Minions: Stripe's one-shot, end-to-end coding agents

Alistair Gray’s architecture follow-up to the first minions post, with the rate now at over 1,300 minion-produced PRs merged a week, up from 1,000 in Part 1. Little of the design is exotic. What makes it work is that each piece already existed for human developers.

Runs happen on devboxes, EC2 instances pre-provisioned so one is ready in about ten seconds. Stripe describes them in the usual DevOps terms: cattle, not pets, standardised and easy to replace rather than bespoke and long-lived. On top sits a fork of Block’s goose agent, running with full permissions inside that box. Orchestration comes from blueprints, which are workflows defined in code that mix deterministic nodes (lint, push) with agentic ones (implement the task, fix the CI failures). Context comes from Cursor-format rule files scoped to subdirectories and from Toolshed, now at nearly 500 MCP tools, with each agent given a curated subset so the tool list does not swamp the context.

Feedback is deliberately layered by cost. A background daemon lints locally in under a second, then CI runs, auto-fixes are applied, and the agent gets one more attempt. After the second CI run the branch goes back to its human operator regardless. Two rounds, then stop.

The conclusion is the same one as Part 1, stated more directly: what’s good for humans is good for agents, and building on this infrastructural primitive paid dividends as a natural home for LLM agents. The isolation is not just safety, it is what makes the runs reliable enough to trust unattended. The knowledge-work counterpart to all this is Kai.