# Building an Enterprise AI Agent: Stripe's Internal AI Playbook How Stripe scaled AI adoption safely with their internal agent Kai - **Source:** [Lenny's Newsletter](https://www.lennysnewsletter.com) - **Published:** 2026-09-07 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/fc2ef888-0f7d-48a4-af48-99ee9ca1b059 ## The Kai AI Agent Stripe built Kai as an internal AI assistant used by over 10,000 employees weekly. Unlike off-the-shelf solutions, Kai was custom-developed to handle Stripe's specific needs at enterprise scale. The system integrates with Stripe's existing data infrastructure while maintaining strict governance controls. Engineering manager Sharadh Krishnamurthy led the development, drawing from his experience across Stripe's core infrastructure teams. ## Why Build Instead of Buy Stripe chose to develop Kai internally rather than purchase commercial AI tools due to unique requirements around security, scale, and integration. Existing solutions couldn't meet Stripe's needs for agentic workflows across their specific tech stack. The decision was driven by the need for deep integration with Stripe's data layer and developer tools, which already contained necessary governance structures. ## Data Governance Framework Kai operates within a carefully designed data access framework where 'projects' serve as governance boundaries, not just organizational folders. This structure ensures agents only access permitted data. Stripe leveraged its existing infrastructure built for human developers, which serendipitously provided the perfect foundation for AI agent operations at scale. ## Skills Platform Architecture Stripe created a skills platform allowing any employee to package workflows as reusable AI capabilities. The system now hosts over 2,000 skills. Each skill operates in a secure sandbox with telemetry for monitoring. This democratizes AI while maintaining control, enabling non-technical staff to safely create and share automations. ## Production Challenges Early versions of Kai nearly took down production systems when agents behaved unexpectedly. These incidents led to implementing safeguards like agentic identity tracking and load shedding mechanisms. The team learned critical lessons about monitoring AI systems in production environments through these challenges. ## Rollout Strategy Stripe phased Kai's introduction carefully, starting with controlled pilot groups before company-wide deployment. The gradual rollout allowed for iterative improvements based on real usage patterns. Adoption was driven by demonstrating concrete productivity gains while addressing security concerns through transparent controls. ## Bespoke Agent Development Beyond the general Kai assistant, Stripe developed specialized agents for particular use cases. These bespoke agents handle domain-specific workflows with higher precision. The architecture supports both broad and narrow AI applications, allowing teams to create custom solutions while benefiting from shared infrastructure. ## Enterprise AI Blueprint Stripe's experience offers a blueprint for large organizations implementing AI. Key lessons include the importance of governance layers, the value of existing infrastructure, and the need for both broad and specialized agents. Their approach balances innovation with responsibility, demonstrating how to scale AI safely across thousands of employees. ## Sources covering this story - https://www.lennysnewsletter.com/p/build-your-own-company-brain-the — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._