From D&D to $8B: How a Lawyer Built Harvey AI

According to TechCrunch, Harvey has become one of Silicon Valley’s fastest-growing AI companies with an $8 billion valuation as of late October 2025, up from $5 billion in June and $3 billion in February. The legal AI startup now serves 235 clients across 63 countries including most of the top 10 US law firms, and hit $100 million in annual recurring revenue by August. Founder Winston Weinberg started as a first-year associate at O’Melveny & Myers before cold-emailing Sam Altman and OpenAI’s general counsel Jason Kwon on July 4th, which led to the OpenAI Startup Fund becoming their first institutional investor. The company has since attracted backing from Sequoia Capital, Kleiner Perkins, Google Ventures, Coatue, and Andreessen Horowitz while growing to about 400 employees.

Dungeons & Dragons to legal revolution

Here’s the wild part: Weinberg’s first real use case for GPT-3 wasn’t legal work at all—it was running Dungeons and Dragons games with friends. But when he got assigned a landlord-tenant case he knew nothing about, he and his co-founder Gabe Pereyra developed chain-of-thought prompting before it was mainstream. They tested their system on 100 Reddit legal questions, and on 86 of them, actual lawyers said they’d send the AI-generated answers with zero edits. That was the lightbulb moment. Basically, they realized this wasn’t just a productivity tool—it could actually transform an entire industry.

The cold email that changed everything

Weinberg didn’t follow the typical Silicon Valley playbook. He had no tech connections, didn’t grow up in San Francisco, and didn’t even know who the top VCs were. So how’d he raise hundreds of millions? He sent one perfectly targeted cold email to OpenAI’s general counsel—because only another lawyer would understand if the outputs were actually correct. The call happened on July 4th morning, and OpenAI’s startup fund wrote a check almost immediately. Weinberg’s fundraising philosophy is refreshingly simple: just make your company perform incredibly well, and the money will find you.

The multiplayer problem

Now here’s where things get really interesting. Harvey’s biggest technical challenge—and potentially its strongest moat—is building what they call a “multiplayer” platform. Think about it: law firms work with multiple clients simultaneously, and accidentally sharing confidential data between competing clients would be catastrophic. They have to solve both internal permissioning and external ethical walls across dozens of countries with different data residency laws. Germany and Australia, for instance, won’t let financial data leave their borders. So Harvey has to maintain expensive compute infrastructure in every country they operate in, even if they only have a handful of clients there initially.

More than just a ChatGPT wrapper

Critics call Harvey a fancy ChatGPT wrapper, but Weinberg argues their real advantage comes from two things: workflow data and multiplayer capabilities. They’re collecting massive amounts of data about how lawyers actually work, which helps them build evaluation frameworks that can self-assess quality. And their platform serves both law firms and corporate legal departments—something no competitor has managed yet. The business model is shifting from seat-based pricing to outcome-based pricing as workflows get more complex. Honestly, if they can actually pull off this multiplayer vision while maintaining accuracy across thousands of documents and multiple stakeholders? They might just redefine how legal work gets done entirely.