Descriptions:
Paola Estefania of Better Auth leads an interactive workshop at AI Engineer tackling a fundamental but underappreciated problem: most AI agents today operate by borrowing the user’s own credentials, effectively granting them CEO-level access to email, calendars, and connected services. The session argues for a paradigm shift — agents should be treated as first-class principals with their own identity, scoped authority, and cryptographic keys rather than invisible proxies hiding behind human accounts.
The workshop walks through three interconnected challenges: discovery (how agents learn which tools and APIs are available, ideally through a directory rather than manual wiring), authorization (granting fine-grained capability-level permissions — read-only Gmail access rather than blanket account control), and identity (issuing each agent its own private key so audit logs can distinguish between ‘user sent email’ and ‘Cursor agent sent email on behalf of user’). The private key approach enables clean revocation: if an agent misbehaves, its credentials can be invalidated without touching the underlying user account.
Using an employment analogy throughout — you give a new hire their own badge and role-specific access, not the CEO’s login — the session makes the security case accessible to developers who may not have considered the attack surface of delegated agent permissions. Better Auth’s implementation draws on OAuth-style token minting extended to the agent layer, with the explicit goal of making agents auditable, revocable, and independently attributable actors rather than extensions of human identity.
📺 Source: AI Engineer · Published July 21, 2026
🏷️ Format: Hands On Build
