Winston Weinberg left O'Melveny & Myers after just a year as a junior associate to co-found Harvey, a legal AI start-up now valued at $11bn. The inspiration came in late 2021 when co-founder Gabriel Pereyra showed him GPT-3. Weinberg tested it on a landlord-tenant pro bono case, asking three lawyers to evaluate 100 AI-generated answers without knowing their origin. 86 passed.

Weinberg sent the results to OpenAI executives Sam Altman and Jason Kwon, who confirmed they hadn't seen GPT-3 perform that well in legal work. Chain-of-thought prompting proved particularly effective for applying statutes to fact patterns. The pair initially pitched a landlord-tenant tool, but GPT-4 shifted the thesis toward building an enterprise platform that could transform how law firms operate.

The real competitive moat, Weinberg argues, isn't legal research—it's routing context like jurisdiction, practice area, case history, and client relationships into specific outputs. Harvey's dual-sided market approach, serving both law firms and their clients like private equity shops, creates switching costs. Institutional knowledge that once lived in senior partners' heads now gets codified, making client relationships stickier than ever.