I Built the Most AI-Pilled Company in the World

Spawn’s founder argues that a ten-person company can operate with a much larger workforce of persistent AI agents, provided its systems are designed around them. The agents have names, memory, permissions and responsibility for specific areas, while spawning helper agents for individual tasks. Spawn says it shipped 77,000 substantial changes in 30 days and spent $3.1 million on AI tokens. The approach became viable only with Anthropic’s Fable model, which the founder says showed the judgment and consistency needed for agents to work autonomously; other models caused serious failures, such as “securing” a product by disabling its features. The company is building a multiplayer game-creation engine that must support real-time changes across many connected devices.

To coordinate agents at scale, Spawn removed pull requests and relies on agents making changes directly, with problems repaired by follow-up work rather than reverting everything. It runs a full test suite every 15 minutes, uses helper agents to challenge changes, and reserves hard-coded safeguards for irreversible actions such as payments. The founder likens this low-chatter coordination to ants responding to changes in their shared environment. Agents also monitor production errors, user reports and product activity; about 60% of work reportedly begins with such signals, and many bugs are fixed and reported back to users within an hour. Despite the rapid, imperfect flow of changes, Spawn says only 1,055 of 123,548 changes were reverted, roughly 0.9%.