A hundred lawyers, a thousand lawyers’ output.

1. 1.

   The machines that now write the world’s software went to work because software was already structured; the same machines stalled in law because law is not.

2. 2.

   So the prize in legal AI is not a better chatbot. It is the structure itself: whoever turns legal practice into something machines can operate on owns the decade.

3. 3.

   The funded platforms cannot build it: software rented identically to every firm cannot hold what makes one firm different.

4. 4.

   Abroad, law firms rebuilt around machines have already shown in public what happens to output per lawyer.

5. 5.

   India prohibits every shortcut those firms used, so here the change arrives through existing firms transforming in place, and no firm can do that alone.

6. 6.

   We are the partner a firm does it with: the machines, the structured record of Indian law, and the engineers, delivered as a service, because a law firm will never employ infrastructure engineers.

7. 7.

   Firms choose this because every other option is a bad trade: knowledge that stays dead in the files, knowledge that leaks into a shared platform, or knowledge that is forbidden from remembering.

8. 8.

   The metric is output per lawyer, the work is won one engagement at a time, and the end of the road is Indian law itself rebuilt as public structure: the statutes, the regulators, the reported record. What a firm builds on that road stays the firm’s.

Everything below expands those sentences, in order.

A managing partner in Hyderabad told me that five years ago, he wanted a thousand-lawyer firm. Then he told me what he wants now: a thousand lawyers’ output. The headcount was only ever the means. Billable hours times bodies was an elaborate way of manufacturing output, and for a hundred years it was the only way. It is not anymore, and the first proofs are already public.

In the United States, Crosby, a law firm built around AI agents and backed by Sequoia, Index, and Lux, crossed a billion dollars of client contracts within a year of coming out of stealth, up from thirty million at launch, and it charges by the contract, not the hour. In England, Garfield became the first firm the regulator ever authorised to provide legal services through AI, in May 2025; thirteen months later it won its first trial, a junior barrister on their feet in court and the machine behind every document, the client paying about £400 in fees to recover £7,000. Pierson Ferdinand launched in January 2024 as the largest law firm debut in American history, 130 partners and no associates; two years on it runs more than 270 partners, crossed nine figures in revenue, and machines do the junior work.

These are not software companies. They are law firms that changed what one lawyer’s day produces, and said so out loud.

Every Indian partner reading this list has the same two reactions. First: that cannot happen here. Correct: the Advocates Act forbids the structures every one of those firms stands on. Second: if my competitor gets there first, I have a problem. Also correct. This essay is about the narrow door between those two facts.

For thirty years, software promised to change legal practice and mostly changed its filing. The reason is not that lawyers resist technology. The reason is that legal work never existed in a form software could touch.

A PDF is a photograph of a page. A Word file is formatting wrapped around text. The thing that actually matters, what a clause means, what a regulator punished, what the law said on the day the deal signed, lives in partners’ heads and dies with the matter. Programmers had it differently: fifty years of files, versions, tests, and structure meant that when the models arrived, they arrived as the last mile on a finished road. Pointed at software, a frontier model works for hours and ships. Pointed at a law firm’s document system, the same model is a genius locked in a junkyard. It can describe every blob it picks up. It can build nothing, because nothing connects to anything else.

That is the entire story of why legal AI disappoints. Everyone is selling the last mile. Nobody built the road.

Structure. A firm’s executed deals broken into typed positions a machine can query. A regulator’s entire published record as data instead of ten thousand PDFs. The statutes a firm practises under as versioned text, so the law as it stood on 14 March 2019 is a lookup, not an afternoon.

Once practice is structure, the machine stops guessing and starts operating. Comparing indemnity caps across forty deals becomes a query, not forty documents read by an associate. Reviewing the other side’s draft becomes a comparison against every position the firm has ever negotiated. And every answer gains a property legal AI has never had: it can be checked, mechanically, against the text it came from.

We have been building that road for Indian law in public. The published record of Indian competition law, over 3,200 orders and appeals, runs as structured data with free public search. The Foreign Contribution Regulation Act runs as the first date-addressable Act of Parliament in India: pick a date, read the law as it stood. Both are open. They are samples of a category: the same structure, built on a firm’s own record, for that firm alone.

