Stripe's Kai enables context-specific safety by applying tool policies at the "Project" level. For an HR project with sensitive data, creating a public document can trigger human approval, while other, less sensitive projects run without this friction.
Deploying AI agents will mercilessly find and exploit your system's weaknesses. Agents "dial up all your failure modes," revealing that investments in infrastructure resilience, load shedding, and monitoring are critical prerequisites for safe agent deployment at scale.
Stripe’s Kai uses "Projects" as a powerful governance layer. Instead of just organizing chats, Projects define context, set spending limits by restricting model choice, and enforce tool usage policies. This allows teams to create safe, pre-configured environments for specific workflows.
Companies with strong, pre-existing developer platforms, data infrastructure, and analytics layers see the highest returns from AI agents. Foundational investments that made humans efficient provide the necessary leverage for AI to operate effectively and safely at scale.
To make data agents effective, Stripe implements a tiered access strategy. The agent first searches existing reports, then a curated analytics layer, and only queries the full data catalog as a last resort. This triage prevents inefficient "brute force" queries and improves answer reliability.
Stripe's internal agent, Kai, started as a V0 built by just 1.5 people in two weeks. This rapid MVP was critical because it made the agent's value tangible. Showing a working prototype was far more effective for gaining organizational buy-in than trying to explain the concept abstractly.
As an internal AI agent's skill library grows, quality and discoverability become major challenges. Stripe actively manages its 2,000+ skills with automated improvement suggestions and usage telemetry to prune unused skills, preventing context bloat and maintaining performance.
The primary challenge in building Stripe's internal AI, Kai, wasn't the technology, but creating governance structures. This ensures employees across a complex, global business can use AI safely and know it will "do the right thing," making governance the true product.
Stripe's Kai succeeds by being a skill-building platform, not just a tool. Users can package successful interactions into reusable workflows ("skills") to share across the company. This democratizes automation and transforms the agent into a system for codifying institutional knowledge.