Everyone will have AI. So what will actually differentiate a law firm?

Kirkland & Ellis has committed half a billion dollars over three to four years to building its own AI platform, with a hundred million of it going out this year. Freshfields is co-building with a frontier model company. The Financial Times reported earlier this month that something like a fifth of large firms have stopped buying off the shelf entirely.

Ropes & Gray has gone the other way and said so publicly. Their view is that the technology is moving too quickly for proprietary development to make sense, and that technology companies will do this better than a law firm can.

Both positions are defensible, and the debate has been framed as build versus buy. We think that framing is a distraction, and that it will look strange in three years.

Here is why. Whichever route a firm takes, it ends up with access to broadly the same underlying reasoning capability. The models improve for everybody simultaneously. A firm that spends five hundred million and a firm that buys a subscription are, in terms of raw capability, closer to each other than either would like to admit. Capability is becoming a utility, and nobody has ever built a competitive advantage out of a utility.

So what is left?

The most interesting detail in the Kirkland announcement was not the money. It was that around two hundred and fifty of their own lawyers contributed to designing the platform. That is not a technology investment. That is an attempt to encode how one specific firm practises law into a system, which is a knowledge exercise wearing a software budget.

That points at where the actual differentiation sits, and it is not the model.

It is whether a firm's own history is usable. Every good firm has decades of negotiated positions, opinions, precedents and matter correspondence. Almost all of it is technically stored and practically unreachable. The firm that can retrieve and reuse its own past work has something no vendor sells and no competitor can copy.

It is whether the work is organised so a system can see it whole. A model reasoning over fragments produces fragmented answers, and most firms hold their matters in fragments.

And it is adoption, which is the least glamorous item on the list and probably the largest. A firm with four hundred licences and thirty genuinely expert users has a return problem that no amount of proprietary development will fix, because the same three hundred and seventy people will underuse a bespoke tool too.

None of those three require half a billion dollars. All three require decisions most firms have not made.

The uncomfortable implication is that a mid sized firm that gets knowledge, structure and adoption right may end up further ahead than a large one that bought the most expensive answer to the wrong question.