**Google Cloud** has unveiled ‘Gemini agents’ that can be used for Gmail, Docs, Slack, and more within 3 days (local time). These agents can run not only on Google’s Gemini, but also on the Claude models from **Anthropic**. Google said, “About 80% of customers are already using our AI products.”
Key points
Process multi-step work inside the WorkSpace app and call it from Slack and Microsoft 365 as well
Automatically select the most suitable model between Gemini and Claude for each task, with additional models planned for future support
Google Cloud has not yet disclosed its pricing policy or official launch schedule
Gemini agent capabilities
In a blog post, CEO **Thomas Kurian** introduced new products based on the keynote address theme of “Gemini at Work 2026.” He described the agent as a “single agent” that can “answer questions, handle knowledge work, generate images and media, and write and execute code.” The explanation is that users only need to “set objectives,” not provide detailed instructions.
Inside Workspace, you can call the agent directly from Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar. In-house personnel can access the same agent in Slack, Microsoft 365 product suite, and via a command-line interface (CLI) as well. The agent runs in the cloud, maintains a single memory across devices, and splits long-running tasks into multiple temporary “sub-agents” for parallel processing over hours to days.
In a separate “co-worker mode,” the agent is granted its own workspace account, email address, and calendar. However, the information the agent can access is limited to the scope that the team allows for sharing. This agent also integrates with external enterprise systems such as Salesforce, ServiceNow, and Snowflake. It is currently in a private preview stage, and it is reported that a gradual rollout is planned for selected business and enterprise workspace customers.
Related article: What is the annual salary of Anthropic CEO? The figure disclosed in the IPO filings
What Curian emphasized was “control” and “cost”
Kurian tailored the entire presentation to “governance (control)” and “cost efficiency.”
He wrote, “Whether the enterprise agent program succeeds or is discontinued comes down to two questions—how well the company can control it, and whether it can afford the costs.”
The Gemini agent is assigned an independent ID per instance, and all actions are logged for auditing. If you set budget caps per project, you can have the agent automatically stop as soon as the quota is exhausted. Kurian argued that by matching models to the difficulty and characteristics of tasks, you can improve the accuracy of complex work while reducing the cost of simple tasks.
However, the key information is still left blank. Google Cloud has not yet disclosed a separate pricing structure for this agent, the timing for General Availability, or the results of independent benchmark tests. Also, the figure that “80% of customers use AI” refers to the AI product lineup overall, and has no direct connection to the actual adoption level of this Gemini agent.
A head-to-head showdown with Microsoft and OpenAI
This announcement came after a chain of competitor launches targeting the same enterprise customers. Microsoft, on September 25, fully revamped Copilot and introduced an always-on “Autopilot” agent. **OpenAI** unveiled the always-waiting “Dots” agent at a developer event four days later. Even the Anthropic that Gemini agents can call was deployed into workspace’s Docs, Sheets, and Slides on October 6.
As competition to launch “always-on AI agents” accelerates in the enterprise productivity tools market, it is expected that when and at what level Google Cloud will present pricing, release timelines, and performance metrics will significantly influence the landscape going forward.
Next reading: OpenAI’s 722 AI math papers have been fully released—now remains the evaluation by mathematicians