Sep 7, 2026 · 50 min
Stripe turns organizational context into an enterprise AI advantage
Build your own company brain: the enterprise AI playbook from Stripe’s engineering team | Sharadh Krishnamurthy
Kai shows that useful enterprise agents depend as much on permissions, data infrastructure, and governance as on models.
- 1Stripe made Kai broadly useful by combining personal context, organizational knowledge, secure tools, and project-based permissions.
- 2Reusable skills turn successful agent workflows into shareable capabilities while reducing repeated setup, token, and time costs.
- 3Agent adoption requires query limits, telemetry, evaluation, and lifecycle controls to prevent automation from overwhelming core infrastructure.
Don't miss
Sharadh demonstrates Kai creating a Stripe adoption dashboard and turning the successful workflow into a reusable skill.
The brief
Sharadh Krishnamurthy explains why Stripe built Kai: making AI useful across a company required replicating organizational context, not merely adding another technical tool.
Kai combines identity, role, time, project data, and company knowledge, while projects signal intent and govern which context, permissions, tools, and models apply.
A live demonstration shows Kai creating a Stripe adoption dashboard, then packaging the workflow as a reusable skill that others can discover and invoke.
Stripe’s agent-ready foundation includes resilient query infrastructure, data catalogs, analytics layers, tool restrictions, and limits designed to contain runaway or excessive queries.
Kai spread rapidly after a first version was built by roughly one and a half people in two weeks, then improved through demonstrations, feedback, and organic adoption.
The broader lesson is operational: enterprise agents need telemetry, evaluation, editable behavior, permission boundaries, and lifecycle management alongside capable models.
Featuring
Mentioned
KaiListen to the full episode and explore every guest, topic, and moment on PodLume.