AI Won't Differentiate Your Law Firm. Thinking Will.

There's a real argument running through legal tech right now about what tools like Harvey and Legora are actually for, and whether they're about to be swallowed by the very AI they're built on.

A quick translation first, because the jargon does the industry no favours.

Harvey and Legora are what the industry calls “wrappers”. They don't build their own artificial intelligence. They sit atop the big underlying models — Claude, built by Anthropic, and ChatGPT, built by OpenAI — and add a layer of legal polish around them.

That's a more precarious place to stand than it sounds. The companies that own the underlying models have more money, wider reach, and one decisive advantage: AI is sold by the unit, and they own the units. They can sell that usage cheaply, even at a loss, to win the market. A business that rents those same units and resells them can never win that fight.

The models are also growing up. They began as something you chat with, ask a question, and get an answer. Now they can act: open the documents, work through a sequence of steps, use other systems, and finish a task rather than just describe it. That ability is what people mean by an “agent”. And here's the direction of travel: Harvey now runs inside Claude itself, with the heavy lifting done by Claude's own agents. What's left of Harvey is largely the screen you log into. And that screen is starting to look a lot like Claude's.

Legora has read the danger more clearly. Its recent purchase of Qura — a business that has built genuinely proprietary legal research — is an attempt to own something the model makers can't simply replicate. Late, perhaps. But the right instinct.

There's one line in all this worth stopping on, because it's the one that matters most. The underlying models remain deeply unintuitive to actual lawyers. Connectors, skills, workflows, log-ins — most lawyers have tuned out before the sentence ends. The mass market is wide open.

Here's where I land differently from most of this commentary.

The lesson isn't “pick the wrapper that survives”. It's that no shared tool — wrapper or raw model — makes a firm different. These tools will be good at processing and at cutting the cost of the work they can handle. But they hand everyone the same vanilla capability with a different logo on it.

Differentiation comes from thinking. From what your firm understands about your clients, no general model ever will — and from building tools that encode that understanding.

And this is the part I most want business development people to hear. This isn't a threat to BD. It's the moment BD has been waiting for.

For years, we've had to describe a firm's edge in a brochure. “We put clients at the heart of our business” was always met with a wry smile — where else were you going to put them? Now we can build systems that genuinely know more about a client than the client realises about itself, communicate that far more easily, and hold real relationships steady with the right prompts and a clear view of who knows whom across the firm.

Differentiation becomes a battle of ideas and actions, not a race to license the same software as everyone else. The wrappers commoditise. Thinking doesn't.

The advantage goes to the firms and the BD teams, who build what they don't share.

So here's the question worth sitting with: if your firm's real edge was something you built rather than something you rented, what would it be?