For more than twenty years, running businesses, I was a heavy user of mid-sized law firms. Commercial contracts, share purchase agreements, M&A, HR and tax matters, the full range. If it carried risk, it went to a lawyer.
In the last three years that spend has almost vanished. AI now does much of the work I used to pay for.
And yet the law firms I work with as clients have not seen their revenue fall. In fact, the UK legal market has just had another record year, with one market estimate putting 2025 revenue at more than £55 billion. If AI were only replacing lawyers, that coexistence would be difficult to explain. So what is actually going on?
THE SET-UP
Why I am not the signal you think I am
It is tempting to read my zero legal bill as the leading edge of a collapse. Resist that. I am a bad example to generalise from.
I have spent a quarter of a century working alongside lawyers. I know roughly where the potholes are. I can look at a contract and judge where the real risk sits, whether it genuinely needs counsel, and where I am carrying the risk myself if I get it wrong. Just as importantly, I am a proficient user of these tools. I know how to get a useful answer out of a model and how to tell a finished one from a half-formed one. That is a skill, and most people do not yet have it. The people who do can quietly remove some work from the market. Most cannot.
So park my anecdote and look at what some firms are beginning to see. It is not necessarily fewer clients. It is clients arriving with an AI-generated legal position already in hand. And that does not always end the conversation. It starts a different, often bigger, one.
All of this growth has happened in the same period AI legal tools went mainstream. That does not prove that AI caused the growth, or that it is not displacing work underneath it. But it means the honest question is not simply whether AI destroys the legal market. It is whether AI might quietly grow it
QUESTION ONE
Do cheaper answers create more work, not less?
There is a 160-year-old idea that explains why this might be. In 1865 the economist William Stanley Jevons noticed that as steam engines got more efficient, coal use did not fall. It soared. Cheaper power meant more industries used it, so total consumption went up, not down. When you make something cheaper, people do not always consume a fixed amount more efficiently. If demand expands sufficiently, they do far more of it.
Jensen Huang, the CEO of Nvidia, makes the same point about intelligence itself. His argument is that as AI keeps getting cheaper and more abundant, we do not use less of it. We find vastly more uses for it, which is exactly why he insists the technology will create more work, not less. Make thinking cheap enough and people think about things they previously left alone.
Legal thinking has just become dramatically cheaper to produce in first-pass form. Apply that logic and the question is not only whether it shrinks demand, but how far it expands it.
My experience says it expands, and the mechanism is what I would call fractal. Every answer AI gives you opens two more questions. Each of those opens two more. Because you can now explore a legal problem quickly and cheaply, you go further into it than you ever would have when every question cost you £400 an hour. And the deeper you go, the more you surface issues you did not know you needed to raise.
There is precedent for this. When ATMs arrived in the 1970s, many assumed bank tellers were finished. Instead, cheaper branch operations meant banks opened far more branches, and total teller employment initially held up and then rose. The machine handled the routine cash, and the human moved up to advice and relationships. It is worth holding onto that story, because it has a sting in the tail I will come back to.
There is also a simple market-size point. Vast amounts of legal need currently go unmet because advice is too expensive to bother with. Around 32% of adults in England and Wales who experience a legal problem receive no professional support. Lower the cost of getting started, and you pull people into the market who were never in it. That is not the existing pie being redistributed. That is a bigger pie.
QUESTION TWO
Does complexity push work back to lawyers?
Growing the volume of legal questions is one thing. The more interesting move is where those questions go.
AI is at its strongest on bounded tasks: summarising and comparing documents, producing a first draft and retrieving information when it is grounded in reliable sources and properly checked. It is at its most limited on the parts that depend on context and judgement: the specifics of your situation, the commercial trade-offs, what you should actually do given everything in play. While models have got much better at this, they can still on occasion invent authorities or state the law incorrectly, so fluency should not be mistaken for reliability.
As the fractal process drives you deeper, the questions change character. They stop being ‘what does this clause mean?’ and become ‘given my position, what is the right move, and what am I exposed to if I am wrong?’ That is judgement under ambiguity, and it is precisely where a lawyer earns their fee.
You can see this in what the market actually pays for. At the upper end of the market, the greatest value sits in commercial advice, contested matters, deals and disputes where the stakes are high and the answer is genuinely uncertain. Few sensible clients would entrust a live M&A negotiation or serious dispute entirely to an unregulated chatbot.
This is why Torsten Slok, chief economist at Apollo, argues that Jevons will mean AI produces more demand for lawyers and accountants, not less, because falling costs expand the market for their work
THE DIVIDING LINE
The routine, letter-of-the-law drafting gets absorbed or repriced. The judgement-heavy, high-stakes, high-ambiguity work rises in relative value. The lawyer's role shifts up the value chain rather than disappearing.
But a larger market for legal solutions does not automatically mean a larger market for traditional law firms. The winners will be the firms that use AI to create a cheaper front door and then convert the complexity it reveals into trusted, accountable advice.
THE CATCH
The moat is liability, not intelligence
Here is where I have to be honest about the other side of the argument, because a sceptic will raise it and they will be partly right.
All of this could be a lagging indicator. Remember the ATMs. For a time the Jevons effect held and teller numbers rose. Then online and mobile banking allowed customers to bypass the branch altogether, and the effect broke. Tellers finally declined. The question for law is whether we are in the happy middle period, or whether the machine is one capability jump away from breaking the pattern.
But I think the thing keeping people walking into law firms is not what most people assume. It is not that AI cannot reason. The models are already highly capable, and getting better fast. It is that AI cannot itself be accountable.
A lawyer sells you more than an answer. They sell you a throat to choke: professional liability, insurance, a regulated name on the advice, someone who is on the hook when it goes wrong. An AI has no professional duties and cannot carry personal responsibility for its advice.
In March 2026, Nippon Life sued OpenAI in the US, alleging that ChatGPT had effectively practised law without a licence. The claim is contested and has not yet established a legal precedent, but legislators are already considering how liability should apply. A proposed New York law, for example, would expose AI platforms to claims where a chatbot effectively impersonates a licensed professional.
For now, someone regulated still has to own the risk. That may be the profession’s strongest defence. But England and Wales already has SRA-authorised AI-led law firms. The real question is how long it will be before one can put its regulated name and professional indemnity insurance behind complex legal advice at scale.
REFERENCES
Sources and evidence
● UK legal-services revenue exceeding £55 billion in 2025, exports passing £10 billion and business and commercial work representing 51% of market value (Research and Markets via GlobeNewswire).
● Jensen Huang on AI agents augmenting rather than replacing workers (Fortune, 19 March 2026).
● The Jevons employment effect and its application to lawyers and accountants, Torsten Slok (Apollo).
● The ATM and bank-teller story, James Bessen (IMF Finance & Development).
● AI in legal services, consumer expectations and unmet legal need (Legal Services Board).
● Reliability and error rates of AI legal-research tools (Stanford HAI).
● Nippon Life v. OpenAI and proposed professional-impersonation legislation (Reuters).
● Authorisation of the first AI-driven law firm in England and Wales (Solicitors Regulation Authority).