation of concerns accelerates development whilst ensuring consistent behaviour across all agents in the system.

The framework should expose well-defined interfaces that adhere to SOLID principles, allowing teams to extend or override default behaviours when needed without modifying core components. This flexibility enables specialisation whilst maintaining system cohesion.

### Creating an Agent Orchestration Layer

Multi-agent systems require coordination to function effectively as a cohesive whole. An orchestration layer provides this coordination, routing requests, managing agent lifecycles and ensuring proper sequencing of operations.

Following the Dependency Inversion Principle, this orchestration layer should depend on abstractions rather than concrete agent implementations. Agents should register their capabilities through standardised interfaces, allowing the orchestrator to discover and utilise their services without hard-coded dependencies.

This approach enables:

- Dynamic composition of agent workflows based on available capabilities
- Graceful degradation when specific agents are unavailable
- Seamless introduction of new agent types without orchestrator changes
- A/B testing of alternative agent implementations for specific functions

The orchestration layer serves as the practical embodiment of our design principles, providing the flexibility and extensibility needed for evolving agent ecosystems.

### Implementing Comprehensive Evaluation Systems

Reliable agent systems require comprehensive testing beyond traditional unit and integration tests. A robust evaluation framework might include:

- Functional testing of individual agent capabilities
- Interaction testing between collaborating agents
- Performance testing under various load conditions
- Security testing for potential vulnerabilities
- Fairness and bias assessments for user-facing agents
- Drift detection to identify model degradation over time

These evaluations should run continuously as part of the development and deployment process, catching regressions early and providing confidence in system reliability. By embracing TDD principles and defining these tests before implementing features, teams ensure that their implementations directly address actual requirements rather than hypothetical use cases.

The evaluation framework itself should follow DRY and SOLID principles, providing reusable components for common testing patterns whilst allowing extension for domain-specific assessments. This approach ensures consistent evaluation methodologies across different agent types whilst accommodating unique requirements.

## The Path Forward: Evolutionary Design for Agent Systems

As we navigate this new frontier of multi-agent systems, it's important to recognise that our approaches will necessarily evolve as we gain experience and as underlying technologies mature. Rather than attempting to design perfect systems upfront, successful organisations will embrace evolutionary architecture principles that allow systems to adapt over time.

This evolutionary approach aligns perfectly with the principles we've discussed:

- DRY encourages creation of reusable components that can evolve independently
- SOLID enables extension without modification as new requirements emerge
- YAGNI keeps systems lightweight and malleable by avoiding premature complexity
- TDD and evals provide confidence that changes don't break existing functionality

By combining these foundational principles with standardised interfaces and protocols, organisations can create multi-agent systems that deliver immediate value whilst remaining adaptable to future needs and technologies.

## Conclusion

The integration of established coding principles with emerging protocols provides a robust foundation for developing multi-agent AI systems that are both powerful and maintainable. By centralising common functionalities, ensuring each agent has clear responsibilities, and rigorously testing interactions, technical leaders can create systems that deliver immediate business value whilst building a platform for future innovation.

For CTOs and engineering leaders, the message is clear: the principles that have guided software development for decades remain relevant in this new era of AI agents. By applying DRY, SOLID, YAGNI and TDD alongside appropriate communication protocols, organisations can avoid the maintenance nightmares and technical debt that often accompany rapid technological adoption.

The future of multi-agent systems lies in balancing sophisticated AI capabilities with sound engineering practices. This balanced approach delivers systems that are not just intelligent but also reliable, maintainable and adaptable to changing business needs, truly augmenting human capabilities whilst remaining transparent and accountable.

As we continue to explore this new frontier, let us bring forward the best of our software engineering tradition whilst embracing the unique opportunities that multi-agent AI systems present. In doing so, we'll build systems that stand the test of time even as the underlying technologies continue their evolution.

---

# At the Crossroads of Culture and Code: What Mid-Sized Businesses Must Face Before AI Delivers Value

Source: https://www.prosperconsulting.ai/insights/at-the-crossroads-of-culture-and-code

> In boardrooms across the UK, a familiar conversation is playing out. Someone asks if the business is doing anything with AI. The answers are often vague. A chatbot here. A small experiment there. Perhaps someone in IT is trying out a language model. These early steps might feel like progress, but…

In boardrooms across the UK, a familiar conversation is playing out. Someone asks if the business is doing anything with AI. The answers are often vague. A chatbot here. A small experiment there. Perhaps someone in IT is trying out a language model.

These early steps might feel like progress, but they often miss the deeper truth. AI adoption is not only a technological shift. It is also an economic and social one.

