How does Google separate temporary agents from persistent ones?

Google's Gemini agent announcement (Google Cloud, 2026-10-08) describes two distinct tiers of delegated agent under one orchestrator. The first is disposable: "it can dynamically create a roster of sub-agents — temporary, job-specific agents, each with their own identity — to tackle multi-step tasks," coordinating "parallel and sequential steps that can run for hours or days." The second is durable: Gemini "can act as a coworker agent — which is more like a team member, with a persistent, defined role and operational presence across days, sessions, and multiple changing responsibilities... Coworker agents have dedicated identities including their own @agents.company.com emails, their own persistent storage, and only have access to the context that you or your team members provide."

The distinction Google draws is about lifespan and scope, not capability: a sub-agent exists for one task and disappears, while a coworker agent is given an organizational identity — a mailbox, storage, a role such as "a project manager within a team" or "an analyst in your finance department" — that persists independently of any single task. The post pairs this with a deliberate restriction: a coworker agent's context access is limited to what people explicitly share with it, rather than inheriting everything the orchestrating Gemini agent already knows.

This pattern extends the role-based framing in Do humans and AI divide labor differently in knowledge work?, which found collaboration modes structured by who picks the task and who decides the method; Google's coworker agent is pitched as occupying a fixed organizational role rather than shifting between modes session to session, though the announcement does not say who — the agent or its human teammates — decides what it works on day to day. It also sits beside Can orchestration layers make coding agents more auditable?, which wraps a fixed executor in a layer built specifically for auditability and recovery; Google's coworker agents claim an equivalent persistent organizational presence without describing any comparable audit trail, oversight mechanism, or recovery path for a job-specific identity that holds its own email and storage.

The excerpt does not establish how the "only has access to the context that you or your team members provide" restriction is enforced or revoked, nor does it address the security exposure of giving an agent a durable company mailbox and storage of its own. It also makes no claim about how well a coworker agent performs the colleague-like work it is positioned for — a gap worth weighing against Why do AI agents fail at workplace social interaction?, where social interaction with colleagues was independently found to be one of the hardest failure modes for today's agents. Until there is independent measurement, the "team member" framing should be read as Google's stated design intent for Gemini, not as evidence that coworker agents work as colleagues in practice.