The Chalk Mark Judge Scott Schlegel Sept 1 2026

Within four days of each other, Thomson Reuters and LexisNexis announced major new agentic capabilities for their legal AI platforms. Thomson Reuters said its next-generation CoCounsel Legal can “reason, plan, and execute at the level of a senior associate,” moving from research and issue analysis to trusted work product. LexisNexis described a similar ambition for Lexis+ with Protégé, moving lawyers from “first idea to review-ready work product.”

I think these developments are remarkable, and I am not suggesting we should necessarily resist them. If these tools can help lawyers work faster, lower costs, and produce better work, maybe we should use them. But that idea of “review-ready” work made me think about an old story about an engineer who was called in to fix a massive industrial machine that no one else could repair. He studied it, made a few adjustments, and put a chalk mark where the problem needed to be fixed. The repair worked, and the owner later received an enormous bill.

When the owner complained about the price for what looked like a few minutes of work, the engineer explained that only a tiny part of the bill was for making the chalk mark. Almost all of it was for knowing where to put it. The mark took seconds. Knowing where to put it took his whole career.

That is the part we need to remember as legal AI gets better. For generations, lawyers developed expertise by doing the work. They researched issues that led nowhere, read cases that turned out not to matter, wrote bad first drafts, watched arguments fail, and slowly learned how legal problems fit together. Much of that work was inefficient, but some of it was part of the apprenticeship.

Legal technology has always removed drudgery. Lawyers once spent days searching warehouses full of boxes for a document that can now be found in seconds. The profession was better for that kind of progress, but the lawyer still had to wrestle with the legal problem and decide how the pieces fit.

These new systems are beginning to do more of the intellectual work, though. They can organize the facts, identify issues, find cases, develop arguments, draft the brief, and check citations and supporting authority. The lawyer increasingly receives what appears to be an extraordinarily good piece of work and merely becomes the reviewer.

But review is not the same thing as supervision. Meaningful supervision requires enough independent understanding to know whether the right issue was identified, whether the procedural posture changes the analysis, whether the right line of cases controls, and whether the answer actually addresses the problem that needed to be solved.

Now imagine the lawyer who begins practicing in 2028 and spends the next twenty years reviewing work produced this way. That lawyer may become extraordinarily good at using AI. But if the machine consistently performs the work through which generations of lawyers developed expertise, how does that lawyer learn to recognize the moment when the machine gets something important wrong?

The answer is not to turn these tools off. It is to be deliberate about when and where young lawyers still have to sit with hard problems, engage with the underlying material, form their own view, and occasionally follow the wrong path long enough to understand why it was wrong.

My concern then is that, twenty years from now, everyone in the room may know how to operate the machine, prompt it, review what it produces, and keep the work moving, but no one will have developed the depth of understanding to recognize when something is subtly wrong.

The machine may produce an extraordinarily good answer. The cases may be real, the citations may be right, and the writing may be excellent. But someone still has to know where the chalk mark belongs.

Thanks for reading [sch]Legal Tech! Subscribe for free to receive new posts.

Subscribed