## Culture Is the Operating System of Your Business

At Prosper, we have seen that every organisation runs on an invisible engine. That engine is culture.

It influences how decisions are made, how change is received, how people learn, and how risk is approached. Culture determines whether a new idea gains traction or gets buried under the weight of precedent.

It is the air people breathe at work. And when that air is stagnant, innovation cannot breathe.

## The Fog of Familiarity: Why Mid-Sized Firms Struggle

Many mid-sized firms have built their success on deep expertise, strong relationships, and trusted reputations. These are real strengths. But they can also create blindspots.

AI brings with it new language, tools, and ways of working. For firms anchored in legacy systems, hierarchical structures, and fixed role definitions, this can feel disorienting.

Instead of exploring how AI might enhance human potential, teams often pull back. Initiatives remain scattered. Interest fades. And even capable organisations find themselves stuck.

## Picture This: Two Companies at a Crossroads

Imagine two firms operating side by side.

In one, AI is mentioned occasionally but never owned. Pilots begin without follow-through. People see AI as a risk. Every step forward feels slow and uncertain.

In the other, AI is part of the organisation's future story. Leaders engage with it. Teams are invited into the conversation. Wins are shared. Learning is encouraged.

The tools are similar. The difference lies in culture and the readiness to use those tools well.

## Building a Culture That Embraces AI

This is not about cosmetic change. It is about shaping a culture that supports intelligent, responsible innovation.

- Value Learning Over Knowing. AI will continue to evolve. So must people. We help foster environments where learning is rewarded over certainty.
- Encourage Psychological Safety. Innovation involves uncertainty. We support environments where questions can be asked and experiments can happen without fear.
- Reimagine Roles. We help communicate how AI can work alongside people, not against them. This opens doors rather than closing them.
- Connect Technology to Purpose. People adopt change more readily when it connects to something meaningful. We always tie AI back to the organisation's mission and impact.

## Practical Steps for Cultural Shift

To bring these ideas into practice, we support mid-sized firms through four key steps.

- Start with Leadership Mindset. Leaders must go first. We help model curiosity and openness. If leaders do not engage, no one else will.
- Invest in Cross-Functional Storytelling. Share wins and lessons across teams. Culture shifts when stories change.
- Build Cultural Infrastructure. We help embed AI awareness into onboarding, training, and performance. We also create space for collaboration between technical and non-technical teams.
- Measure What Matters. We do not just track adoption. We track engagement. Curiosity, excitement, and involvement are leading indicators of cultural readiness.

## The Payoff: Growth and Resilience

Firms that lead with culture do more than adopt AI. They build the internal capacity to adapt.

In a world shaped by fast-moving technology, that flexibility becomes the ultimate competitive advantage. It enables firms not only to survive disruption but to drive it.

## Conclusion: Start with Culture

If your business is serious about AI, start by building the right foundation.

- Align your AI goals with business strategy
- Ensure leadership is involved and committed
- Prepare your people and evolve your processes
- Establish a repeatable framework for smart adoption

AI will not transform your business in isolation. But with the right approach, it will unlock new value.

---

# Why AI Adoption Fails: The Organisational Challenge Facing Mid-Market Businesses and What To Do About It

Source: https://www.prosperconsulting.ai/insights/why-ai-adoption-fails

> Introduction: The Unstoppable Wave of AI Artificial Intelligence (AI) is no longer a futuristic concept; it is already reshaping the core of knowledge-based work. Over the past two years, my team and I have studied, consulted with, and advised more than 100 mid-sized businesses on AI readiness,…

## Introduction: The Unstoppable Wave of AI

Artificial Intelligence (AI) is no longer a futuristic concept; it is already reshaping the core of knowledge-based work. Over the past two years, my team and I have studied, consulted with, and advised more than 100 mid-sized businesses on AI readiness, adoption, and implementation. In each engagement, ranging from marketing agencies to accountancy firms, law practices, engineering consultancies, and beyond, one observation stands out: **succeeding in the AI era is far less about the technology itself, and far more about how effectively an organisation can adapt to it**.

For years, technology in mid-market companies merely played a supporting role, generally lightening the load for human teams. However, the rise of advanced AI solutions, spanning natural language processing through to data-driven analytics and of course generative AI, has elevated technology into a leading creator of value. This shift demands a parallel transformation in how mid-sized firms operate, transitioning from **technology-supported** to **technology-centric.**

To understand why this shift is now crucial, let us first look at the competitive forces pressuring mid-sized firms from all angles.

