Every firm has now heard that the future is AI-native. Most have drawn the wrong conclusion from it.
The common reading is that a new species of law firm is arriving to replace the old one, and that keeping up means buying the best legal AI product on the market. Both halves of that are wrong. AI-native describes an operating model, not a product category, and the firms most likely to end up genuinely AI-native are existing firms that redesign how they work. The venture-backed newcomers are proving the model. They are unlikely to be the ones who capture most of the market.
That distinction matters because it changes what you should spend money on.
What AI-native actually means
An AI-native law firm is one where AI execution is the assumption the firm is built on, rather than a capability bolted onto a firm that already exists.
The AI Firm Index, the public directory tracking this category, puts it precisely. Its founder, Lupl co-founder Matt Pollins, defines AI-native firms as those “built from the ground up around AI-enabled workflows, with pricing, intake, delivery, and team structure all built with AI execution as the foundation.” The directory launched in March 2026 with 23 firms, reached 40 by April, and passed 50 by late June, 31 of them in the United States.
The markers Pollins uses are operational, and they are worth reading as a checklist rather than a description:
- Intake happens through agents, automated systems or a client channel, rather than through a partner’s inbox.
- Pricing is published online and quoted immediately.
- Work reaches a model for first review before a human lawyer engages with it.
- Team structure assumes that split, rather than absorbing it.
Notice what is absent from that list. There is no mention of which model the firm uses, which vendor it buys from, or how large its technology budget is. Pollins is explicit that the definition covers established firms redesigning around AI, not only new entrants. That is the part most coverage skips.
The new firms prove the model works, in a narrow band
General Legal is the cleanest example of the model executed properly. It was founded in 2025 by Ryan Walker, the former CTO of Casetext, with Javed Qadrud-Din and J.P. Mohler, both Harvard Law graduates who practised at Fenwick & West, Cooley and WilmerHale. This is the team that built CoCounsel, and they chose to start a law firm rather than another product.
The operating details are the interesting part. A client signs up on the website, completes a form, signs an engagement letter that creates a real attorney-client relationship, and then works with attorneys through Slack. A contract review that used to cost $2,000 is priced at $500 and delivered in under an hour. Six months out of Y Combinator the firm had around 400 clients and had launched a venture financing practice. It also runs an MCP server, so a client’s own AI agents can engage the firm directly. Ryan Walker has talked through the reasoning at length on LawNext.
Norm Law shows the other route in: capital and credibility. It launched as the law firm arm of Norm Ai, which has raised more than $140 million from Blackstone, Bain Capital, Vanguard, Citi and Marc Benioff. Its chairman is Mike Schmidtberger, who left Sidley Austin at the end of 2025 after seven years chairing its executive committee, and joined a two-month-old firm the following month. When a lawyer of that standing moves that fast, the signal is about where the model is heading rather than about one firm.
The wider set follows the same shape. Garfield became the first purely AI-based law firm authorised by the Solicitors Regulation Authority in England and Wales, working in small-claims debt recovery. Crosby pairs licensed attorneys with proprietary AI for commercial contract review at fixed per-document pricing and has raised $85.8 million. Eudia Counsel operates under Arizona’s alternative business structure programme with non-lawyer ownership. The International Bar Association has published a useful survey of the regulatory changes that made these structures possible.
Read those examples together and the pattern is obvious. Every one of them has picked work that is high volume, well specified and easy to scope: contract review, standard commercial agreements, small debt claims. That choice is deliberate, because published pricing requires predictable scope. The category is real, and its current boundary is narrow. We have written separately about what happens when legal AI companies decide to become law firms.
Big Law is answering with capital, and that is the part people misread
In 2026 Kirkland & Ellis committed $500 million over three to four years to build its own generative AI platform, with more than $100 million of that in the first year. The firm’s revenue is $10.6 billion, so the commitment is roughly 1% of turnover, funded from current revenues and taken out of partner distributions. Around 250 lawyers, including 100 partners, are shaping the platform, with 180 technology professionals mapping firm workflows into it. There is a Palantir partnership attached and on-premise GPU infrastructure behind it.
Firm chair Jon Ballis described the goal as taking “the collective intelligence of our institution” and deploying it across the firm.
Other firms are taking narrower versions of the same bet. Freshfields signed a deal with Anthropic in April 2026 to co-develop legal AI tools, with an option to license them to other firms later. Cleary Gottlieb acquired the generative AI company Springbok in March 2025 to build custom tooling in-house.
Here is the misreading. Most people look at $500 million and conclude that becoming AI-native is expensive, therefore out of reach, therefore a problem for later. That gets the lesson exactly backwards.
