Mathew PaquetChecklist · 4 min read
The AI Coworker First-Week Checklist
Before an AI agent gets its own work email, decide who manages it, what it sees, what it can send and spend, and how you switch it off.
On October 8, 2026, Google announced the Gemini agent. It works two ways: as a personal assistant that works for you, and as a coworker agent that works with your team.
You describe the role you need, say an events coordinator, and Gemini creates it. Google says it gets its own Workspace account: an email address, a calendar, a Drive, and a spot in the company directory. People add it to a chat space or tag it in a doc, and its edits show up under its own name in the version history.
It's a limited preview. No price yet, and no date for everyone else. So this is about what you decide before you turn it on.
Why the account matters
Most AI assistants today work inside your own account. It reads your inbox and drafts your replies, so everything it does, it does as you. A coworker agent is a separate identity. Google says it sees only what you share with it, and every action it takes goes into an audit trail attributed to the agent, not to a person. You can finally tell what you did from what your AI did.
The gap. A log tells you what happened. It doesn't tell you who answers for it. Google's launch post explains how an administrator grants an agent its permissions. It doesn't describe who owns the agent, or how you take that access away. Microsoft, building the same kind of agent identity, requires at least one sponsor per agent: a named person accountable for what the agent is for and whether it should still exist.
Fill this in before you create the agent. If a line stays blank, you're not ready.
Manager (name on it): [NAME]
Backup when the manager is away: [NAME]
It can see: [ONE FOLDER] and [ONE CHAT SPACE]
Email outside the company: waits for [NAME] to read it
Spending cap alert goes to: [NAME]
Who can switch it off: [NAME OR ROLE]
Switch-off tested on: [DATE]
Picture a twelve-person marketing agency that creates that events coordinator. Each day hides one decision.
Two of these are controls Google showed on stage: a rule that any email going outside the company waits for a human to review it, and budget caps. Google says you set a hard limit on a project's AI spend in the Cloud Billing Console. When it's hit, that project's agent pauses, and someone resumes it with one click in the console.
On Friday, expect something to be missing. On Google's own stage, the agent's launch readiness document missed some recent meeting updates. The presenter left it a comment, the way you would with a new hire, and it fixed it.
The last decision is the one Google didn't write down. Say the agent's manager leaves. The agent doesn't. It still has an email address, a calendar full of threads, and access to every folder anyone ever shared with it.
Then actually try it once. Write the date on the agent card.
Regulators are asking the same thing. The same day Google announced this, the UK's data protection regulator opened a call for evidence on AI agents and asked companies using them to respond, not just the ones building them. Its director of technology regulation said the fact that AI agents act with autonomy is not an excuse for poor compliance.
The product does the logging. The accountability is still yours.Sources
Mathew's LibraryLibrary card
Keep reading free with your library card
One email opens this checklist and all 3 resources in Mathew's library. No account, no password.
Write down who would manage your first AI coworker, who backs them up, and who is allowed to switch it off.
Saving works best in your phone's browser. A copy is also in your inbox.