## Why Mid-Sized Firms Are Under Threat: Three Competitive Pressures

Today's mid-market businesses, particularly those reliant on knowledge work, are feeling pressure from three main sources:

- **Start-Ups Born AI-Native.** New entrants launch with AI inherently built into their service offerings from day one. These firms often undercut or unbundle traditional services, delivering sharper solutions at lower cost. I would wager that ChatGPT is already offering more legal advice on a day-to-day basis (in terms of volume) than any single law firm.

- **Large Enterprise Competitors.** Major players, multinational corporations and global professional services giants, can invest billions into AI, significantly lowering their cost base while unlocking new revenue streams. For example, large accountancy firms now utilise advanced AI-assisted audit tools to compete for mid-market work they previously bypassed, thereby challenging mid-sized firms on both price and efficiency.

- **Technology Vendors Becoming Competitors.** Software providers, which once supplied mid-sized businesses with off-the-shelf tools, are now offering AI-powered services directly to the end market. In-house counsel are already becoming a strong market for AI infused legal drafting, researching and precedent tools that have been the traditional preserve of Law firms.

The net result? **Mid-sized companies are under immense pressure** to adopt AI or risk being outmanoeuvred by more technologically advanced rivals.

Yet, despite the urgent need to respond, simply finding the right AI tool is rarely the core issue. The deeper challenge lies within the organisation itself.

## It's Not About the Technology, It's About Organisational Transformation: Closing the Capability Gap

_"The real challenge isn't adopting AI; it's building the strategy, leadership mindset, processes, policies, and people strategies that fundamentally reshape the business to operate in an AI-driven world."_

Many mid-market leaders assume their difficulties stem from failing to discover the 'right' AI platform or tool. In reality, we witness that **technology itself is seldom the root issue**. Far more common are failings within the organisation, such as:

- **Weak Strategic Alignment.** AI initiatives are too often relegated to the IT team or treated as side experiments. Without clear goals tied to revenue, cost reduction, or client experience, adoption falters.

- **Insufficient Leadership and Cultural Readiness.** Senior leaders might still view AI as an optional extra, rather than a competitive necessity. Meanwhile, employees who fear AI's impact on their jobs may resist adoption, undermining the benefits.

- **No Repeatable Approach.** Many mid-sized firms tackle AI with a scattergun strategy, trialling a tool here, engaging a vendor there, without a robust framework to test, implement, or scale successes. This typically leads to short-lived experiments that fail to generate lasting returns.

In other words, **it is an organisation's capacity to integrate, adapt, and scale AI solutions that truly determines success**, not the specific technology. Even an excellent tool will underperform if the business cannot systematically evaluate its value, implement it properly, and encourage adoption.

How can mid-sized firms address these organisational shortcomings and catch up with the AI curve? It begins by rethinking their entire approach to technology.

## From Technology-Supported to Technology-Centric: Mapping the Shift

Traditionally, mid-sized firms defined themselves as 'people businesses', with human expertise providing the greatest value while technology took a secondary, supportive role. This model is rapidly becoming outdated in an era where AI can replicate, and even surpass, human performance in certain tasks.

## The Technology Value Creation Curve

One way to think of this shift is by imagining a curve that maps how quickly firms adopt, integrate, and derive value from new technology:

- **Left Side, "Technology-Supported".** People and their expertise drive most of the value; technology is simply an add-on or peripheral.

- **Right Side, "Technology-Centric/Driven".** Technology is embedded into the organisation's DNA, forming a continuous feedback loop with human expertise. AI does not replace humans; it complements and amplifies their capabilities.

Companies stuck on the left, treating AI as an afterthought, will progressively fall behind in productivity, cost-effectiveness, and overall competitiveness. Meanwhile, those who shift to the right side build "fast-follower" capabilities, rapidly adopting and profiting from emerging solutions.

**The key to success isn't just acquiring better technology; it's developing the organisational capability to integrate, adapt, and scale it efficiently.**

Businesses that get this right will move up the technology value creation curve seamlessly. Those that don't will remain stuck, constantly experimenting, but never truly evolving.

A fast follower isn't a company on the bleeding edge experimenting with unproven technology. Rather, it's a company that evaluates emerging technologies and adopts them when they're robust enough to fulfil a particular function. Fast followers get in early when there's sufficient evidence that the technology will create value.

By adopting early, these firms create initial value and profit, then reinvest those funds in the next advancement. This creates a self-reinforcing cycle: technology adoption leads to value creation, which funds further adoption, widening the competitive gap.

So, if becoming technology-centric is the goal, why do so many mid-sized firms stumble when they try to adopt AI? A common pitfall is the lack of a structured, strategic approach.

