Building on Stripe's first Minions account, the implementation follow-up explains how unattended coding combines model judgment with fixed software steps. Its blueprints let an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → decide how to implement a change, while ordinary code runs required operations such as linting and pushing the branch. Local checks catch inexpensive fixes before the larger test pipeline runs. If that pipeline fails, the agent gets one repair attempt before a second run and human handoff. This gives the agent freedom over the implementation while keeping required checks and the stopping point explicit.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
