Stripe Unveils Kai, Its Internal AI Agent and Governance Structure

Key Takeaways

1Stripe uses Kai, an internal AI agent, every week with over 10,000 employees

2Governance is based on projects, a secure sandbox, and assessments

3The skills platform allows any employee to create flows, with up to 2,000 flows mentioned

4The developer-focused infrastructure has been repurposed for the agents

Why it matters : Stripe's example highlights the concrete implementation of a large-scale AI agent, with safeguards and an architecture tailored to the enterprise.

Stripe opens the black box of Kai, its internal AI agent. Between governance, a secure sandbox, and a skills platform open to employees, the company describes what it has built internally, why it did so, and how far the system is used in practice.

Safeguards, Incidents Averted, and Project Governance

Lessons have been learned when it nearly happened that agents disrupted production systems. At Stripe, projects serve as a governance mechanism and are not just records. The system relies on tools, skills, and a secure sandbox. The quality of skills, assessments, and telemetry are processed to frame usage. The topic also addresses agent identity, load offloading, and the management of unwanted agents. The functioning of projects is detailed to explain this framework.

Internal Adoption and Scaling of Workflows

More than 10,000 employees use Kai each week. The skills platform allows each employee to generate a workflow, and the situation where 2,000 workflows are recorded is mentioned. The integration of Kai within the team is explained. A live demonstration shows Kai building a dashboard, followed by an iteration on that dashboard.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Choice to Build Internally and Data Architecture

Stripe built Kai from scratch instead of purchasing tools. The infrastructure designed for human developers turned out to meet the needs of agents. The structuring of the data layer is described to enable secure queries at scale. The content discusses what Kai knows about a default user and what can be controlled or disabled. Custom agents for specific use cases are presented, along with the workflow for building skills.

Mentioned Technologies and Contacts

Several tools are mentioned: Trino, Anthropic, Gemini, and Cursor. Sharadh Krishnamurthy can be contacted on LinkedIn via the profile sharadhk. Claire Vo is associated with ChatPRD, a personal website, a LinkedIn profile, and an X account.