## Why Haphazard AI Adoption Falls Short: The Pitfalls of Ad Hoc Implementation

It is common for mid-sized firms to approach AI in a piecemeal fashion:

- **Asking a Small Team to "Play" with ChatGPT.** Without strategic oversight or clear objectives, these experiments can fall by the wayside.

- **Delegating AI to IT Alone.** AI impacts customers, business models, processes, and people strategies. Treating it as a purely technical issue ignores its wider strategic implications.

- **Underestimating AI's Potential.** If leadership only sees AI used for minor tasks such as drafting emails, they may conclude it is not transformative enough to justify major investment.

These approaches rarely produce meaningful value because they lack **executive alignment, cultural buy-in, and a structured process** for testing and scaling. The same organisational roadblocks resurface with each new piece of technology, leading to repeated underachievement.

To break out of this cycle, mid-sized firms need a systematic, repeatable framework, one that addresses strategic alignment, implementation, and cultural adoption in tandem.

## Building a System for Ongoing Success: The ODTA Framework

To truly climb the technology value creation curve, mid-sized firms need **a repeatable, systematic approach**. In developing a structured response to these recurring organisational barriers, we've worked extensively with mid-sized businesses to understand what actually enables successful, sustainable technology adoption. This led to the creation of our **Outcome-Driven Technology Adoption (ODTA)** framework. What emerged from that work is a set of eight core components that we believe are essential for any organisation, regardless of the methodology they use, if they are to make the shift from being technology-supported to truly technology-driven. These elements are not tied to ODTA alone; they represent the foundational work any mid-market firm must do to overcome the structural and cultural inertia that so often stalls progress.

ODTA core components:

1. **AI & Technology in the Business Plan.** If AI is not visible as a board-level priority, with budget, KPIs, and timelines, it will never gain real traction.

2. **High-Level AI Literacy & Leadership Commitment.** Board members and senior executives must develop enough understanding of AI to advocate for it and set realistic expectations.

3. **People Strategy for an AI Era.** Employees should be trained and reassured about the opportunities AI offers, rather than fearing displacement. Ongoing education and clear communication about AI's role in the business are essential.

4. **Process Benchmarking & Evaluation.** You need a clear baseline for existing workflows to assess the benefits AI can bring, from cost savings to enhanced quality.

5. **Vendor Evaluation & Business Cases.** Establish a consistent method for vetting AI solutions, covering strategic alignment, technical feasibility, ROI potential, and ease of integration.

6. **Project Greenlighting & Governance.** Create transparent criteria for deciding which AI projects receive funding, how they are monitored, and how results will be measured.

7. **Structured Adoption & Rollout.** Pilot new AI tools in controlled conditions, measure the outcomes rigorously, and scale up successes. Communicate wins widely across the organisation.

8. **Data Audit & Strategy.** AI thrives on high-quality data. Reviewing how your organisation collects, cleans, stores, and governs data is vital before rolling out AI.

By embedding these elements into business-as-usual processes, you can transition away from disjointed technology experiments and towards a deliberate, scalable strategy.

Even with a structured framework, some leaders remain hesitant. What if AI's rapid progression stalls, or the return on investment isn't immediate? Let's explore why that worry is largely unfounded.

## Why There's No Downside to Becoming Technology-Centric

Leaders sometimes worry about committing significant resources to AI, asking what happens if AI's rapid evolution hits a plateau. The reality is that:

1. **Existing AI Capabilities Are Already Game-Changing.** Even incremental progress in AI can bring substantial productivity gains across professional services, manufacturing, and administrative processes.

2. **Organisational Strengths Are Multi-Purpose.** The processes, governance frameworks, and cultural openness developed to accommodate AI will continue to serve your business well, regardless of how AI technology evolves in the future.

Put simply, **the greater danger lies in failing to adapt**. As soon as a competitor develops a robust approach to AI adoption, they will gain productivity advantages, pass on cost savings to clients, and invest profits into further innovation. Falling behind in this cycle can be devastating for a mid-sized business.

Having established both the urgency and the framework for AI-driven change, the final step is to take decisive action before the competitive gap widens further.

## Conclusion: Your Next Steps Before the Gap Widens

Mid-sized knowledge-work firms stand at a pivotal juncture. The future will not be shaped by the number of employees a company has, or by its traditional reputation, but by **how swiftly and intelligently it adopts and integrates AI**.

