How I AI with Claire Vo

Episode: How Stripe built “minions”—AI coding agents that ship 1,300 PRs weekly from Slack reactions
Guest: Steve Kaliski, Stripe Engineer
Date: March 25, 2026

Episode Overview

This episode dives into how Stripe uses “minions,” AI-powered coding agents, to automate engineering workflows. Host Claire Vo and engineer Steve Kaliski break down how these agents handle everything from kicking off pull requests via Slack emoji reactions to orchestrating complex machine-to-machine (M2M) payments—all with minimal human intervention. The conversation covers the practical realities of deploying AI in engineering organizations, infrastructural investments that pay off, and hints at the broader future of agentic software.

Key Discussion Points & Insights

1. What Are Stripe’s “Minions”?

• Definition and Purpose:
  • Minions are AI coding agents living inside Stripe’s codebase and dev environments.
  • They automate coding tasks based on prompts that often originate from natural places like Google Docs, JIRA, or Slack.
  • Activated simply by Slack emoji reactions, they spin up isolated cloud development environments, make code changes, and submit PRs for human review.
• Impact:
  • Stripe lands ~1,300 minion-initiated PRs per week.
  • Reduces the “activation energy” required to start (and sometimes finish) engineering work.

"At Stripe, we're landing about 1300 PRs that have no human assistance besides review per week."
— Steve Kaliski [00:00]

2. Lowering Friction and Increasing Velocity

• Friction in Orgs:
  • Large companies accumulate operational and communication friction that slows innovation.
  • Minions help cut the friction by making it easy for anyone to kick off work—no deep technical expertise required.

"When you're in larger organizations, there's so much friction that can come between a good idea and getting it into the world."
— Claire Vo [00:19]

• Developer Experience (DX) as an Enabler:
  • Investments in developer tooling (cloud environments, great docs, CI) empower both humans and AI agents.
  • Good developer experience makes both human and agent engineers more effective.

"What's good for the developer is good for the agent."
— Claire Vo [09:40]

3. How Minions Are Activated and Work in Practice

• Slack-Based Activation:
  • Engineers (or anyone in the org) can trigger a minion from Slack by reacting to a message.
  • The minion provisions a dev environment seeded with the relevant prompt, accesses internal tools/docs, and attempts a “one-shot” resolution.

"All I have to do now is add a reaction...and then we get a reply in here saying your minion for pay server—it's the repository—a new branch...has been created."
— Steve Kaliski [05:51]

• Parallelism and Cloud:
  • Multiple minions can run in parallel in isolated, fully-provisioned cloud environments.
  • This is crucial for scaling and keeping engineers’ local machines unburdened.

"No matter how juiced these laptops are, you get three or four work trees in, and it starts to sound like an airplane taking off."
— Claire Vo [00:51 / 17:45]

4. The Importance of Developer Productivity Investments

• Dedicated Teams:
  • Stripe has a longstanding developer productivity team focused on tools, CI, environments, and internal DX as a product.

"That team cares equivalently about engineers at Stripe being successful and being able to build things quickly."
— Steve Kaliski [16:29]

• Human Review Remains Critical:
  • All agent-initiated PRs are subject to the same CI and human review pipelines.
  • Strong test coverage and deployment safety mechanisms (e.g., blue-green deployments) are essential.

"Whether the text has been written by Steve or Steve's robot, you still want that CI environment that's providing confidence that the code that's being changed is safe."
— Steve Kaliski [00:35 / 22:18]

5. Extending Minion Usage Beyond Engineering

• Company-Wide Enablement:
  • The low-barrier Slack interface means PMs, designers, and others can initiate minion work with little technical knowledge.

"For whether you just want like a proof of concept or you're going to make a docs change...you can probably write out in plain text the thing you want to occur."
— Steve Kaliski [24:00]

6. Agents as Economic Actors: Machine-to-Machine Payments

• Demo: Planning a Birthday Party with AI (“Claude Code”)
  • Steve demonstrates using an agent to plan a PM’s birthday party, end-to-end, including venue, invitations, and offsetting carbon emissions.
  • The agent makes real payments to third-party services (for browser sessions, venue info, physical mail) via Stripe’s machine payment protocol.