Structure is invisible, so we built the machine that makes it visible. On our racks, a firm employs computers. Not chatbots: computers. Each one exists for a single matter. It has read everything you gave it, its notes are in order, its drafts are versioned, and every morning it checks whether the law moved. No one logs into it, because it has no login, no screen, and no keyboard. It does not need a user. It is the user.

You can watch one work today. A partner briefed a machine at 11:00 in the morning with a distribution dispute: three research questions, a memo, a notice, a markup of the other side’s draft. At 12:27 the machine refused its own draft, because seven quotations did not match the source record, character for character. It fixed them, verified 170 of 177 claims against original text, flagged the rest in the memo’s final section, and returned the work at 12:50. One hundred and ten minutes. The bill was ₹8,211.46, itemised. The matter is an invented one, built to be shown in public. The run is not: the work, the refusal and the bill are real and unedited, and the full record is open to anyone.

Two things about that run matter more than its speed. The machine could refuse, because its work runs against structure instead of against guesswork; checking a quotation is arithmetic, not judgment. And nothing left for the other side without a lawyer deciding it should: the machine produces work product, never advice. The lawyer verifies, the lawyer signs. That division is permanent, and it is the entire design. It is also why a public product exists at all: one machine on one matter is how a lawyer judges the work before a firm trusts it with a practice.

Every route those firms took is closed here. Outside capital inside a law firm, non-lawyer ownership, a regulator that authorises legal services delivered through AI: the Advocates Act and the Bar Council rules prohibit all of it, and there is no sign that changes.

So in India the wave has exactly one legal channel: existing firms, transforming in place. And a firm cannot do it alone, because the engineers who build machine infrastructure do not exist at legal-industry salaries, and a lone hire without a finished stack spends two years rebuilding what already runs before delivering a single win.

Everywhere else in the world, regulation is legaltech’s excuse. In India it is the moat, for the firms that move. The route the law leaves open is a partner that brings the road, the machines, and the engineers into your walls, as a service. That is the company we built. Machines are employed, never admitted. The output is the firm’s. The signature is always a lawyer’s.

A firm’s edge is what its partners know: how they negotiate, what their regulator actually punishes, which clauses survived which fights. Today every way of putting that knowledge to work is a bad trade. Keep the documents away from AI, and the knowledge stays dead in the files. Upload them to a shared platform, and the edge seeps into a vendor that serves your competitors the same software it serves you; the leading platforms’ own security promises concede that what they hold, they must forget. Either way, nothing compounds.

There is a fourth option: structure the firm owns, on machines that answer to no one else, compounding privately, matter after matter. The knowledge finally works, and it works for exactly one master. For the giants, renting the same software as every other giant is a wash. For the ambitious firm outside the top tier, it is worse than a wash: renting what your bigger competitor also rents preserves the gap forever. Owning your own leverage is the only move that closes it.

The split is permanent, and worth stating plainly. What a firm builds with us stays the firm’s: its record, its playbooks, its negotiated positions, exportable on the day it asks. What we keep is the road: the machines and the tools that run them, the public record of Indian law we rebuild in the open, and everything we learn about doing this work, which makes the next firm’s machines better and cheaper than the last. The firm’s knowledge compounds privately. Ours compounds across firms, without carrying anyone’s papers.

Output per lawyer. It is the number managing partners already maximise; the pyramid of juniors billing hours was a way of faking it with headcount. Every engagement we take is a before and after on that number, measured on the firm’s own matters. Not hours saved, not seats adopted: what one lawyer, with machines, produces by Friday.

A hundred-lawyer firm does not need to become a thousand-lawyer firm. It needs the output. That is now a buying decision, and in India it is a decision only a firm can make for itself, inside its own walls, with a partner the law permits.

The firms that own their leverage are going to eat the firms that rent it. We arm the side that chooses to own.

If you are a lawyer: brief a machine on a matter tonight, and read its record in the morning. Every matter runs sealed, and nothing you upload is used to teach the system.

If you run a firm: we build this inside your walls, on your record, for you alone. The first conversation is with me.