Descriptions:
Stripe software engineer Steve Khaliski joins the How I AI podcast to walk through how Stripe has built and deployed a fleet of AI coding agents — internally called ‘minions’ — now contributing approximately 1,300 pull requests per week with no human involvement beyond code review. Khaliski demonstrates the end-to-end workflow live: triggering an agent via a Slack emoji reaction or Jira ticket, running multiple agents in parallel across isolated cloud environments, and feeding results through Stripe’s existing CI/CD pipeline with blue-green deployment and rollback capabilities. The system is built on top of Stripe’s developer productivity infrastructure — the same team that manages git workflows, editor tooling, and development environments, now increasingly focused on AI-native developer experience.
A key insight throughout the interview: the activation energy for starting engineering work has dropped to near zero. Khaliski describes rarely opening a text editor to begin a task anymore, instead kicking off agents from Slack on the subway commute and reviewing output by the time he arrives at the office. The video also includes a live Claude Code demo spending just over $5 to fully plan a birthday party as a proxy demonstration of complex multi-step agentic planning.
Khaliski’s most actionable guidance targets engineering leadership: investing in cloud-based and virtual development environments is the single highest-leverage move for organizations wanting to scale AI coding agents in 2026. Running multiple work trees locally — even on high-end hardware — creates hard CPU and memory ceilings. For CTOs and VPs of Engineering, he argues this infrastructure investment unlocks parallelism that no amount of model capability can compensate for without.
📺 Source: How I AI · Published March 25, 2026
🏷️ Format: Workflow Case Study