"We imagine a future where third-party services are going to want to sell into these kinds of experiences and that those interactions will cost money. So we have to equip our agents with the capacity to spend..."
— Steve Kaliski [25:02]

• Costs and Economic Model:
  • Task completed for about $5.47 including an offset for carbon emissions.
  • Illustrates the shifting economy where agents transact not just in tokens (for API/compute) but in dollars (for third-party purchases).

"It feels very natural over time to see the token and the dollar side by side...I planned a birthday party for $5.47. That doesn't seem too bad."
— Steve Kaliski [34:10]

• Impact:
  • Lays the groundwork for agent-to-agent commerce and new API-first microservice businesses.

"It’ll be really interesting to build a business where your primary consumer sort of wants an ephemeral interaction, and instead you could focus on just a hyper useful single API and monetize that directly and make your audience primarily agents."
— Steve Kaliski [35:48]

Notable Quotes & Memorable Moments

• On Activation Energy:

  "A lot of where our work begins...could be in a Google Doc...or we're talking about something in Slack. And then you end up in a text editor when it's time to actually do the work or make the final tweak. And it’s just felt very natural."
  — Steve Kaliski [03:21]

• On Prompt Writing:

  "I love your system prompt. So sophisticated. It says implement this task completely: and then just whatever you put in, no mistakes."
  — Claire Vo [11:40]

• On AI Assistance for Non-Devs:

  "Being able to just click an emoji or tag the robot...we’re trying to see more non engineer usage there."
  — Steve Kaliski [24:00]

• On agent-initiated economic activity:

  "I told [Claude] to research Gen Lee, find a place to have the birthday, send invites, and then offset the energy usage with Stripe Climate."
  — Steve Kaliski [25:02]

• On being polite to AIs:

  “This sounds crazy. Like I have made a concerted effort to always be polite...I don't want to, like, I'd rather have to do a little bit extra work than have it on the record that I was mean—because you never know.”
  — Steve Kaliski [39:07]

Important Timestamps

• [00:00] – Stripe’s weekly 1,300+ minion PRs and frictionless code activation (Slack, docs, tickets)
• [03:21] – How minions lower personal and organizational barriers to starting work
• [05:51] – Live demo: Clicking a Slack emoji triggers a minion to update Stripe docs
• [09:40] – “What’s good for developers is good for agents” and value of DX
• [13:42] – Why Stripe builds on “Goose” (open source agent harness)
• [16:29] – Stripe’s developer productivity team and its role
• [18:58] – Advocating for cloud/virtual dev environments over local machines
• [22:18] – All minion code flows through robust human and CI review
• [24:00] – Slack interface lowers technical bar for engaging minions company-wide
• [25:02] – Demo: Planning and paying for a birthday party start-to-finish through agents
• [32:16] – Recap: Agent pays for access, services, and tokens—economic model emerging
• [34:10] – Reflections on the economics of token/dollar costs in agentic automation
• [35:48] – Vision: New businesses targeting agent-consumers, pure-API interfaces
• [39:07] – Steve’s approach to prompting: politeness, iteration, and guidance (“dad teaching to ride the bike”)

Tone & Final Takeaways

The tone is practical, lightly humorous, and “demystifying”—making advanced AI workflows approachable for a general audience. The conversation stays grounded in real engineering and organizational contexts. Both host and guest stress that investments in developer experience benefit both humans and AI agents, and hint at a future where agents transact and build “all the way down,” opening new business and productivity paradigms.

Final advice from the episode:

• Prioritize developer experience and cloud-based dev environments.
• Be ready for a world where AIs and humans collaborate and transact fluidly via APIs and agent-driven workflows.
• Approach prompting and agent interaction with clarity—and maybe a little bit of politeness, just in case.

Where to Learn More

• Stripe’s AI/Minion Docs and Blog: swepe.dev, docs.swepe.com/payments/machine
• Follow Steve: @stevekaliski on Twitter
• How I AI: howiapod.com (all episodes).