How the Most AI-Pilled Company in the World Works

The episode opens with headlines on legal AI, market pressure, and AI governance, then focuses on Spawn founder Jacob Sanberry’s playbook for running a 10-person, agent-first company. CIOs should care because Spawn shows how persistent agent fleets, automated feedback loops, and AI judges can materially change software delivery, operating costs, and org design.

Key points

• Spawn runs a persistent agent fleet with named roles, not temporary copilots, and shipped 77,000 code changes last month.
• The company spent $3.1 million on tokens last month, about 30 times payroll, on 5,000 CPUs, 17 TB RAM, and 150 GPUs.
• Spawn removed pull requests, runs tests every 15 minutes, and lets agents fix bugs directly in the codebase.
• AI judges convert two years of human decisions into precedent; 11,000 decisions in eight days helped raise output from 1,000 to 7,000 changes daily.
• Human work shifts to taste, deleting obsolete work, and setting ambition; closed-loop feedback is essential for agentic systems.
• A separate headline noted legal AI startup Crosby trying to buy assets and data from bankrupt Weil, Gotshal & Manges.