Contract review has quietly automated itself for a decade. Clause libraries replaced blank pages. Version control replaced "final_v7_ACTUALFINAL.docx." Search replaced flipping through a binder. And yet the actual act of redlining (proposing a change, explaining it, and waiting for the other side to accept or push back) still happens almost exactly the way it did twenty years ago: one lawyer, one document, one clause at a time.
That's strange, because redlining is the most repetitive part of the job. A firm's limitation-of-liability position doesn't change deal to deal. Neither does its stance on indemnification carve-outs, or its house view on governing law. What changes is the counterparty's draft, and the work of reconciling that draft against a position the firm already holds is exactly the kind of pattern-matching that's slow for a person and fast for a system built to do it.
The bottleneck was never "finding" the issue
Ask any associate what takes the longest in a review, and it's rarely spotting the problem clause: most experienced lawyers can flag a bad indemnification provision on sight. The time goes into the mechanics around it: cross-checking it against the firm's playbook, drafting alternative language, making sure the change is consistent with three other clauses in the same document, and writing a note that explains the position clearly enough for opposing counsel to act on it without a call.
The redline itself was never the hard part. Making it defensible, consistent, and fast was.
That's the part worth automating: not the judgment call about whether a clause is acceptable, but the mechanical work of proposing the fix, checking it against precedent, and keeping the rest of the document consistent once it's made.
What changes when a system does the mechanical part
When first-pass redlining is automated well, the associate's job shifts from drafting the change to reviewing it: reading a proposed edit, deciding whether it fits the deal, and adjusting it if not. That's a faster loop, but it's also a fundamentally different kind of work: less transcription, more judgment. It's closer to what a senior lawyer already does when they review a junior associate's markup.
The caveat is that none of this works if the system's suggestions arrive as a fait accompli. A redline a lawyer can't see the reasoning behind, or can't easily reject, isn't actually saving anyone time. It's just moving the verification work somewhere less visible. The suggestions have to show up as tracked, attributed changes, the same way a junior associate's markup would, so the reviewing lawyer is doing the same job they always did: deciding what to accept.
Where this is headed
The interesting version of this isn't "AI that redlines for you." It's a system that knows a firm's playbook well enough to flag departures from it automatically, propose language consistent with past deals, and keep a document internally consistent as changes ripple through it. That's a narrower, more useful claim than "AI understands contracts," and it's the one worth building toward.
That's the bet behind JURISPAI's document automation: not replacing the lawyer's judgment about what a clause should say, but closing the gap between having that judgment and getting it onto the page.