Stripe's Kai shows how to build AI agents that mirror your org's DNA

Sep 7, 2026 · Lenny's Podcast

🎧 PodShort 49 min squeezed to 2 AI / ML New

Full episode from Lenny's Podcast

Quotable Moments

Agents are very creative at bringing your infra down. It turns out that agents just like dial up all your failure modes.

The harder problems are in trying to replicate the way a company works at scale. And Stripe is an incredibly complex business.

The one cool things about projects that can be very concrete for people is the idea of tool policies.

Key Insights

Building a custom AI agent like Kai is driven by the need to replicate and scale a company's internal processes and knowledge, not just to provide AI capabilities.
Kai is designed to be context-aware, understanding who the user is and their role within the company's organizational structure, which allows for more personalized and effective AI interactions.
Projects in Kai serve as a governance mechanism, allowing users to define specific tasks, appropriate tools, and safety controls for AI usage, which is crucial for enterprise-scale AI.
Stripe invested heavily in foundational infrastructure like developer experience, data platforms, and analytics layers, which significantly de-risked and accelerated their AI agent development.
The tri-layered approach for AI agent interaction—starting with existing reports, then using analytics, and finally falling back to raw data—provides a smart and efficient way for agents to retrieve information.
Agents, when not performing as expected, can sometimes default to brute-force methods, highlighting the importance of robust error handling and user-defined guardrails.
The key to successful AI adoption in enterprise is not just providing the AI, but establishing strong governance and understanding how the AI fits into existing company workflows.
A multi-turn conversational approach for AI agents allows for more complex task completion and deeper collaboration between the user and the AI.

Metrics Mentioned

28% X growth in AI tool spend over the last year. (Across over 500 engineering organizations.)
86%+ of Stripe users use Kai. (Indicates strong internal adoption and reliance on the AI agent.)
Kai usage covers 70-80% of user workflows. (Shows the broad impact and integration of Kai across various tasks.)

RevBots.ai View:

ARM orgs should study Kai's layered approach: reports → analytics → raw data retrieval.
SaaS Hoarders take note: 28% growth in AI spend proves bolt-ons don't scale.
Project governance in Kai is a blueprint for AI Sprinklers to add real controls.
Tab Hoppers see the gap: Stripe's prior infra work enabled rapid AI deployment.