How legal workflows should change in the generative AI era
Hello, this is Legal Agent.
Many legal teams have started using generative AI, yet the overall workload has not dropped as much as expected. Summarizing a contract, asking what a clause means, and drafting a revision are all things general-purpose AI can now do reasonably well, but the time spent confirming the request, finding past materials, and entering tracked changes into Word remains.
What gets lost the moment text is pasted into a browser
Pasting a clause into a browser to check its meaning or get a first-draft comment works fine for that narrow question. Real contract review, though, happens on a Word file carrying the counterparty's tracked changes and comments, plus business-unit context such as "we can't move on this point" or "we'll work with this counterparty again." That context disappears the moment only a fragment of the contract is pasted elsewhere, so the AI's answer can look reasonable while still falling short of what the file actually needs.
Word and the matter folder are where the work actually happens
Legal work is largely conducted through rounds of edits on a Word file: revising language, leaving tracked changes, explaining the reasoning in a comment, and responding to the counterparty's own comments, all while keeping the tone appropriate for both the counterparty and internal readers. The matter folder also holds past drafts and the sticking points from similar deals, tacit knowledge that in-house lawyers draw on constantly. Bringing generative AI into legal work means bringing it into that folder and that Word file, not just into a chat window next to them.
Let AI prepare the draft, let the lawyer finish it
A liability cap is not determined by contract type alone; it depends on deal size, bargaining power and past disputes. The realistic model is AI doing the groundwork, surfacing issues, comparing against past templates and drafting candidate language, while a lawyer with corporate legal experience finishes it using the company's own priorities. That combination, not AI output used as-is, is what actually holds up in practice.
Separate what goes to the counterparty from what stays internal
An AI-assisted workflow is not just about generating text faster; it needs to decide where in the Word file a comment lives, who it is for, and how strongly it should be worded. A comment meant for the counterparty, asking for a cap to clarify liability, should read very differently from an internal note flagging that the current terms could expose the company to broad damages relative to the deal size. Keeping those separate is what lets the business unit move immediately on what comes back.