▶ Watch ↗18:33
Conference affiliation: WorkOS
Isabelle Williams works on identity and access management, including how applications admit AI agents acting on behalf of people. Her work spans enterprise identity governance at Microsoft Entra ID, API authorization at MongoDB and agent access at WorkOS, where she was a product manager in October 2026.
Before WorkOS, Isabelle led MongoDB’s identity and access management platform and launched its OAuth authorization server, which powered application and agent access to MongoDB’s API and MCP server. Earlier, she worked on enterprise identity governance at Microsoft Entra ID.
Isabelle approaches agent authentication partly as a product onboarding problem. If people delegate product discovery and setup to personal agents, an application that requires a browser signup or a manually obtained API key can interrupt that work before the person sees its value. Her Waypoint demonstration, a travel-planning application she built with Cursor, explores how an agent can get started, connect its work to a customer and operate within permissions the application enforces.
She uses auth.md, an open protocol with an application-hosted Markdown file describing registration and API access. The file helps agents discover how to use a service; the application must still authorize requests. Isabelle demonstrates two ways to connect that access to a person:
In Waypoint, Isabelle demonstrates the second approach with an agent that creates an account, prepares a shareable itinerary and books dinner after receiving identity-sharing and booking consent. She advocates open standards for these interactions so agents and services can work together across providers.
For corporate travel, Isabelle gives agents separate tokens rather than handing them a user’s credentials. Each agent’s permissions are the intersection of the user’s permissions and a predefined agent blueprint’s ceiling. That distinction lets the application attribute actions to an agent and revoke its access independently of the user’s session.
Her funding-agent example makes the limit visible. Asked to fix a traveler’s $250 shortfall, the agent reallocates the remaining $70 but cannot increase the overall budget. The user retains that permission and raises the budget before the agent completes the task. The example explains a specific division of responsibility: the agent can redistribute approved funds, while changing how much the company may spend remains a human-authorized action.
▶ Watch ↗18:33
Loading more talks and highlights…