Many people thought frontier models would kill vertical AI startups. Then Anthropic launched Claude for Legal… and Harvey is still worth $11B.

Everyone assumes it has some secret legal model the labs can't match. It doesn't. It runs on the same models you and I can access through an API.

And Harvey kept growing anyway. Revenue reached $190 million by January, up from $100 million just five months earlier. It now serves over 100,000 lawyers, including half of the largest law firms in America.

So if it isn't the model, what is it?

It's everything built around the model.

Harvey didn’t spend the last few years training a better LLM. It spent them embedding itself inside the world’s largest law firms. Every deployment is customized. It integrates with the firm’s documents, internal knowledge, permissions, review process, and workflows.

Anthropic can ship incredible intelligence. What it doesn’t do is work with every firm to redesign its workflows, integrate into its existing systems, pass procurement, train employees, and earn trust over years. That’s a completely different business.

The lesson isn’t about legal AI. It’s that foundation models are quickly becoming commodities. The enterprise value is being created in the workflows, integrations, and trust built around them.

Anyone can access the same model. Very few companies can replicate years of deeply embedded customer relationships.