Stripe’s ‘Minions’ Automate Coding Tasks, Producing Over 1,300 Pull Requests Weekly

Stripe, a leading financial infrastructure platform, has deployed a system of autonomous coding agents called “Minions” that are significantly boosting developer productivity. These agents are now generating over 1,300 pull requests per week, all of which are reviewed by human engineers but contain no human-written code [InfoQ]. This represents a substantial increase from the 1,000 pull requests generated during earlier trials [InfoQ].

How Minions Work

Minions are designed to execute one-shot, end-to-end tasks, differentiating them from interactive coding assistants like GitHub Copilot or Cursor [InfoQ]. Tasks can originate from various sources, including Slack threads, bug reports, and feature requests [InfoQ]. When a task is assigned, a Minion orchestrates the work using “blueprints,” which combine deterministic code with flexible agent loops to produce the necessary code, tests, and documentation [InfoQ]. The process culminates in a pull request submitted for human review.

Built on Existing Infrastructure

The success of Minions isn’t primarily due to advancements in the underlying AI model, but rather the robust infrastructure Stripe had already built for its human engineers [ByteByteGo]. This infrastructure, developed over years, provides the foundation for LLMs to operate effectively at scale [Stripe]. The system evolved from an internal adaptation of Goose, a coding agent originally developed by Block [InfoQ].

Blueprints and Reliability

Minions are orchestrated using blueprints, which define workflows by breaking down tasks into subtasks handled either by deterministic routines or by the agent itself [InfoQ]. Stripe engineers describe blueprints as a blend of agent skills and code, ensuring both efficiency and adaptability [InfoQ]. Reliability is further reinforced through CI/CD pipelines, automated tests, and static analysis, ensuring generated changes meet engineering standards before human review [InfoQ].

Supporting a Trillion-Dollar Ecosystem

The code managed by Minions supports over $1 trillion in annual payment volume at Stripe, operating within complex dependencies involving financial institutions, regulatory frameworks, and compliance obligations [InfoQ]. This highlights the importance of reliability and correctness in deploying autonomous agents at this scale.

Key Takeaways

• Stripe’s Minions are autonomous coding agents generating over 1,300 pull requests weekly.
• All code changes are human-reviewed, but no code is human-written.
• Minions excel at well-defined tasks like configuration adjustments and dependency upgrades.
• The system’s success is rooted in Stripe’s pre-existing infrastructure for human engineers.