Kirkland is not spending half a billion dollars on model access. Model access is a commodity and the price of it falls every quarter. It is spending that money to encode its own institutional knowledge and its own workflows into a system it controls. The expensive part is the mapping: 250 lawyers and 180 technologists working out how the firm actually does what it does, and turning that into something a machine can execute against. The asset being built is the firm’s own work product, not the model.
That is a strategic insight available to a firm of any size. The price tag attached to it at Kirkland reflects Kirkland’s scale, its practice count and its decision to build the infrastructure layer itself. It does not set the price of the idea.
The judgment problem, which nobody has solved
Thomson Reuters Institute has published one of the more honest treatments of this shift in its work on the AI-first law firm, and it raises the question most vendor material avoids.
AI-native delivery tends to produce what the piece calls a technician archetype: lawyers working at high volume, at speed, often without a continuing client relationship. That is a coherent way to run a firm. It sits awkwardly beside the trusted-adviser model, where the value comes from knowing a client well enough to anticipate what they will need.
The sharper problem is developmental. Judgment in lawyers has traditionally been built by producing bad first drafts and having a senior lawyer take them apart. If the first draft now comes from a model and the junior’s job is to review it, the training loop that produced senior judgment has been quietly removed. The article’s suggestion is that firms have to rebuild that loop deliberately, including asking, at each step, what cognitive function is being handed to the model.
Any firm claiming to be AI-native without an answer to that question has automated its production line and left its succession plan to chance.
Why mid-size firms are better placed than they believe
The assumption that mid-size firms are losing this race deserves checking against the numbers.
In Clio’s Legal Trends research for mid-sized firms, 86% reported that they had adopted AI. More striking, mid-size firms captured close to 5% demand growth in the second half of 2025, while the Am Law 100 struggled to reach 2%. That is the widest gap between those segments since the global financial crisis. Meanwhile law firm technology spending grew 9.7% and knowledge management spending 10.5%, the fastest real growth the industry has recorded.
Mid-size firms hold three structural advantages here. They have fewer legacy systems to unpick. Their decision chains are short enough that a managing partner and a small group can commit to a change of operating model in a single meeting. And they answer to one profit-and-loss account rather than to a federation of practice groups each defending its own way of working.
The honest constraint is different from the one usually stated. It is not that mid-size firms lack ambition or adoption. It is that they lack eight-figure technology budgets, dedicated innovation teams and the governance infrastructure to deploy AI safely across a firm. Those are real gaps, and they are gaps in capability rather than in intent.
What becoming AI-native actually requires
Strip out the capital expenditure and five things have to change. None of them is a software purchase.
- Intake. A client should be able to start a matter and understand what it will cost without waiting for a partner to be free. That means scoping rules written down rather than held in someone’s head.
- Pricing. Published or rapidly quoted pricing forces scope discipline. It is the single hardest change, and it is the one that makes everything downstream possible.
- Delivery. First pass by model, review by lawyer, with the split written into the workflow rather than left to each fee earner’s preference.
- Knowledge. Your precedents, your past advice and your matter history have to be machine-readable and searchable by the system, because that is the only asset a competitor cannot buy.
- Supervision. A recorded, auditable answer to who checked the model’s output and against what standard.
This is the gap Vettam was built to close. We build mid-size firms a system of their own, shaped around the workflows they already run, sitting on their existing stack and their own matter data, and owned outright rather than rented. The point is to give a firm the operating model that Kirkland is buying, without the half-billion-dollar bill or the vendor lock-in that comes with renting someone else’s roadmap.
Six questions that tell you where your firm actually stands
Answer these honestly and you will know more than any maturity assessment will tell you.
- Can a client get a price from your firm before a lawyer looks at the matter?
- Does work reach a model before it reaches a person, as a rule rather than as an exception?
- Is your precedent bank machine-readable, or is it a folder structure?
- When a matter closes, does the firm learn something, or does one lawyer?
- Do you sell an outcome or an hour?
- Can you show, for any piece of AI-assisted work, who reviewed it and against what standard?
A firm answering yes to five or six of those is AI-native, whatever its headcount. A firm answering yes to one or two has bought software.
Where this actually lands
The future of law firms is AI-native. It will not arrive as a wave of new firms replacing old ones, and it will not be bought at a price only the Am Law 20 can pay.
It will arrive as a slow separation between firms that rebuilt their operating model and firms that bought a licence and called it done. The new entrants have proved the model works. Kirkland has shown what the asset really is. What remains is for everyone in between to notice that the expensive part of the exercise is understanding your own firm well enough to encode it, and that this is work only your firm can do.
The firms that start that work now will be AI-native before the firms waiting for the budget to appear.
