AI Agents as Product Front Door
Reed McGinley-Stempel, CEO of Stytch, on authentication for AI agents
This is really a distribution shift, not just an auth feature request. When answers move from a company website into Claude or ChatGPT, the old model of making users come to a branded UI breaks down. That forces content rich and workflow rich companies to expose their product through agent friendly interfaces, then add OAuth, consent screens, scopes, and audit trails so an agent can read or act without getting full unsupervised access.
-
The Stack Overflow example matters because it shows what happens when AI becomes the front door. External data points and academic work both show meaningful traffic and activity declines as developers shifted simple question answering into chat products instead of visiting the site directly.
-
For most SaaS companies, this is new plumbing. Stytch describes the core job as turning an app into an OAuth 2.0 identity provider for remote MCP servers, so users can grant an agent narrow permissions like read docs, create a ticket, or update a record, and revoke that access later.
-
The competitive angle is that identity vendors are converging on the same wedge from different starting points. Clerk makes signup and profile flows easy for modern app teams. WorkOS grew from enterprise SSO and directory sync. Stytch is trying to own the broader layer that combines login, delegation, fraud controls, and agent detection.
The next step is that every important software surface becomes dual mode, one path for humans in the app, one path for agents through MCP and OAuth. As that happens, identity moves closer to the product roadmap and the CEO agenda, because distribution inside AI clients will matter as much as search ranking and web conversion once did.