Stripe's Internal Agent Kai: Built by 1.5 Engineers in Two Weeks, Now Serves 10,000+ Employees
Summary
Sharadh Krishnamurthy (Stripe) explains that Kai's hardest problem was governance, not the model: 'projects' configure default models/tool access/human-review requirements per team, ~2,000 reusable skills exist with telemetry to flag which are stale, and prior investment in data/developer infra was what made scaling agents safe (Stripe nearly had agents overwhelm production before adding agent-specific identities, load shedding, and priority signals).
Why this signal
A named, enterprise-scale agent deployment case study matching Neo's confirmed-high-signal calibration exactly — governance-by-project and skill telemetry are concrete, adoptable patterns for however Xometry scales its own agent rollout.
Original email
Lennys Newsletter · 9/7/2026
How I AI: GPT-6 Astra is a banger + Stripe's AI playbook + Grok Bot vs. OpenClaw & Lessons from spending $20,000 on Devin