There’s a confident thesis going round legal tech, and the sharp versions of it are good, so I want to give it its due before I take issue with it. It runs like this: AI commoditises legal drafting, and at the same time kills the demand for checking the draft, so routine commercial work just skips the lawyer. A business reads a confident AI draft, decides the risk is fine, and signs. No lawyer, no check, done. On the first half of that, that drafting is commoditising fast, I think they’re right.
My problem is with one assumption sitting underneath it. The thesis quietly assumes everyone inside a business weighs legal risk the way a founder or Exec does. They don’t, most lawyers inside organisations get this. Once you see that, a lot of the argument comes apart.
The thesis pictures a clean, firm-level calculation. The company weighs the cost of a lawyer against the odds of a dispute, decides the risk is acceptable, moves on. That’s exactly how a founder thinks, because a founder owns the company and the downside. But a company never actually signs anything. A person does. That person isn’t actually pricing the company’s risk at all. They’re pricing their own name on it. The review afterwards, the “why didn’t you run this past legal,” the mark against their judgement, their career if it goes wrong.
So their rational move is to hand the decision to someone whose job it is to own it. That’s just how risk works when it lands on an individual actually doing the day-to-day, rather than a balance sheet.
And better AI doesn’t blur this out. It’s the opposite, it focuses it. Once the model spits out the draft in seconds, the slow step is the sign-off, and the sign-off is precisely the part nobody (except maybe burn conscious founders) wants to carry alone. The wish for a named, accountable person isn’t an old habit that fades as the models get better. It’s structural. Anyone who’s worked in-house feels this, you ironically feel super valuable to the individuals within the business but its difficult for that value to be translated at an organisational level until something goes wrong. It’s because the organisation speaks a different risk language to the individual. It always has.
The poster child for the new world is usually the founder writing her own privacy policy. That happens but she’s often a sole trader who answers to no one. She isn’t the GC, or the procurement lead, or the Head of People, and that’s where the actual volume of contracts sits. Founders will take the risk because they own the downside. But they’re a thin slice of all commercial-legal work, and their posture on this invariably changes as they scale. Good founders take legal seriously, and good companies tend to follow.
There are a few analogies I initially drew on: planes/pilots, etc but the overall point I think hits the nail on the head is that automation in a high-liability job doesn’t remove the human. It raises the bar on who that human has to be, and cuts how many you need. Ok fine, a few lines on aviation. Autopilot flies almost the whole flight. It didn’t lower the qualification of the person in the seat, it raised it, because now you’re paying them to handle the exceptions and to hold the certificate the regulator and the insurer insist on. The cabin crew never started flying the plane to save money. Not because you couldn’t train them, but because customers and regulators wouldn’t have it. The constraint was never whether they were capable. It was who was allowed to be accountable.
The cleanest parallel is prescribing. Drug information has been free for decades and is trivial to get now (especially with LLMs). Diagnosis is increasingly automatable. Yet, prescribing is still locked behind a licensed human, for one reason: someone regulated has to own the decision. You can read all night. You still can’t prescribe yourself the thing that matters. The information got cheap and the gatekeeper stayed, because the gate was never about scarce information. It was about who’s accountable when the call is wrong. Legal verification sits in exactly that spot. Granted, the ultimate pain might seldom be life or death but to companies it can still have significant consequences.
And here’s the detail that points to the answer. Nobody keeps a full-time pharmacist on staff. You use one as and when you need it. The expertise is essential, specialist, licensed, and consumed in small doses. That’s the natural shape for a gated, high-trust function most people only need now and then. It’s also, I think, where legal is going.
If the duty can’t be held by a machine, or by a platform that disclaims it in its terms, it has to be held by a regulated human. The traditional version of that human, the full-time lawyer or the firm on retainer, is priced for heavy continuous use and passed on to clients who can afford it. It does nothing for the business that needs occasional, targeted, accountable cover. That’s the gap the disruption thesis actually exposes, and it’s the one a fractional model fills. The entry cost drops because you engage the lawyer as and when needed, so carrying a named duty costs something proportionate to the matter rather than to a salary. A deterministic layer does the mechanical checking, so the marginal cost of the human standing behind the output falls towards the cost of the exceptions. And nobody needs to be full-time, because fractional supply matches fractional demand. That’s where the market is headed and that’s what we are building for at Correm.
So the conclusion isn’t that lawyers survive because the work is too hard for the machine. The machine can do plenty of it. Lawyers survive because the accountability can’t sit with a machine or a platform, someone regulated has to hold it, and the moment someone can hold it cheaply, the market scales instead of stalling. The numbers shrink. Fewer lawyers needed. But the ones who remain are required, and they’re leveraged across far more automated throughput. Fewer pilots, more capable, each across more aircraft (sorry).
The liability has to land somewhere. Whoever catches it cheaply takes the value the data giants are too exposed to touch and the model platforms won’t go near. The lawyer doesn’t vanish. The lawyer becomes the pharmacist. Fewer of them, specialist, on demand, standing behind the machine.
But let’s not be naive, its all still very early. Whilst I’m offering a forecast I’m not reading from a thermometer. The way legal risk is actually carried, by individuals, against their own names, makes the clean disintermediation story unlikely, and points to a fractional, verification-backed, duty-carrying layer as where this settles. Let’s see what legal tech announcement next week makes me rethink!