**If you lead or work within a mid-sized organisation, here are practical steps to take now**:

1. **Secure Senior Leadership Buy-In**: Build your AI literacy and then imagine what AI will make possible and the changes it is likely to create over the next 1, 3 and 5 years for your sector and business. Leadership teams quickly realise the need to act and prioritise following this exercise. Technology must become a board-level strategic priority with clear executive sponsorship.

2. **Assess Your Position on the Technology Value Creation Curve**: Objectively determine where you are on the curve and identify your organisational capability gaps. Resolve to fix these foundational issues before racing to roll out new technology.

3. **Adopt a Structured Framework Like ODTA**: Whether you choose ODTA or another structured methodology, implement clear processes to identify, test, measure, and roll out technology solutions effectively. This provides the scaffolding for consistent success rather than isolated wins.

4. **Invest in People and Processes**: The most advanced AI in the world will not fix a broken organisation. Upskill teams, map critical processes, and establish governance that enables rather than hinders AI adoption. Actively share technology successes internally to build confidence, buy-in, and momentum.

Whilst it may be tempting to hold off until AI matures further, your competitors, be they lean start-ups or global powerhouses, are already embedding AI at the heart of their operations. Ultimately, **the winners in the coming era will be those who treat AI not as an optional tool, but as an engine of organisational transformation**. By moving decisively now, developing your AI strategy, aligning your teams, and creating the right processes, you can elevate your mid-sized firm to become truly technology-centric and ready to capitalise on whatever innovations the future may bring.

---

# Reimagining Software Delivery: One Human, Many AI Agents

Source: https://www.prosperconsulting.ai/insights/reimagining-software-delivery

> Introduction: A Vision of Future Development The intersection of software development and AI agents presents an area ripe for exploration and innovation. Traditional governance practices, often designed for human-centric workflows, demand reimagining in an era where AI teams can handle operational…

## Introduction: A Vision of Future Development

The intersection of software development and AI agents presents an area ripe for exploration and innovation. Traditional governance practices, often designed for human-centric workflows, demand reimagining in an era where AI teams can handle operational delivery at unparalleled speed and efficiency.

This article explores these transformative possibilities, shifting between narrative and editorial perspectives to illustrate how AI-driven software delivery can evolve traditional frameworks. It examines how agile methodologies like Scrum and SAFe can adapt for AI teams, the new roles humans play in this paradigm, and the broader implications of building software at AI speed.

While Sarah's story serves as a grounding narrative, this article expands into a broader thought leadership discussion, exploring the potential strategies and challenges organisations might face as they move towards this future. By blending practical insights with visionary thinking, it offers readers both inspiration and actionable pathways for embracing AI-driven development.

## Sarah's Story

Sarah Chen never set out to revolutionise software development. As a seasoned Product Manager with fifteen years of experience at companies ranging from nimble startups to tech giants, she had grown intimately familiar with the challenges of traditional software delivery. The endless coordination meetings, the communication bottlenecks, the constant balance between speed and quality - these were simply accepted as inherent costs of building software at scale. Or so everyone thought.

Now, as she leads product development at FastTrack, a rapidly growing e-commerce platform, Sarah is pioneering something radically different: a development organisation where she serves as the sole human leader, orchestrating teams of AI agents that handle everything from code generation to testing and deployment.

## The Evolution of Software Development

Sarah's journey toward AI-driven development began with a crisis. FastTrack was growing exponentially, but their development processes couldn't keep pace with market demands. Despite having talented teams and well-established Agile practices, they were struggling with the same issues that plague many software organisations: long development cycles, communication overhead, and the constant challenge of maintaining consistency across multiple teams, how to address low value long tails of requests and technical debt.

"I remember the moment it clicked," Sarah recalls, sitting in her office overlooking Northampton's bustling tech corridor (it's not the biggest corridor). "We had just finished another exhausting day of sprint planning meetings, and I realised we were spending more time coordinating work than actually doing it. There had to be a better way."

Instead of a calendar packed with stand-ups and planning sessions, she sees a clean, data-rich interface showing the progress of her AI development teams. Over the past 48 hours - the equivalent of a full sprint in this new model - her AI agents have designed, implemented, and tested three new features for the checkout experience. The system is requesting her strategic input on two product decisions and flagging one potential ethical consideration regarding user data handling.

This isn't far science fiction, it's getting quite near.

As AI capabilities advance, we're approaching a future where a single human Product Manager could effectively orchestrate an entire development organisation staffed primarily by AI agents.

But this raises intriguing questions: Could established frameworks like Scrum and SAFe be adapted to structure this new way of working? How would traditional ceremonies and artefacts evolve? Most importantly, could this approach actually deliver better software faster?

