The thin layer of judgement is the whole business

There’s a phrase I keep coming back to, in my own writing and in conversations with firms, and I want to give it a piece of its own because I think it’s the most important idea in legal AI. It’s the thin layer of judgement.

The idea is simple. Any piece of legal work is mostly assembly and application - gathering, checking, drafting, applying a known method - sitting underneath a thin layer of genuine judgement, the part that’s actually contingent, actually hard, and actually what the client is paying for. AI is rapidly getting good at the thick base. What it can’t do, and won’t for a long time, is the thin layer on top. And as the base gets automated, almost all the remaining value concentrates into that thin layer. Which means the thin layer of judgement isn’t a leftover. It’s the whole business.

Why the layer is thinner than lawyers think

When I say the judgement layer is thin, lawyers often bristle, because it sounds like I’m saying their work is mostly mechanical. I am, a bit - and I say it as someone who did the work. Be honest about a typical piece of legal work and most of it is not genuine, contingent judgement. It’s assembly and the application of known methods to recognised situations - skilled, valuable, but repeatable. The genuinely contingent judgement, the part where the answer isn’t in any precedent and being wrong is expensive, is a real but thin slice on top.

This is uncomfortable because the profession’s self-image and its pricing both assume the whole thing is judgement. The blended hourly rate charges judgement prices for the assembly too. The status comes from being a person of judgement. So “most of it is assembly” feels like an attack. It isn’t. It’s the most important strategic fact in the industry, because it tells you exactly where the value is going to concentrate as the base gets automated - and it’s a much smaller, much more defensible place than the profession currently prices for.

The thin layer is where everything valuable lives

Here’s why I say the thin layer is the whole business rather than a problem. As AI automates the thick base, the base commoditises - it gets cheap, fast, and roughly equal across everyone with access to the same tools. What doesn’t commoditise is the thin layer of judgement, because that’s the part the model can’t do. So all the differentiation, all the defensibility, all the premium, concentrates there.

This is genuinely good news, properly understood, and most firms read it as a threat because they’re attached to the volume of the base. But a thin layer of genuine judgement that’s defensible and high-value is a better business than a thick base of assembly that was always going to be automated. The firms that thrive are the ones that recognise the thin layer is where they live now, and organise the entire business around making that layer as good as it can possibly be - rather than mourning the base, which was never the real value, only the bulk.

What this means for building products

For anyone building legal AI, the thin-layer idea is the most useful design principle I have. It tells you precisely what to build and what not to.

Build to automate the thick base. The assembly, the application, the marshalling, the first-pass drafting - that’s where the model creates real value and where the time savings live. Go after it aggressively.

But do not try to automate the thin layer. The judgement is where you stop. Not because the model can’t generate a plausible answer - it can - but because the thin layer is precisely the part a responsible human must own, where being confidently wrong is catastrophic, and where the whole value depends on someone being accountable. A product that respects the thin layer does the base and stops, presenting a prepared, evidenced piece of work for human judgement. A product that tries to automate the thin layer crosses the one line that gets it rejected and, occasionally, gets someone hurt.

The best legal products are therefore thin-layer aware by design. They know exactly where the base ends and the judgement begins, they automate everything up to that line ruthlessly, and they hand the judgement to the human deliberately, with everything the human needs to make it well. Getting that line right, for each task, is most of the design work - and it’s the part you can only get right if you understand the work deeply enough to know where the judgement actually lives, which is not where it looks like it lives from outside.

The strategic punchline

If the thin layer of judgement is the whole business, two things follow for anyone in legal services.

First, get out of the base. The assembly work you’re charging judgement prices for is on a timer. Automate it yourself, price it as a product, and stop pretending it’s where your value lives - because the market is about to be able to see that it isn’t.

Second, invest everything in the thin layer. Make your genuine judgement the best it can be, because that’s the part that survives, differentiates, and commands a premium when everything around it is commoditised. The firms and the products that win the next decade are the ones that understood, early, that the thin layer of judgement was never the leftover after automation. It was the point of the whole exercise - and the automation just stripped away everything that was hiding it.