How Stripe built Kai: a company-wide AI agent on Deep Agents

Summary

A case study of how Stripe built Kai, an internal company-wide AI agent available to all 5,000+ employees, to handle synthesizing data, brainstorming, drafting, and analysis across internal systems. Stripe's AI platform team layered Kai on top of Deep Agents (an open-source agent harness from LangChain), focusing their custom work on domain-specific workflows rather than reinventing foundational agent infrastructure. The platform grew from quarterly adoption targets to 83% of Stripe using it weekly within weeks, driven by serving non-engineers with tools that fit their actual work instead of asking them to learn developer tooling. Read the full case study.

Key Claims

• Stripe built Kai's initial version in one week by layering domain-specific workflows on top of Deep Agents; the time savings proved the ROI of their prior investment in Python infrastructure.
• Kai achieved 83% weekly adoption across Stripe within a month by targeting non-engineers (sellers, finance analysts, business operations) with an interface matching how they actually work, rather than requiring them to master developer tooling.
• Middleware primitives—filesystem backed by S3 for persistent context, sandboxed code execution kept separate from agent execution for security, and context summarization for long-running sessions—enabled Kai to scale reliably in production.
• Stripe runs a federated skills model where individual teams own and maintain skills (domain-specific agent modules); with 500+ internal tools and 1,000+ skills, the team avoids loading all tools upfront and instead uses two-pass selection: LLM selects relevant skills, then skills gate which tools load.
• Stripe's Kai sees higher adoption rates from non-engineering business functions (Marketing 95%, GTM 87%, Finance) than from engineering, despite being built by engineers.

Quotes

• "What if everybody had an always-on, production-ready assistant?"
• "The Deep Agents layer solves all the non-Stripey problems, so that we can focus on solving the Stripey agent problems."
• "Kai completely solidified people's belief that Deep Agents is the way to go."
• "My Stripe career is divided into before and after Kai."