What changed
Glean has made Agent Identity generally available. Instead of running every agent through the person who launched it, Glean can give each agent its own governed account in the systems it connects to. Administrators manage service credentials and Knowledge Profiles, while builders attach approved credentials to an agent. Glean also added dedicated Slack and Microsoft Teams app support for agent identities.
The same October release makes Glean's Platform Skills API generally available. Its 14 operations let integrations discover, validate, import, sync, and version skills programmatically, which gives the identity layer a more useful surface to govern: not only who an agent is, but which reusable capabilities it can carry.
Why it matters
The important product decision is separating human intent from agent execution. A user may ask an agent to prepare a support response, inspect a repository, or update a CRM record, but the downstream system should be able to tell whether the action came from a person, a named agent, or an integration acting on the agent's behalf. That distinction improves audit trails, incident response, access reviews, and the ability to revoke one agent without disabling a whole employee account.
It also changes the permission model. A shared human token tends to accumulate broad access because it has to serve many workflows. An agent identity can be scoped to one job, one audience, and one set of tools. Glean's Agent Identity documentation describes the service-credential path, and its GitLab integration guide shows the security pattern in more detail: use a dedicated service account, limit it to the necessary group and token scopes, inject credentials into outbound requests, and require approval before writes.
The product takeaway
For builders, agent identity should be designed as a first-class domain object, not added later as a logging field. Give each production agent a stable name, owner, purpose, permission envelope, credential set, approval policy, and revocation path. Record the agent identity alongside the human request that initiated a run, then make the boundary visible in the UI and audit trail.
Glean's launch is a useful signal for where enterprise agent products are heading: the differentiator is no longer only how well an agent reasons. It is whether the surrounding product can make autonomous work attributable, bounded, and reversible across the systems where the work actually happens.