San Francisco’s office vacancy rate sits at historic highs. Tech companies are dumping real estate faster than they can renegotiate leases. Yet Harvey, a legal AI startup, just expanded its downtown headquarters to 150,000 square feet.
This isn’t some desperate grab at cheap space. Harvey raised $200 million at an $11 billion valuation before signing the lease. They’re doubling down on physical presence at a moment when most AI companies are going remote-first or hybrid-minimal.
What This Tells Us About AI Infrastructure
As someone who builds bots for a living, I watch how AI companies structure their operations. The tools we use to build intelligent systems don’t care where we sit. You can train models from a coffee shop. You can deploy agents from your bedroom. The technology is location-agnostic.
But Harvey’s expansion suggests something different is happening in legal AI. When you’re building systems that handle sensitive client data, attorney-client privilege, and regulatory compliance, the infrastructure requirements change. You need secure environments. You need spaces where lawyers and engineers can work side by side, iterating on prompts and testing edge cases in real time.
Legal work isn’t like writing marketing copy or generating images. A hallucination in a legal brief can cost millions. A misinterpreted statute can tank a case. The stakes demand a different development process.
The Real Estate Signal
Harvey’s move tells me they’re hiring aggressively. You don’t lease 150,000 square feet for a small team working three days a week. This is a bet on sustained growth and in-person collaboration.
For bot builders, this matters. It signals where the serious money is flowing in AI. Not into consumer chatbots or content generation tools, but into specialized vertical applications where accuracy and reliability trump speed and scale.
The legal industry has been notoriously slow to adopt new technology. Lawyers still bill by the hour. Firms still organize around partner hierarchies that predate the internet. But when a legal AI company can command an $11 billion valuation, something fundamental is shifting.
What Bot Builders Can Learn
Harvey’s success points to a specific strategy: go deep in one vertical rather than broad across many use cases. Legal AI requires domain expertise that takes years to build. You need to understand civil procedure, case law, regulatory frameworks, and how different jurisdictions handle similar issues.
This isn’t a weekend hackathon project. It’s not a wrapper around GPT-4 with a legal-themed interface. Building reliable legal AI means creating custom training pipelines, validation systems, and safety checks that don’t exist in general-purpose models.
The real estate expansion also suggests Harvey is building for enterprise sales cycles. Law firms move slowly. They need extensive pilots, security audits, and partner buy-in before adopting new tools. You can’t service those relationships entirely over Zoom. You need account teams, implementation specialists, and support staff who can show up in person.
The Contrarian Bet
Harvey’s expansion is a contrarian move in a market that’s supposedly moving toward distributed work and lean operations. But contrarian bets are often where the biggest returns hide.
For those of us building AI systems, the lesson is clear: the future of AI isn’t just about better models or faster inference. It’s about building the organizational structures and physical infrastructure to deploy those models in high-stakes environments.
San Francisco’s office market might be struggling, but Harvey’s expansion shows that certain types of AI work still benefit from centralized, in-person collaboration. The question for other AI companies is whether they’re building the kind of products that justify that investment.
Sometimes the smartest move is to zig when everyone else zags. Harvey is betting that physical presence matters in legal AI. Given their valuation and growth trajectory, they might be onto something.
🕒 Published: