Who controls the keys: setting credential limits for AI agents that access mail, calendar and browser

Connecting a personal AI agent to email, a calendar or a browser means deciding where your identity boundary sits and how far the agent can act within it. There are two basic models: delegated identity, a scoped and revocable grant issued by the provider, and shared identity, where your own password or session is handed over. Shared identity is the default in many young products, yet revoking it can mean changing your password and logging out everywhere. The author argues products should clearly label each connection's credential type, scope and revocation cost. Actions also fall into different risk classes, from read-only access to reversible and irreversible writes, and the irreversible ones should require human confirmation.