← All episodes

How Stripe deploys 1,300 AI-written PRs per week

Name: How Stripe deploys 1,300 AI-written PRs per week
Uploaded: 2026-03-25T12:00:00Z
Channel: How I AI

March 25, 2026 | 12 products mentioned

How I AI host

Steve Kaliski guest

Watch on YouTube ai-assisted software development internal developer tools agentic workflows machine-to-machine payments developer experience optimization code generation at scale cloud computing infrastructure

Stripe engineer Steve Kaliski demonstrates how Stripe deploys 1,300 AI-generated pull requests per week using internal agents called "minions" that handle development tasks autonomously. The episode explores two major innovations: how minions integrate with Stripe's developer infrastructure to accelerate engineering workflows, and how machine-to-machine payments enable AI agents to purchase third-party services dynamically while completing tasks like planning a birthday party for under $6.

Key takeaways

• Activation energy for starting work drops dramatically when development prompts can be triggered from Slack, Google Docs, and Jira tickets rather than requiring engineers to open text editors.
• Cloud-based development environments are critical for running multiple parallel AI agent tasks simultaneously without overwhelming local machine hardware.
• Strong developer experience infrastructure directly enables agent capabilities—agents inherit the benefits of internal documentation, CI/CD tooling, and well-designed APIs that engineers use.
• Minions were built on top of Goose, an open-source agent harness, customized with Stripe-specific MCP servers for accessing internal tools, documentation, and test data.
• AI agents can now act as economic actors by spending money on third-party services through machine-to-machine payment protocols, enabling agents to purchase browser sessions, web searches, and mail services as needed.
• Code review remains essential even with AI-generated PRs because CI environments, test coverage, and blue-green deployments provide confidence that agent-written code is safe to deploy.

Recommendations (12)

VS Code uses

"It's going to set up a VS Code server so I could connect to it just through the web or locally"

Steve Kaliski · ▶ 8:56

Goose uses

"Boot up an instance of goose that's basically the harness that's going to run through all this"

Steve Kaliski · ▶ 11:36

MCP (Model Context Protocol) uses

"Most things that could have an MCP server have an MCP server. So we're able to interact with a lot of the internal data we have"

Steve Kaliski · ▶ 12:40

Cursor uses

"We also make Cursor and Cursor and tools like that widely available to engineers at Stripe"

Steve Kaliski · ▶ 14:14

Claude uses

"I can boot up Claude and I can have sort of the typical Claude experience with all the Stripe MCP tools available"

Steve Kaliski · ▶ 15:38

BrowserBase uses

"We've actually paid BrowserBase to create a new browser session. So, I didn't sign up for BrowserBase beforehand. I'm just paying for this one session."

Steve Kaliski · ▶ 27:36