AI in the legal industry has reached the stage where the gap between the headlines and the daily reality is wide. The trade press describes a profession being remade. Walk into most firms and you find a few specific tools doing a few specific jobs, a lot of cautious experimentation, and several things that got switched off after a month.
This note is the grounded read on where AI in the legal industry is actually real, drawn from the adoption work behind our law reports.
What is genuinely in use
Document review on a firm's own files is the most established use. Pulling the relevant clauses out of a long contract so an associate reads the three pages that matter instead of forty. It is unglamorous and it works, which is why it stuck.
Intake and triage is the second. Drafting a first response to a prospective client, running a preliminary conflict check against the firm's matter history, flagging the matter type. A human still decides whether to take the case; the software gets the firm to that decision faster.
Drafting routine correspondence is the third. Engagement letters, status updates, standard clauses. The lawyer edits every output, but the blank page is gone. These three uses share a feature: each removes grunt work without asking the tool to exercise legal judgement.
What got switched off
Legal research without a human checking every citation is the use that bites. The pattern repeats across firms: someone trusted an AI research summary, a citation turned out to be fabricated, and the firm now reads every authority before it goes in a filing. That rule is correct, and it erases most of the time the tool promised to save.
Client-facing bots answering legal questions are the other retreat. The liability is obvious once said out loud, and most firms that tried it pulled it back to "we will have someone call you."
The lesson the industry is converging on is that AI is safe where a qualified person reviews the output and dangerous where it is trusted unsupervised. The firms moving fastest are not the least careful, they are the ones who put the checks in the right places.
Where this is heading
Adoption is widening at the task level rather than replacing roles. The realistic near-term picture is more firms using the established tools for intake, document review, and drafting, with research assistance trusted only under human verification, and regulators expecting disclosure where AI touches client data.
For a firm deciding where to start, the safe entry points are the three established uses, integrated with the practice-management system already in place. The report behind this note covers adoption across firm sizes, what is working, and the benchmarks for what realistic use looks like in 2026.