A theory of the future of law

This one performs the work

Every previous wave of legal technology digitised the work around the law. Document management moved the files, practice management moved the time recording, e-discovery moved review into a browser. The work itself - reading, drafting, judging - stayed with the lawyer, and so did the business model built on top of it.

This wave is different in kind, not in degree. The explosion of large language models handed every lawyer a machine that reads, drafts and reasons, and that is the catalyst for the shift we are now living through. For the first time the technology performs the work itself - reading the data room, drafting the first cut, applying the checklist - rather than moving it somewhere more convenient. When technology changes where work happens, you get a cycle. When it changes who does the work, you get a structural shift, and structural shifts reprice industries.

The tools already making noise - Harvey, Legora and the wave forming behind them - are the leading edge of this, not the whole of it. In our view they are the beginning of the impact rather than the extent of it: what they prove is that legal work can now be performed by software at all. The repricing that follows from that fact is still mostly ahead of us.

Cycles change your tools. Shifts change your business.

Margins are squeezed from every side at once

Each pressure on its own would be manageable. Clients, with a clearer view of what AI makes cheap, expect efficiency to show up in fees. Fixed and capped arrangements keep spreading. The cost of the talent pyramid keeps rising while the leverage model that funded it - juniors billing hours on work AI now does in minutes - quietly erodes underneath it.

What makes this moment different is simultaneity. The pressures arrive together and they compound: pricing pressure makes leverage harder to fund, thinner leverage weakens the training ground, and every efficiency a firm demonstrates becomes the client's new baseline. In our view no single pressure is fatal. The combination is what reshapes the market.

Your client runs the same models you do

In-house legal teams now have access to the same AI tools as their external firms - sometimes better, because they can wire them straight into the business. The first contract review, the first-pass risk answer, the routine policy question: increasingly these are handled inside, in minutes, without a matter ever being opened.

That does not make external counsel obsolete. It changes when the phone rings. The call comes later, on narrower questions, with the routine work already done. And every call that never happens is not just unbilled work - it is a lost touchpoint, and touchpoints are where relationships, and the next instruction, come from.

There is a symmetry here worth sitting with. If building has become cheap enough for you, it has become cheap enough for your client - in-house teams will not just use AI, they will build with it. Read defensively, that means they need you less. Read the other way, it is the opening: the firm that turns its expertise into products can sell them to the very clients now building for themselves, and be the one they buy from before a start-up gets there first.

The platforms firms buy are becoming firms

The same platforms being sold to law firms are also the enabling layer for AI-native competitors. This is not hypothetical. Orbital Witness - a legaltech company - launched Farringdon, a regulated legal services business. A technology company became a legal services provider, because once the technology performs the work, acquiring the licence to sell that work is the shorter step.

And it is no longer a single example. Garfield AI became the first fully AI-powered law firm authorised by the SRA. Eudia stood up a regulated, AI-augmented legal business in the United States for enterprise contract and diligence work. Y Combinator has gone further still, publicly urging founders to start AI-native law firms and compete with incumbents directly rather than sell them software. When the people who fund the next decade of companies point at your market and tell founders to attack it, that is not a forecast. It is a starting gun.

We think the same gravity will act on Harvey, Legora and every platform that reaches scale. Their investors will eventually ask the obvious question: why sell tooling to law firms at software prices when the legal work itself is the far larger market? Not every vendor will cross that line. But the pressure to cross it is structural, and firms should plan as if some will.

There is a fair objection here: software companies rarely move into services. Salesforce never became a sales agency - the services would only drag its margins down. Usually right, because normally the software and the service are two different things, so the vendor stays in software. Legal is the exception. Once the software performs the work, the software is the service, and the step from selling the tool to selling the outcome is a short one.

Look closely at an AI-native law firm and what you see is not a firm that happens to use AI. It is legal expertise, encoded, wrapped in software, and sold at product economics. That is the competitive threat in one sentence:

An AI-native firm is a productised firm.

The moat did not disappear. It moved.

Read the above and the conclusion looks bleak: the vendors have the technology, the engineers and the capital. But that reads the board wrong. When the core intelligence is a commodity - rented by the token, the same few frontier models available to Harvey, to Legora, to an in-house team, to you - the model stops being what sets anyone apart. Defensibility drains out of the technology layer. It has to settle somewhere, and it settles in the two things a rented model cannot hand you: proprietary expertise, and the data that expertise generates.

This is what the vendors are actually chasing. Not your work for its own sake - the two assets that now constitute the moat: legal expertise and client trust. That is precisely why they are turning themselves into firms. You start with both. So the real contest was never whether a firm can out-build Harvey. It cannot, and it does not need to. The question is who closes their gap first: the vendor acquiring expertise and trust, or the firm acquiring the ability to build. The collapsing cost of building is what makes your side of that race winnable.

When everyone runs the same models, the model is not the moat. The expertise is - and that is already yours.

Defend the base. Claim the new ground.

The comfortable instinct is to retreat upmarket: let the commodity work go and keep the premium advisory. We think that reads the funnel backwards. Commoditisable work is where relationships start - the routine review that becomes the restructuring mandate, the lease portfolio that becomes the development joint venture. It is the entry point to everything the firm prizes.

Lose the routine work and you do not just lose its revenue. You lose the pipeline that feeds the premium engagements you kept. Defending the base is not nostalgia for volume work - it is defending the top of the funnel.

Productise the expertise

There is one move that addresses both fronts at once: turn what the firm knows into products clients can buy repeatedly. A monitoring platform with the firm's judgement encoded in its tests. A diligence agent tuned to a client base. A compliance product that turns bespoke advice into a fixed-fee line. Products defend the routine work by making the firm the natural home for it - faster, more consistent, and backed by the firm's name - and they create revenue that scales without headcount, which the billable hour never will.

Say the quiet part, because every partner is already thinking it: a fixed-fee product eats the hours a junior used to bill. It does. But those hours are leaving regardless - the client who can run the review in-house, or buy it from an AI-native firm, will not keep paying by the hour out of loyalty. The only real choice left is who does the cannibalising: you, at product margins, or someone else, at no margin to you at all.

And there is an honest difficulty: productisation requires product craft, and law firms have never needed it before. Very few firms carry the product managers, designers and applied-AI engineers this work demands - and the best of those people generally choose technology companies, where the equity, the tooling and the peer group live. That is not a criticism of firms. It is a hiring-market reality that any credible plan has to deal with.

That was decisively true in 2022. The collapsing cost of building has quietly repriced it. You no longer need fifty engineers - you need a few good ones and something worth building. Look at who is already shipping: Tacit built its own contract platform; Avantia built its own AI; small outfits standing on the very same commodity models everyone else rents. If a start-up with none of your expertise can build that, a firm sitting on decades of precedent and a book of live matters certainly can - with a team you could count on one hand. The craft is real and still scarce. The scale it once demanded is not.