Harvey — the legal AI that big firms pay serious money for — launched its next generation this week. The headline feature isn’t a smarter model.
It’s more robust memory.
Their system can learn a firm’s lawyers’ word choices, their drafting style, how they structure their work, and then use all of that as context inside the firm’s workflows. You can read the details at Artificial Lawyer.
Why they built it
This is Harvey’s best defense against being disintermediated by the foundational model makers — Anthropic, OpenAI, and whoever comes next.
Think about what that means. The models keep getting better, and they get better for everyone at the same time. So the model can’t be what makes Harvey worth paying for.
What’s left is everything the tool knows about how a particular firm works. That part doesn’t come free with the next release.
What it portends for the rest of us
Almost certainly, you’re not buying Harvey. It’s built and priced for firms with a procurement department.
But the logic doesn’t only apply to Harvey. If accumulated context is the durable part, then the advantage doesn’t come from which model you pay for. It comes from how much of your own work and judgment you’ve put into whatever tool you already use.
You can build a smaller version of what Harvey built, in Claude or ChatGPT, for basically what you’re already paying for those tools. A project or a custom instruction set that holds:
Your own writing — a few letters or briefs you were happy with, so it drafts like you instead of like a chatbot.
Your standing instructions — how you want work structured, what you never want it to do, the words you avoid.
The documents you keep rebuilding — your forms, your templates, the things you recreate from scratch more often than you’d admit.
True, this takes thoughtful setup and ongoing tweaking. It isn’t a weekend project you finish. So most lawyers won’t bother. But considering the long-term value, it’s almost certainly worth doing.
The alternative is renting it
If you’re at a big firm, you can pay Harvey to handle all of this for you. That’s a real option, and for some firms it’s the right one.
But the further down that road you go, the more of your firm’s working knowledge lives inside someone else’s product. Your preferences, your precedents, the way your people actually work — all of it sitting in a system you license rather than own.
I’ll leave it to you to decide what’s wrong with being dependent in that way. I’m not sure it’s a problem for everyone. I am sure it’s worth deciding on purpose rather than by default.
Where’s the transferrable value in a firm?
Which brings me back to something I’ve been chewing on.
Buying a small law firm has always meant buying the client list and hoping the clients stayed. But if a growing share of a firm’s value sits in its systems — in context that took years to accumulate and can’t be handed over in a closing binder — that changes what a buyer is actually buying.
I don’t have a clean answer. It seems like the kind of shift that shows up slowly and then all at once.
;-)
Ernie
P.S. Setting this up is exactly the kind of thing we work through together in the AI Workshop — twice a week, hands-on, instead of everyone figuring it out alone