Stripe's Knowledge AI Platform treats agents as shared infrastructure

Enterprise agents seem to need shared execution infrastructure, domain-owned skills, and task-specific access controls, not just many one-off assistants.

Logged at IST: 2026-07-31 10:35 IST

What it is: Stripe engineering introducing Kai, its internal Knowledge AI Platform for non-coding knowledge work across sales, finance, support, compliance, operations, and engineering.

Gist: Stripe says coding agents worked well for software because the workflow shape is relatively uniform: edit files, run tests, commit. Knowledge work was messier. Teams had built more than 4,000 no-code micro-agents, but those became hard to monitor and maintain, while coding agents created security and support issues for non-engineers.

Kai is Stripe's answer: a shared agent platform with surface-agnostic APIs, AgentStudio for domain-owned skills and governance, and a secure execution environment. The execution layer runs on Kubernetes with a per-session sandbox, a multi-tenant virtual filesystem, sandboxed code execution for analytics, long-session task management, and access-control rules that consider the task context, not only what the user token can access.

The adoption numbers are the hook: Stripe reports 83% weekly active usage, 1,000+ skills and tools, sessions as long as 932 turns, more than 5,000 daily data-analysis sessions, and GTM impact metrics including 2x sales activity and 39% more closed deals for account executives in weeks they used Kai.

Newsletter angle: Strong agents/platform-engineering item because it frames enterprise agents as shared infrastructure plus domain governance, not a pile of one-off assistants. The most interesting bit is the access-control boundary: “what should this task be allowed to view given this context?” is a harder and more useful question than “what can this user access?”