The Pint and the Corgi

I will take that pint, Quentin Solt. Next round is on me, regardless of where any of this lands. I will happily admit that our #TrojanHorseDebate has sharpened my personal thinking on legal tech and refined our investment thesis at Grayhair Venture Partners, and that is obviously worth a lot more than a couple of pints.

On Tenet: cost, not control

You have reached for control because control is the more flattering explanation, and because it survives longer as an argument than a spreadsheet does. But look at what actually happened. Harvey's margins (allegedly) went negative because token consumption rose twenty-fold and every one of those tokens was purchased at retail from someone with more pricing power than Harvey has. Sovereignty is not at issue here, we are faced with unit economics that require improvement.

Now, we at GrayHair have been backing companies with superior architecture (better accuracy per dollar, not just better accuracy) since we started writing checks in this space, precisely because it was obvious that renting frontier intelligence at frontier prices for commodity tasks was not necessarily the soundest business model. But, equally, we knew that you can fund an unsustainable cost structure indefinitely if the capital keeps flowing. What you cannot do is call the moment the capital gets nervous a "strategic pivot." Bloomberg's article is the invoice; nothing more, nothing less.

Call it control over your own destiny if you like the phrase (I won't begrudge you the flourish, and I do admire your writing style), but it is still a pricing decision, not governance.

On building your own stack

I have said this before and Tenet does not change it: there is nothing fundamentally wrong with a hybrid approach, building some and renting some. Where I would beg to differ (slightly) is on the shape of the contest you have drawn. No law firm is competing with Anthropic or OpenAI, and none should be foolish enough to try — the people keeping that particular genie on some kind of leash are not the people Kirkland is going to out-hire, and that is not the layer at which Kirkland is trying to compete. Kirkland is trying to out-build Harvey (if that; we have it on reliable authority that Kirkland is buying/renting a lot of solutions, so the whole notion of "owning the stack" is a bit of a red herring).

For everyone without Kirkland's revenues, the answer is - categorically, absolutely and resoundingly - not to own the stack. It is to never stop watching it. That is the actual significance of firms appointing so many Heads of AI alongside KM Managers (as recently pointed out by Ted Theodoropoulos. From where I sit, this is an admission that horizon-scanning has become a permanent function, not a project with an end date. Even a firm that can afford the best off-the-shelf platform today will find it stops being Good Enough in eighteen months, because the practice underneath it is being rebuilt in real time. Standing still is the one strategy guaranteed to fail.

On insurance

Orbital already has AI outputs underwritten for residential property transactions in the UK — a live policy, not a promise. Corgi is doing something similar more generally. And Crosby has announced it will insure its agents' outputs, moving toward autonomous work its lawyers no longer review line by line, though the carrier and the evals underpinning it are still being worked out with regulators. That is not evidence the human backstop is quietly becoming unaffordable, as you would have it. It does, however, offer some evidence that genuine service-delivery innovation is already insurable, in at least one case, and getting there in others — which is exactly the kind of structural advantage worth backing rather than an ambient risk worth fearing.

On the frontier

Your 4K/8K metaphor is elegant, but I cannot fully agree with it. It assumes legal capability has a fixed ceiling, and the only question is how cheaply you can rent your way up to it. That may have been true when Harvey showed up, and the whole proposition was "better chatbot than the associate had." It stopped being true roughly the moment we started underwriting company brains, real data-driven business and legal insights (shout out to Andrew Mancilla at OurFirm.ai for their predictive analytics relating to judicial philosophy) and digital twins (another shout-out to Lewis Z. Liu, whose Twin-1 captures a senior partner's judgment for asynchronous access). This is not compression of existing capability - it opens the door to a head-spinning potential of what is possible, an expansion of what is for sale in the first place. The Legal Capability Frontier is not a fixed resolution you are metering access to. It is a boundary that keeps moving, and it is moving (in part) because investors like us are underwriting the startups moving it, going beyond LegalTech 2.0 (which, by the way, does not cease to be valuable simply because the next wave has appeared).

That is also my answer to your best question — what does a firm sell once the client can buy the advice, the insurance and the regulatory comfort without the firm? Whatever currently sits past the frontier, because the frontier relocates faster than commoditization can catch it. The pie is not being redivided among fewer players. It is expanding, and the expansion is the opportunity.

It is also why I remain adamant that "landlords need tenants". Freshfields did not sign with Anthropic because Anthropic had run out of things to do. It signed because Anthropic needed domain expertise and distribution it does not have, and a law firm was the fastest way to acquire both (remember that part in the early installments of our #TrojanHorseDebate, where I mentioned that some firms are bound to break ranks and provide access to frontier models, even if some firms decide to stonewall?) A landlord that needs its tenants to help build the building is not yet ready to evict them (and I appreciate that you will feel that the operational word here is "yet").

On Jevons — turn it back upstairs

You are right that consumption expands as intelligence gets cheaper, and I do not believe that this was ever in dispute. But, as you pointed out: Harvey's tokens went up (allegedly) twenty-fold and it responded by going and building its own supply. The labs' response, roughly ninety minutes apart, was to cut prices on exactly the tier Harvey had just fled to. That is not the behavior of an entrenched landlord calmly collecting rent. That is two competitors racing each other down a cost curve because neither trusts the other to hold the line — which is precisely the treadmill Harvey was trying to get off. "Confirmed dependency," as you put it, cuts in both directions. Nobody upstairs is as comfortable as the rent cut makes them look. Not sure there are any "landlords" in this market that are sitting there looking pretty, certainly not at some companies that are currently raising money at an eye-watering valuation well north of $1 trillion...

Two things I am not answering here

The pyramid point deserves more than a paragraph tacked onto the end of someone else's argument, so I'm not going to try. How judgment gets built when the repetitions that used to build it disappear is a real question and I don't think either of us has an honest answer yet.

Same with pro bono — I have a different view on the announcement (I do think that Harvey is enabling more pro bono capacity, not renting the firms' reputations to build one, and - while I certainly like the strategy to get closer to certain portions of the market, I am not yet confident that it makes sense given the value proposition), but that is a debate on its own terms and deserves its own round rather than a rebuttal squeezed in here.

Where that leaves us

I do not think the market settles into one story any more than it settled into one law firm business model a generation ago. Some firms will sell volume, some will sell judgment, some will do both under different letterhead (Cooley GO exists precisely because commoditizing your own expertise and defending the premium version of it are not mutually exclusive strategies). That plurality is not a sign that the profession has not figured out its answer. The process of experimenting with business models is the answer.

What do you think?