Gemini gets its own email address: Google bets on the identity layer

Google’s new Gemini agent for business gets its own Google Workspace account. That means its own email address, calendar, Drive and entry in the company directory, and every action it takes is logged against the agent rather than a person. It also routes work to Anthropic’s Claude when that suits the task better. For IT and ops teams, the practical consequence is that you now have to provision, audit and eventually offboard an AI agent the way you would a member of staff. Google has said very little so far about the offboarding part.

What Google announced

At a Google Cloud event on Thursday 8 October, Google launched a unified Gemini agent that plans and executes work from a single interface. Google Cloud CEO Thomas Kurian described it as an agent you give “objectives, not just instructions”. It plans the work, uses custom skills and tools, and connects to internal systems, with connectors for Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake. It can also work with any MCP server, whether inside or outside the company network. Users follow its progress in a “tasks inbox”, which shows its reasoning, the sub-tasks it has handed to subagents, the skills it has loaded and the code it has written.

The identity piece is what stands out. Each agent knows who sits on which team, their time zones, who approves what and what is on people’s calendars. You reach it by tagging it, emailing it, sharing a file with it or adding it to a group chat. According to The Next Web’s analysis of the governance model, each coworker agent gets a cryptographically attested identity that is stamped into its logs and into any virtual machine it starts. Administrators approve role-based permissions, and all traffic goes through what Google calls an AI network firewall. Long-running jobs carry on after the user’s laptop is closed.

Google is launching for businesses first, with consumers to follow. CNBC reports that versions for financial services and legal work are in preview, with government, healthcare and retail versions coming soon.

The tell: Google routes to Claude

By default the agent chooses which model to use for each task. Today that means Google’s Gemini models and Anthropic’s Claude models, and Google says open-source and private models will be added to the picker later. Google putting a competitor’s models inside its own flagship agent is not a sign of weakness. It shows where Google thinks the lasting advantage sits. If models can be swapped behind a router, the hard thing to replace is the layer above them: the directory entry, the permissions, the calendar context and the audit trail.

That changes how you should think about switching costs. The question is no longer whether another vendor’s model scores better. It becomes whether you are prepared to unpick an agent from your org chart, approval chains and access-control reviews. Once an AI shows up in your directory, it gets treated like an employee and not like a tool.

Why it matters for Irish, UK and EU businesses

The launch coincided with regulatory pressure in the UK. Hours before the announcement, the UK Information Commissioner’s Office named Google among ten developers it had secured data protection changes from. It also opened a call for evidence on agent risks, with responses due on 20 November. Richard Nevinson, the ICO’s director of technology regulation, said agents acting with autonomy “is not an excuse for poor compliance”.

Giving each agent its own logged identity does help answer the accountability question regulators keep asking. It is much easier to demonstrate control over an agent you can find in the directory than over an agent running under someone’s personal OAuth token. The gaps are real, though. Google’s post names Deutsche Telekom as a European data sovereignty customer, but it gives no regions, no pricing and no general availability date. The European Commission’s proposed Cloud and AI Development Act would also place sensitive public workloads behind EU ownership requirements that Google Cloud cannot currently meet.

The largest gap is offboarding. Google explains in detail how an administrator grants an agent permissions, maps its identity through OAuth into external systems and caps its spending. It does not explain how you take any of that away. That matters when, as The Next Web reports, Gartner expects 40% of enterprises to pull autonomous agents back by 2027 because of governance gaps.

What to do now

Don’t wait for general availability before working out your position. Agents with identities will arrive through Google, Microsoft or whichever SaaS tools you already pay for. Here is what we would put in place first:

1. Treat agent identities as joiners, movers and leavers. Each agent needs a named human owner, a documented scope and a revocation procedure that you have actually tested. That includes the OAuth grants it holds in Jira, Slack or Snowflake.
2. Start with read-heavy, low-blast-radius work. Reporting, triage and drafting are good first jobs. Put anything that sends, pays or approves behind explicit human-in-the-loop checkpoints.
3. Map data flows before connecting systems. Multi-model routing means a task could be handled by Claude and not by Gemini. Your DPIA should cover that. Our guide to data security and privacy in AI projects is a good place to start.
4. Keep your context portable. Skills, prompts and process definitions should live somewhere you control, so that switching vendors later is a migration and not a rebuild.

If your core systems are older than Workspace, the connectors are only part of the job. Our notes on integrating LLMs into legacy enterprise systems cover the parts no vendor demo will show you.

Frequently Asked Questions

Does the Gemini agent only use Google's models?

No. By default it chooses the best model for each task. That currently means Google's Gemini models and Anthropic's Claude models. Users can also select a model manually, and Google says open-source and private models will be added later.

How are the agent's actions audited?

Each coworker agent has its own Workspace identity, and its actions are written to an audit trail attributed to the agent, not to a person. The Next Web reports that the identity is cryptographically attested and stamped into the agent's logs and any virtual machine it starts.

Is it available in the EU yet?

Google has not published regions, pricing or a general availability date. Its announcement names Deutsche Telekom as a European data sovereignty customer, but European buyers should check data residency before committing.