Crosby: The Inner Workings of a Neofirm — New Ontologies
An AI-native law firm that's 10x faster than traditional firms isn't just using better tools—it's a new organizational form where elite lawyers orchestrate AI agents, fixed fees replace billable hours, and profits fund R&D instead of partner bonuses.
Read Original Summary used for search
TLDR
• Crosby reduced contract review from 58 minutes to 4:53 by building custom eval systems, encoding client preferences as living playbooks, and creating infrastructure for long-running stateful AI agents
• The "neofirm" model breaks traditional law firm economics: corporate structure vs partnership, fixed fees vs billable hours, reinvesting profits in R&D vs distributing to partners
• Lawyers at Crosby run evals and shadow engineers—they're "legal engineers" who orchestrate AI armies rather than doing rote work, creating "lawyer utopia" where professionals only solve intellectually tricky problems
• The hardest technical frontier is turning law from non-verifiable (judgment-based) to verifiable through game theory and reinforcement learning—modeling client preferences and counterparty negotiation patterns
• Trust remains deeply human: clients still want to know who's responsible, and the best lawyers are becoming more relational as transactional work gets automated
In Detail
Crosby is an AI-native law firm that represents a fundamental reimagining of professional services structure. Founded by Ryan Daniels (Stanford-educated attorney) and John Sarihan (early Ramp engineer), they've raised $85M and serve customers like Ramp, Cursor, and Clay. The "neofirm" model combines elite lawyers from Kirkland & Ellis and Wachtell with engineers from Ramp and Stripe, charging fixed fees instead of billable hours and reinvesting profits into R&D rather than partner distributions.
The technical challenges are substantial. Law is a non-verifiable domain where two smart lawyers can legitimately disagree about the "correct" answer. Crosby is building custom eval systems that automatically funnel failure cases, encoding client preferences as living playbooks that remember instructions across all formats, and creating infrastructure for long-running stateful agents that can't simply crash and restart. The long-term vision involves game theory and reinforcement learning—understanding how to negotiate against the same counterparty for the 11th time by modeling their preferences and patterns. By owning the full service delivery, they collect unique interaction data from lawyer evaluations across the entire negotiation lifecycle.
The cultural innovation is equally important. Lawyers at Crosby are "legal engineers" who run evals and build playbook schemas. Engineers shadow lawyers for weeks to understand why they accept certain redlines. The goal is "lawyer utopia"—where professionals only do expert work, orchestrating armies of AI agents while working reasonable hours. But trust remains deeply human: clients still want to know who's responsible, and the best lawyers are becoming more relational as transactional work gets automated. As one lawyer put it, "a contract review is a conversation between two humans—it entails feeling." The piece suggests this is a blueprint for how all knowledge work professions will transform in the AI age.