Everyone agrees AI will transform law. Not interesting.
Transformation implies survival. Something comes through. The interesting question isn't whether AI changes legal work. It's what remains irreplaceable when it does, and why. Most people stop before they get there.
The answer surprises people.
A Framework
Legal services break into four categories:
- content (what is the law?)
- advisory (how does it apply to me?)
- advocacy (deals, disputes, negotiation)
- process (producing the documents and deliverables).
Plus a hidden fifth: insurance (transferring liability to a credentialed third party).
What matters isn't the categories. It's the order in which they commoditize, and which one doesn't.
The First Four
Content went first. Google finished most of it. You stopped calling your lawyer to find out what California CCPA requires. Competitive pressure to acquire clients led every law firm to publish memos making the law legible online.
Process came next. I am watching it happen in real time. It started long ago with IronClad automating document construction. Then things changed. During a recent acquisition, Lawyers on working calls said "Harvey's cooking on the draft, should have it in two hours." Eight months ago that meant an all-nighter for associates and a redline the next day. The pace changed visibly. It hasn't made legal work cheaper ... yet. I have a theory about why, but I'll save it.
Advisory follows. The combination of good context, good prompts and skills, and a maintained knowledge base already produces output that would have required a senior associate two years ago.
Advocacy, being someone's avatar at the table, the human face of the argument, takes longer. Not because AI can't do it. Because the legitimizing function of a named human expert still carries real weight. I expect that to erode faster than the profession expects as the Overton Window shifts. But it hasn't gone yet. In the end, Advocacy will resolve to Ritual (do yourself a favor and read Sarah Perry's classic What is Ritual).
The Fifth Quadrant
Insurance survives.
Not the narrow financial kind, though that exists too. The more important function: transferring liability and uncertainty to a credentialed third party — and the deterrent value of having done so.
When a regulator comes knocking and we made a call in-house, we are exposed. But if Kirkland wrote a legal opinion, the calculus changes. Kirkland doesn't want malpractice. Kirkland defends its position. The regulator's cost of enforcement just went up. The legal opinion isn't analysis. It's a deterrent.
This function survives not because outside counsel adds irreplaceable analytical value. They often don't. The structural logic of liability transfer doesn't dissolve just because the underlying analysis gets easier. As content and advisory work move into in-house agentic systems, insurance becomes the primary reason to engage outside counsel. More explicit, more narrow, more clearly valued. Maybe more highly valued.
What This Reveals
The most important thing AI replaces isn't work. It's obscurity.
Legal work has always had a judgment layer buried under process. The associate billing hours on a research memo wasn't the valuable thing. The partner's judgment about what that memo meant for the client's decision has always been the valuable thing. Process just looked expensive and important because it was time-consuming and opaque.
When process compresses, the judgment layer becomes visible and primary. Which means the most valuable thing a legal department can do right now isn't adopt better tools. It's figuring out what their judgment actually consists of and building systems that extend it.
I call this context farming. What principles guide how we think about a class of problem? What internal precedents exist, decisions made, reasoning preserved? How do we make that available to the systems we're building?
At AngelList, we've maintained legal tickets with full reasoning and comments for seven-plus years. We knew context is gold but couldn't mine it systematically. That record now becomes one of our most valuable assets. Not for what any individual entry contains, but for what the whole reveals about how we think.
Where This Lands
I don't expect the legal industry to collapse as dramatically as headlines suggest (though, there will be blood for many firm models). But the value driver of the industry changes completely.
Outside counsel will increasingly not exist to do work in-house teams can't do. They'll exist to do work in-house teams shouldn't do — for liability, optics, and deterrent value. Narrower mandates, clearer value.
For in-house teams, the implication follows directly: the value of the department in five years depends almost entirely on the quality of the knowledge infrastructure built today. The lawyers who survive the transition won't necessarily master the best tools. They'll understand what their judgment consists of and extend it systematically.
The most dangerous position for a legal department right now isn't falling behind on AI. It's continuing to treat process as the product when the process is already, quietly, becoming free.