Skip to content
← All AI workflows

Stripe: turn scoped coding tasks into reviewed pull requests

Stripe reports more than 1,300 agent-produced pull requests merged each week after human review. Its internal system combines isolated workspaces, repository rules and automated checks. The count does not measure a productivity gain or code quality.

From the source

Public article page from Stripe, Alistair Gray
Stripe, Alistair Gray · Captured 2026-10-14Read the original

How it works

  • The original account describes isolated developer environments and repository rules. These require infrastructure and permissions, not just a coding command-line tool.
  • Blueprints combine fixed automated checks with model-driven implementation and repair. Toolshed supplies approved tools within Stripe's internal controls.
  • Stripe reports more than 1,300 merged pull requests each week containing agent-produced code, with human review.
  • Task complexity, failed attempts, review effort, costs and defect comparisons are not disclosed. Human-reviewed does not mean free of human work.

Results

Stripe's Minions Part 2 reports more than 1,300 merged pull requests each week. Unattended coding agents produced all the code in these requests, and humans reviewed it. This is a throughput count. It does not establish a percentage productivity gain, cost reduction or defect-rate improvement. The documented architecture uses isolated developer environments, scoped rule files and a central tool server. It combines fixed automated checks with model-driven implementation. This is an internal architecture pattern, not a public ready-to-run system.

Workflow guide

  1. Select a narrowly defined issue from an approved Slack, command-line or web source. Record its linked context and acceptance criteria.

  2. Prepare an isolated developer environment. Exclude privileged personal workspaces, production secrets and unrelated task files.

  3. Provide applicable repository rules and approved tools. Restrict filesystem and network permissions to the task.

  4. Ask the agent for an implementation plan. Authorize local code changes only after a person accepts the scope.

  5. Run relevant tests and automated checks. Prepare a pull request for human review without bypassing repository security gates.

  6. Compare a fixed issue sample with the previous process. Record failed attempts, task time, correction effort, defects, fees and review time.

Sources and downloads

Discuss this workflow →

Book a call

A free 30 minute call. Tell us where you are with AI, what your IT allows, and what management needs to see. We'll tell you honestly what we would do first.