Strip away the demos and the dread, and artificial intelligence has done exactly one thing to legal work so far: it made reading at scale cheap. A model that can be pointed at a matter bundle and asked questions — with the passages behind each answer — changes the economics of the first pass over documents. It does not change who is answerable for the advice.
What actually changed
The shift is the combination of two older ideas. Retrieval has existed for decades — search inside documents is not new. Reasoning-quality language models are the newer half. Joined together, retrieval-augmented generation lets a system find the relevant passages and write an answer that names them. For a profession whose raw material is text and whose gold standard is the citation, that combination is not a gimmick; it is the shape of the job.
Three tasks absorbed it first, and they are the same three our platform pages describe: first-pass review of bundles, drafting grounded in a firm's own precedents, and interrogation — asking a matter questions and getting the span back. Not litigation strategy. Not judgement calls on privilege. The reading layer.
What did not change
Responsibility did not move. A partner who signs an opinion answers for it as fully as before, whatever tool produced the first draft. Privilege did not move either — the duty not to disclose confidential material now extends to what a tool transmits, which is why architecture matters more than marketing: a system that keeps processing on the device cannot leak what it never sends.
The machine changed the cost of reading. It did not change the owner of the answer.
Where it lands in a firm
- Due diligence and review: the first pass over thousands of pages becomes hours, with every finding carrying its source for counsel to verify.
- Drafting: first drafts assemble from the firm's own precedents instead of a blank page — or worse, a public model trained on who knows what.
- Client questions: “what does clause 9 actually say” becomes answerable in seconds, with the clause attached.
- Knowledge: the firm's accumulated matters become queryable without any of them leaving the building.
The honest caveats are on this site in full: no published benchmark yet, no independent audit, and a demonstrable boundary rather than a promised one. What we can say is directional — the firms adopting reading-layer AI are not replacing lawyers. They are moving the lawyers from the reading to the deciding, which is where the fees and the judgement were always supposed to live.