## From Human Teams to AI Agents

Traditional software development relies on human expertise at every level: developers writing code, testers validating functionality, architects designing systems, and product owners refining requirements. Each role requires significant coordination, leading to the adoption of Agile methodologies to manage complexity and enable iterative delivery.

The introduction of AI development agents fundamentally shifts this paradigm. Instead of coordinating human schedules and managing communication overhead, we can create a seamlessly integrated system of specialised AI agents, each focusing on specific aspects of the development lifecycle:

- **AI Product Owner Agents**: Transform strategic goals into detailed user stories and acceptance criteria
- **AI Developer Agents**: Generate, review, and optimise code
- **AI Tester Agents**: Create comprehensive test scenarios and validate functionality
- **AI Architect Agents**: Maintain technical standards and ensure system coherence
- **AI UX Agents**: Design and validate user interfaces
- **AI Integration Agents**: Manage code merges and deployments

These agents don't just work faster - they work differently. They share context through vector databases, maintain perfect recall of previous decisions, and operate continuously without the natural breaks human teams require.

They still need to have a co-ordination, governance and control structure. Ironically as we are trying to get agents to mimic and integrate into human practices, perhaps the very management and delivery methods that have stood the test of time, Agile, Scrum, test driven development, TQM? Could provide frameworks for this agentic supervision.

## Accelerated Timelines: Redefining "Agile"

### The New Speed of Development

One of the most profound changes in an AI-driven development environment is the compression of traditional timeframes. When we remove human constraints from the equation, the pace of development accelerates dramatically:

**Traditional Sprint (2 weeks) becomes 48 hours:**
- Hours 0-12: Initial development and unit testing
- Hours 12-24: Integration and system testing
- Hours 24-36: User acceptance testing and refinement
- Hours 36-48: Final validation and deployment preparation

**Release Trains (8-12 weeks) compress to 2 weeks:**
- Multiple 48-hour sprints run in parallel
- Continuous integration and deployment
- Regular strategic review points with the human PM

This acceleration isn't just about speed - it's about maintaining quality while moving faster. AI agents can:

- Perform continuous testing as code is written
- Automatically detect and resolve integration conflicts
- Generate and update documentation in real-time
- Maintain consistent code quality across all components
- 'meetings of agents' will be succinct in comparison to human meetings for clarifications on things such as 'architectural runway', 'backlog preparation'.

## Adapting Agile Practices for AI Teams

### Scrum in an AI-Driven World

While the timeframes change dramatically, the core principles of Scrum remain valuable - they just manifest differently:

**Daily Stand-ups Become Continuous Monitoring:**
- AI agents provide real-time status updates
- Blockers are identified and resolved automatically
- The human PM receives daily summaries and alerts for strategic decisions
- There is probably value in overnight cycles to allow execution time (and cost effective access for batch runs)

**Sprint Planning Evolves:**
- AI Product Owner agents analyse the backlog and propose sprint commitments, per estimated based on previous cycles by the 'developer agents'
- Machine learning models predict velocity and capacity
- The human PM reviews and adjusts based on strategic priorities
- Dependencies are automatically mapped and managed

**Sprint Reviews Transform:**
- AI agents generate comprehensive demo environments
- UAT can be largely automated, even using in browser Agents
- UX / UI design can be reviewed based on event analytics
- Automated metrics show progress against business goals
- The human PM evaluates strategic alignment and user value
- Feedback is immediately incorporated into the next planning cycle

### SAFe in an AI-Driven Environment

The Scaled Agile Framework (SAFe) requires more substantial adaptation, but its hierarchical structure could prove useful as a guide for organising AI development at scale. At the portfolio level, where traditional SAFe focuses on aligning teams with business strategy, the dynamics shift dramatically. The human Product Manager becomes a strategic orchestrator, setting high-level objectives that cascade through the AI ecosystem.

Rather than quarterly portfolio reviews filled with presentation decks and stakeholder discussions, the PM engages with sophisticated AI Portfolio Agents that continuously analyse market data, user feedback, and technical trends. These Portfolio Agents synthesise actionable insights, allowing the PM to make data-informed decisions quickly and efficiently.

At the programme level, an AI Release Train Engineer Agent takes on coordination responsibilities with precision and foresight, delegating to AI Agent Sprint teams. Dependencies are predicted and addressed proactively, ensuring a continuous flow of development.

At the team level, multiple AI development teams work in parallel with automated coordination, leveraging shared knowledge bases (potentially as vector or wiki sites) to integrate and optimise outputs.

