Garrett Galow is a product leader at WorkOS developing the identity infrastructure and internal tools that make AI agents useful and governable inside companies. His work addresses two connected enterprise problems: giving software reliable access to business information and keeping that access accountable to existing security policies.
Galow studied electrical and computer engineering at the University of Texas at Austin and participated in its Longhorn Startup Program. He worked on Microsoft Azure before joining Cloudflare, where he became a director of product responsible for areas including APIs, user services, trust and safety, and customer growth. His Cloudflare product writing covered dashboard single sign-on, multi-user access, infrastructure automation, OpenAPI schemas, and scoped API tokens that limit automated credentials to particular resources and permissions.
At WorkOS, he has advanced that concern with controlled access into agent identity and internal AI applications:
- Agent registration through auth.md. His auth.md proposal defines how agents can discover a service’s registration requirements and establish whom they represent. Signed identity assertions or user-confirmed registration provide alternatives to improvised API-key handoffs while preserving credential scoping and revocation.
- Enterprise single sign-on for AI agents. Galow treats MCP’s repeated OAuth consent screens as both an administrative burden and a security risk: disconnected tokens can remain usable after an enterprise session ends. His Cross-App Access demonstration shows Claude Code obtaining Figma access through an existing Okta relationship using identity assertions and token exchange. He distinguishes this authentication mechanism from fine-grained authorization, which requires additional policy and scope controls.
- Self-service business intelligence with Studio. The internal Studio workspace connects a LangGraph-based agent to Snowflake, Linear, and Notion so employees can answer operational questions and create shared tools without repeatedly requesting custom SQL. Its safeguards include just-in-time schema guidance, primary-source grounding, and validation that generated queries return meaningful results. Studio’s reusable data widgets become executable JavaScript interfaces that refresh directly from connected services, avoiding a new model invocation for every update.
Across these systems, Galow’s central product question is practical: how can an agent acquire enough authority to do useful work without escaping the identity, permissions, and verification mechanisms organizations already depend on?
