The Model Wasn't Better. It Was Approved.
Legal AI is sold as a revolution in intelligence. Too often, the product is permission and the concept is branding.
This week, the future of legal work arrived on Teams. It was presented by a “Legal Solutions Architect”, which in ordinary English meant a legally qualified trainer with a better title and a sales deck.
My firm is trying to make up ground on AI adoption. We have done the predictable things. We have formed “tech groups” populated by people who regard a fully customised Android phone as a technical qualification and created a platform for quietly smug lawyers (junior in rank, yet senior in self regard) to catalogue the shortcomings of technology they would never permit anyone to identify in themselves.
Then came the demonstration.
This company in particular has branded its way into the legal lexicon. Its name is now ubiquitous across UK law firms, spoken less like a vendor than an inevitability. The marketing has been extraordinary. The product was solidly underwhelming.
We were told that the version shown to us sits on top of a customised GPT model and, for external legal information, relies on publicly available government sources. It could extract key points from heads of terms, compare them against a draft and assemble a tidy report. Useful, certainly. Transformative, no. It was document review without commercial understanding: competent at organising what was there, unable to understand why it mattered. Essentially working at NQ level.
A partner asked the obvious question. How was this meaningfully different from Copilot, which appeared to do much of the same work without some of the constraints? No convincing answer arrived.
The more awkward comparison was with the general-purpose AI tool I already use, admittedly on an expensive subscription. It is vastly better. It reasons better, writes better, handles context better and is useful across a far wider range of work. Yet nobody had gathered to admire it. It lacked the right enterprise costume.
And that, I suspect, is the real product.
The vendor we saw may not have built the best intelligence (it certainly hasn’t) – but it has built an intelligence a law firm can approve. It offers data controls, security assurances, procurement comfort and the sacred word “legal” stamped across the box. The profession is not necessarily buying superior intelligence. It is buying institutional permission to use something half as good as it could if it hadn’t created so many artificial regulatory landmines for itself.
Law firms have manufactured so many layers of risk, governance and internal anxiety that they now deny themselves the best tools and pay heavily for safer-looking substitutes. The market has noticed. There is no need to build the best product if you can persuade the profession that yours is the only one safe enough to buy.
Put a general capability behind enterprise controls. Give it a legal logo. Send in an architect. The collective jaw drops on cue. It’s all mind-blowingly dumb.
Our own little task force will keep meeting, of course, led by the usual coalition of junior enthusiasm and senior self-regard: people confident enough to direct a technological transformation they do not fully understand, in a profession nowhere near as technically literate as it imagines itself to be.
Still, there is another Teams briefing in the diary.
I cannot wait.