## Building Collective Intelligence: How AI Teams Learn and Share

The advantages of AI-driven development isn't just in the speed - it's in the way these systems learn and share knowledge. Imagine walking into a software development office where every decision ever made, every line of code written, and every user interaction is perfectly remembered and instantly accessible. This is the reality for AI development teams.

When Sarah transitioned to AI-driven development, her biggest concern was maintaining consistency. In traditional development, knowledge often resides in scattered documentation, team members' heads, and countless Slack messages. But AI teams draw on vector databases to store and share every decision and insight, ensuring consistent understanding and application across projects.

AI agents don't just recall past decisions - they analyse patterns and apply lessons learned, creating a feedback loop that improves every subsequent iteration. This collective intelligence ensures that AI teams not only operate faster but also make better decisions over time.

## From Concept to Production: A Feature's Journey

To understand how this works in practice, let's follow a feature from inception to deployment. Sarah identifies a common user pain point: customers often abandon their shopping carts during a lengthy checkout process. She proposes a streamlined one-click purchase option for returning customers.

Within minutes of her entering the concept into GitHub as an issue, AI product agents analyse similar features, propose detailed user stories, and propose when and how to add it into the sprint cycles.

AI developers, testers, and UX agents work in parallel, each leveraging shared knowledge to optimise their outputs. Continuous testing ensures that issues are identified and resolved in real-time, with minimal need for human intervention.

The result? A fully tested and validated feature ready for deployment within a short cycle time, with Sarah only intervening for strategic decisions and final approval. Importantly the short cycle times of inflight developments are also preserved, still using the backlog -> sprint -> done cycle.

## Maintaining Quality at the Speed of AI

Quality assurance in an AI-driven environment is woven into every step of development. AI systems analyse code, test functionality, and monitor performance continuously. When issues arise, they propose solutions based on past successes, often resolving problems before human oversight is required.

Sarah's role in quality assurance shifts to strategic decisions: assessing business value, ethical implications, and alignment with organisational goals. The result is a development process that maintains high quality without compromising speed.

Failed cycles can automatically raise and add issues to the trackers.

It's worth remembering that 'AI is not FREE', engineering decision making and cycle management is still needed.

## Governance in an AI-First World

The question of governance initially kept Sarah awake at night. How could she maintain control over a development process that moved so quickly? The answer lay in creating clear boundaries between human and AI responsibilities. Sarah focuses on strategic decisions, while AI teams handle implementation details. Automated documentation ensures transparency and accountability, creating a living history of the product's evolution.

Importantly, Sarah always has sign off on feature cycles, product backlogs going into sprints, and all key decisions.

## The Reality of AI-Driven Development: Challenges and Evolution

Transitioning to AI-driven development required rethinking infrastructure, metrics, and organisational culture. Start small, gradually expanding AI responsibilities while maintaining strong human oversight. Learn to measure success in terms of value delivered and user problems solved, evolving traditional KPIs to reflect the new paradigm.

The journey is that success isn't just about technology - it's about balancing AI efficiency with human judgment, ensuring that both complement each other to achieve better outcomes.

## Charting the Path Forward

Here are what I suspect will be common findings for success:

1. **Start Small:** Begin with contained features and gradually expand AI responsibilities.
2. **Build Infrastructure:** Invest in computing resources, robust CI/CD pipelines, and comprehensive monitoring.
3. **Develop Expertise:** Train PMs in AI oversight and establish clear governance frameworks.
4. **Scale Gradually:** Expand to larger features, increase automation, and adjust timeframes as confidence grows.

## Conclusion: The Future of Software Development

The concept of a single human PM orchestrating teams of AI developers represents more than just an efficiency improvement - it's a fundamental reimagining of how software gets built. By combining accelerated development cycles, continuous automation, and strategic human oversight, organisations can achieve unprecedented speed and quality in software delivery.

While challenges exist, the potential benefits make this approach worth exploring for organisations ready to embrace the future of software development. Those who start preparing now, building the necessary infrastructure and expertise, will be best positioned to thrive in this new paradigm.

---

# Implementing AI Agents? Your Business May Already Know How

Source: https://www.prosperconsulting.ai/insights/implementing-ai-agents

> A Pattern-Based Approach While organisations rush to implement AI agents, they're overlooking something crucial: the patterns for successful AI agent implementation already exist in their businesses. After working through numerous AI agent developments, I've noticed a striking insight - we're…

## A Pattern-Based Approach

While organisations rush to implement AI agents, they're overlooking something crucial: the patterns for successful AI agent implementation already exist in their businesses. After working through numerous AI agent developments, I've noticed a striking insight - we're reinventing decision-making patterns that businesses have refined over decades. This recognition could transform how we approach enterprise AI agent design.

