Minions: Stripe’s one-shot, end-to-end coding agents | Stripe Dot Dev Blog
- URL: https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents
- Saved: 2026-04-10
Across the industry, agentic coding has gone from new and exciting to table stakes, and as underlying models continue to improve, unattended coding agents have gone from possibility to reality.
Minions are Stripe’s homegrown coding agents. They’re fully unattended and built to one-shot tasks. Over a thousand pull requests merged each week at Stripe are completely minion-produced, and while they’re human-reviewed, they contain no human-written code.
A typical minion run starts in a Slack message and ends in a pull request which passes CI and is ready for human review, with no interaction in between. Engineers often spin up multiple minions in parallel.
Why Stripe built its own system:
- Stripe’s codebase spans hundreds of millions of lines across large repositories
- The stack is unusual (for example Ruby without Rails plus Sorbet)
- There are many internal libraries and strong compliance / financial correctness constraints
- Existing developer productivity foundations already existed around source control, environments, code generation, CI, and more
How minions work at a high level:
- Minion runs start in isolated pre-warmed devboxes that are ready in about 10 seconds
- The agent loop is built on a fork of Block’s goose
- Deterministic steps are interleaved with agentic work for git, linting, testing, and other required actions
- Rule files are mostly conditional and attached by directory or pattern rather than globally
- Minions use MCP to gather context such as internal docs, build status, ticket details, and code intelligence
- Stripe built a centralized internal MCP server called Toolshed with hundreds of tools
- Local feedback is preferred before CI; CI is used selectively and bounded to a small number of rounds