Facts are friendsAI and the legal profession
A structural analysis, not a sales pitch.
1.Why this page exists
1.1.I have spent four decades advising businesses through corporate finance, venture capital and complex transactions, most recently as a consultant solicitor and as founder of L5 Executive Services. I have no product to sell you here and no platform to promote. What follows is an analysis of what is actually happening to the legal profession as artificial intelligence matures, and what I think firms, in-house teams and individual lawyers should do about it.
1.2.My approach is evidence-based and deliberately contrarian. I am not persuaded by the comforting version - that AI simply expands demand for legal talent and everyone benefits proportionately. Nor am I persuaded by the panic. The truth, as usual, sits in less comfortable territory.
2.The three foundations, and which ones are moving
2.1.A law firm has historically rested on three things: a near-monopoly on knowledge (the statutes, the cases, the procedural detail an ordinary client cannot navigate alone), credentialling (the regulated professional as a guarantee, and someone to sue if it goes wrong), and judgment (advocacy, relational intelligence, the wisdom about people and situations that resists systemisation).
2.2.Artificial intelligence mounts a serious challenge to the first. A partial challenge to the second. For now, it leaves the third largely standing. Firms that migrate toward judgment, and away from the knowledge monopoly that machines are dissolving, are the ones I expect to survive this in recognisable form.
3.What is already under way
3.1.Kirkland & Ellis has signed platform deals with Palantir and Syllo, encoding institutional knowledge, practice group by practice group, into infrastructure it does not fully control.
3.2.Legora has partnered directly with Ironclad, moving toward the client rather than the firm. Harvey is training open-source models on firms' own workflows - ostensibly so firms own their intelligence, though a workflow once encoded is also a workflow that can be reproduced elsewhere.
3.3.Garfield AI, an SRA-authorised AI-driven law firm, has already won a regulated civil trial, with the legal work done by AI and the advocacy by a human barrister. This is a genuine access-to-justice moment, and also a preview.
4.The uncomfortable part
4.1.The comfortable assumption is that AI displacement starts at the bottom - the paralegal, the junior analyst - and works up slowly, if at all. I think this is only partly true, and partly a fiction maintained by those on the higher rungs.
4.2.The economics of replacing a senior partner, or a strategy consultant, are considerably more attractive than the economics of replacing a trainee. The AI does not need to be perfect. It needs to be better value than the expensive human. The people who have built their identity most thoroughly on professional standing may be the people whose standing is most exposed.
5.What I think this means for firms and individuals
5.1.For firms: the defensible ground is judgment, relationship and accountability, not document production or research recall. Firms funding platforms that encode their own workflows should ask carefully who else, eventually, gets to use that encoding.
5.2.For individuals: foundational understanding of how a field actually works, not just fluency with current tools, is becoming valuable again in the way it was for the programmers brought back to fix Y2K-era systems. Depth outlasts the tools built on top of it.
6.Where to take this further
6.1.This analysis continues weekly in the #TrojanHorseDebate series, an ongoing exchange with Lev Loukhton, and in longer form on Substack under #FactsAreFriends.
6.2.If you are navigating LLM strategy for a firm or in-house team, or thinking through the ethics and economics of algorithmic disruption in professional services, I am glad to have that conversation directly.