Google Cloud has unveiled Gemini Agent, a work agent that runs across Google Workspace, Microsoft 365 and Slack and routes each job to either Gemini or Anthropic’s Claude models, in private preview with general availability expected by the end of October.
Google Cloud unveiled the Gemini Agent on 8 October 2026 at its Gemini at Work event, describing it as a single agent for work that can plan and complete multi-step jobs across Google Workspace, Microsoft 365 and Slack, according to summaries of the announcement. It is in private preview, and Google expects general availability by the end of October 2026, starting in North America.
The agent is built to route jobs between different AI models. Google says that it sends each job to the model that fits best, using its own Gemini models and Anthropic’s Claude models today, with OpenAI and open-weight models to follow, according to a summary of coverage. A report in another outlet describes this as a work agent that can run for days, though its details could not be reviewed.
A distinctive feature is that “colleague agents” get an identity of their own. According to the summary, each agent has an account with an email address, a calendar, a Drive and an entry in the company directory, and acts under that identity, so it sees only what someone shares with it. People can call on the agents by writing @Gemini in Gmail, Docs, Sheets, Slides or Chat, by writing to them directly or by sharing a file with them.
That design addresses a concern in the use of AI agents at work, which is what data they can reach. An agent with its own account and permissions can be managed in the same way as an employee account, and its access can be limited and reviewed. The sources reviewed do not describe the administrative controls in detail.
The announcement came a week after OpenAI’s DevDay, where it introduced its own always-on agents called Dots. Both companies are pursuing agents that work across tools, though they differ in structure: Google’s agent works across other companies’ software and models, while OpenAI’s Dots connect to applications through its own plugin system, according to the respective reports.
The willingness to route work to Claude is the most notable detail, since Anthropic is a competitor of Google in AI models. It reflects a view that enterprises want to use more than one model, and that the agent layer, not the model, is where the value lies, a point made in the title of a Futurum analysis headlined that the agent is put ahead of the model.
The sources reviewed do not give the price of the Gemini Agent, the list of countries where it will be available after North America or any independent tests of its performance. The claims about what it can do are Google’s, and a private preview means that only selected customers have used it so far.
For companies that use Microsoft 365 as well as Workspace, the promise of a single agent across both is the main attraction. It will depend on how well the agent works with the other suite in practice, which only wider testing can show.
If general availability arrives at the end of October as planned, businesses will be able to try the agent within weeks, and the results will show how far the idea of cross-platform AI colleagues goes.
Image: Outside the Googleplex in Mountain View, California (file photo). Photo: David Nagle / Wikimedia Commons (CC BY-SA 4.0)