## Why Most AI Agent Implementations Miss the Mark

The current approach to business AI agents often starts with technology rather than wisdom. Companies dive into neural architectures and prompt engineering while overlooking fifty years of proven decision-making patterns. This isn't just inefficient - it's unnecessarily risky.

Consider this: when implementing AI agents, most organisations try to build decision-making frameworks from scratch. Yet their existing business processes already contain sophisticated patterns for handling uncertainty, managing risk, and ensuring quality outcomes.

## The Pattern Recognition That Changes Everything

Let's examine how modern AI agent patterns mirror existing business frameworks. Take the ReAct pattern (Reason+Act) that's revolutionising AI agent development:

- Observe the situation
- Think about implications
- Act on the decision
- Reflect on outcomes

Sound familiar? It should. Military strategists have used the OODA Loop (Observe, Orient, Decide, Act) for decades. Business leaders use Plan-Do-Check-Act. These aren't coincidences - they're proven patterns for making decisions under uncertainty.

## Strategic Implications for AI Implementation

This pattern recognition has profound implications for how organisations should approach AI agent design:

First, development speed increases dramatically, 60% faster, 40% fewer errors, and usually there is a back catalogue of 'unsolved problems' which can be a starting point.

Second, risk management becomes more robust. When we map existing business controls to AI agent design, we get more reliable systems from day one. Risk management is best designed into a process, preventative, than as a gate and recovery external step.

## From Theory to Practice: Making Patterns Work

Consider how this transforms customer service AI agent implementation:

**Traditional Approach:** "Let's train an AI to handle customer queries."

**Pattern-Based Approach:**

- Map existing service decision patterns
- Identify proven control points
- Translate successful escalation frameworks
- Build in established feedback loops

## The Future of Business AI Agents

The organisations that will succeed with AI agent implementation won't be those with the most advanced technology. Success will come to those who best apply proven organisational wisdom to new capabilities.

This isn't about diminishing AI's transformative potential. Rather, it's about building AI agents that benefit from decades of business learning about:

- Decision-making under uncertainty
- Risk management in complex environments
- Quality control in high-stakes situations
- Effective human-system collaboration

We don't really need to invent a new 'method for defining process descriptions for AI Agents'. LLMs and agents are being developed to augment human process, as such by parallel many of the ways that we have managed and designed the end to end business processes are readily adaptable to Agentic AI flows.

## Essential Questions for Leaders

Before implementing any AI agent system, ask:

1. What decision patterns have proven successful in your organisation?
2. How do your best performers handle exceptions and edge cases?
3. Where does human judgement add the most value?
4. What existing controls need to be preserved or adapted?
5. What is your existing employee risk tolerances for delegated decision making?

## Looking Forward: The Next Evolution

As AI agent technology evolves, the competitive advantage will increasingly come from how effectively organisations translate their existing process wisdom into AI capabilities. This means:

- Understanding your current decision-making patterns
- Identifying which controls are crucial for quality
- Knowing where human judgement truly adds value
- Building systems that enhance rather than replace proven practices
- Re-use of agents consistently. I still remember a board meeting which went astray when e-commerce, marketing, finance and supply chain all had different 'The Weekly Sales number'.

## A Call to Action

Review your current AI agent initiatives through this lens. Are you reinventing patterns that already exist in your organisation? Could you accelerate implementation by adapting proven processes rather than building from scratch?

The future of AI agent implementation isn't about discarding organisational wisdom - it's about translating it into new capabilities. The question isn't just "How do we build AI agents?" but "How do we embed our hard-won insights into these new systems?"

What patterns from your organisation could accelerate your AI agent implementation? I'd love to hear your experiences and insights. If you are struggling, maybe I can help?

---

*#AIAgents #BusinessStrategy #AIImplementation #EnterpriseAI #AIAgentDevelopment #BusinessProcess #Innovation #ArtificialIntelligence #DecisionMaking #ProcessDesign*

---

# Resources

Source: https://www.prosperconsulting.ai/resources

- [AI Mastery for the Boardroom](https://www.prosperconsulting.ai/resources/ai-mastery-for-the-boardroom): A practical one-day programme for senior executives who want to save time, sharpen decisions, understand what AI now makes possible, and lead their organisation's AI agenda with confidence. Three sessions in 2026: Nottingham and London.
- [AI and Professional Services](https://www.prosperconsulting.ai/resources/ai-professional-services-briefing): Applying AI for growth, from compliance to competitive advantage. A 37-page strategic briefing for partners and directors at UK legal, accountancy, and advisory firms.

---