What is an AI-native law firm?

The phrase now means two different things, and the difference matters if you are deciding what to do about it.

In the trade press, an AI-native law firm is a new entity: founded in the last two or three years, often with outside capital, built to deliver a narrow class of legal work at a price a traditional firm cannot match. Law360’s May 2026 report on these firms describes the model plainly: routine tasks that used to go to junior associates, contract drafting above all, are done by models, the work is not billed by the hour, and the firms remain a niche while the ethics questions around them are worked out. The Thomson Reuters Institute’s account of one such firm adds the caution that a lower price does not by itself create demand, and that a profession split between relationship advisers and high-volume technicians has to answer where the next generation of judgment comes from.

That is one meaning, and it is a real one. It is not the one this site uses.

The working definition

An AI-native law firm is a firm in which the way legal work is produced has changed, and the change is visible in the workflow rather than in the software budget.

The test is not whether the firm has AI. Every firm has AI; it arrived inside the research platform, the document management system and the office suite whether or not anyone chose it. The test is whether you can point to a step in the matter lifecycle, intake, research, drafting, review or billing, and say: this step is now done differently, here is who is accountable for it, and here is how we know it works.

By that definition a 40-lawyer regional firm can be AI-native and a venture- backed start-up with “AI” in its name can fail to be. The start-up may simply have moved the unexamined first draft from an associate to a model, with nothing downstream changed. That is a cost structure, not an operating model.

Five markers you can check

These are the things to check. Each is a yes or no.

1. A step changed, and it has a name. Not “we use AI for drafting” but “the first draft of the lease schedule is generated from the precedent bank and the term sheet, and the associate’s job on that draft is review, not authorship.” If nobody can name the step, the tool is being used as personal convenience, and personal convenience does not survive the champion moving practice group.

2. Tenancy was decided before the tool was. The firm knows whose systems client information sits in when a model processes it, whether the vendor may train on it, and where it is retained. This is the question the ABA’s Formal Opinion 512 puts under the duty of confidentiality, and it is the one most firms answer last. An AI-native firm answered it first.

3. There is a named review gate. Before an AI-assisted output leaves the firm, a specific person checks it against a specific standard, and for anything with citations that standard includes resolving each authority and confirming it supports the proposition. Opinion 512 frames this as the duty of competence applied to the tool; in practice it is a step in a sequence with an owner.

4. The system is scored against known answers. Somewhere in the firm there is a set of questions with answers the firm has already verified, and the workflow is run against them before it is trusted and again when anything changes. A number a risk committee can act on. Without it, “it works” is an anecdote.

5. There is a policy, and it is a working document. It says what is prohibited, what needs review and what is allowed by default. It is short enough that people read it, and it changes when the firm learns something. A ban is not a policy; it is an admission that nobody has looked.

A firm with all five is AI-native by the definition above, whatever its size, age or capital structure. A firm with none of them is a traditional firm with licences.

What it is not

It is not a firm that replaced associates. The reporting on AI-first firms raises the right worry: if the routine work goes to a model, where does a junior lawyer learn judgment? The answer in a firm that has done this well is that the associate’s routine work moved from producing the draft to interrogating it, which is closer to what a senior lawyer does and, done deliberately, a better apprenticeship than the one it replaced. Done carelessly, it is no apprenticeship at all. The marker is whether the firm designed that transition or let it happen.

It is not a firm with a chatbot on the website, an “AI practice group” or a press release. Those may be true of an AI-native firm; they are not evidence of one.

And it is not a destination. A firm becomes AI-native one workflow at a time, and the first workflow teaches it most of what it needs for the second.

How a traditional firm gets there

The path this site describes has five stages, and they are the same on every page: diagnose, architect, build, prove, govern.

Diagnose first. Find where client information is already going, which is the job of an AI readiness assessment. Then architect one workflow, decide its tenancy and its review gate, build it, prove it against known answers and put it under a policy that people will actually follow. The case studies show what that produces, including where it failed.

The order matters. Firms that start with the tool end up with a pilot that dies quietly. Firms that start with the workflow end up with a step that has a name.

The one-question test

If you want a single question to ask your own firm, it is this: which step in a matter is now done differently because of AI, and who signs off on it?

A firm that can answer in one sentence is AI-native or close to it. A firm that answers with a list of products is not, and the good news is that the gap is an engineering problem, not a cultural one.