Supabase engineers report 60% of code written by AI agents

Supabase internal engineering data indicates a rapid increase in AI code generation adoption, rising from 10% of engineers writing 80% of their code with agents to 60% of engineers doing so within six months. The company has no formal AI policy, allowing engineers to freely use tools like CloudCode and Codex. This shift has led to a significant increase in testing requirements, with engineering effort moving from building features to testing specific use cases to ensure reliability.

Supabase has re-architected its CLI and error messages to be agent-friendly, providing specific instructions for agents to remedy errors rather than just reporting them. The company runs internal evaluation suites to test which models perform best with Postgres and Supabase APIs. Paul Copplestone noted that agents are now capable of self-managing database instances and executing complex tasks, such as building personal finance or health tracking applications, without human intervention in the code architecture. Copplestone believes that the current state of AI models is sufficient for engineers to rely on them for substantial portions of their workflow, provided that robust testing and agent-optimized error handling are in place. "We don't have an AI policy internally. We just let... people do whatever they whatever they want and we give them access to to whatever tools they they want to use," Copplestone said.