We opened our Logically Drafted session in Chicago last week with a short talk on the future of drafting, then handed over to a panel of innovation and AI leads – Will Gaus, Laurie Miller, Allison Harbin and Stephen Christiansen and expertly chaired by Jack Recinto of Ice Miller. The most honest moment came early. Around nine in ten people in the room said they feel behind everyone else on AI.

Nobody is behind. That feeling is manufactured: by press release and procurement, by a market that announces innovation far faster than it delivers it. Everyone is still working out what their tech stack should look like. The firms seeing real success aren't the ones who bought fastest. They're the ones who spent the time planning rather than just purchasing.

Planning sounds soft until you see what it involves. Deciding the firm's position before automating anything. Teaching a system how you actually write, and which clauses you always negotiate out – months of work, not a switch you flip. An honest audit of your own precedents before you let a model anywhere near them. The unglamorous work that no demo shows you is exactly the work that decides whether any of it pays off.

Which is why the adoption numbers deserve more suspicion than they get. A high percentage of user logins tells you people have opened the tool. It doesn't tell you whether the use is good use, or what it actually costs. If AI saves an hour of drafting and adds an hour of review, the time hasn't gone anywhere – and most ROI conversations still don't put review time on the other side of the ledger.

And not every job is a nail. AI is probabilistic by design. There are tasks its architecture simply isn't built for, and on the ones it can do, a fluent answer can read exactly like a correct one. A lot of what gets called adoption is really internal education: teaching people which tool fits which job, and when to reach for structure rather than a blank prompt. The jobs it suits are the repetitive, well-bounded ones; the jobs it doesn't are the novel, high-stakes calls where being plausibly wrong is most expensive. Knowing the difference is most of the skill.

None of this is pessimism. The mood in the room was appetite for what comes next. The firms further along aren't shrinking their associate intake; they're taking on clients and matters they could never previously justify – the smaller client, the lower-value piece of work that used to lose money – and handing juniors more significant work than they'd otherwise see in their first years. Used on the right jobs, AI widens what a firm can profitably do. That is a business model question long before it is a technology one.

There is one part no tool answers, and it ran underneath the whole conversation. Law is an apprenticeship. If AI does the routine work that juniors used to cut their teeth on, where does the next generation build judgement? We still teach children mathematics in a world full of calculators, for the same reason. The most valuable thing a firm sells has always been the lawyer's brain, and a firm that automates the training ground without rebuilding it somewhere else is quietly eating its own future.

The hype is loud, and it gets louder every quarter. But nobody is as far behind as the noise suggests, and the advantage won't go to whoever buys first. It goes to the firms that plan, that pick the right tool for the right job, and that keep judgement at the centre while everything around it speeds up.

That is the work behind the hype, and it is the work that lasts.