Google Cloud unveils Gemini agent that works in Workspace like a team member

On October 8, 2026, at the Gemini at Work 2026 event, Google Cloud presented a product called the Gemini agent. The company describes it as a single, universal agent for work that answers questions, handles office tasks, creates images and media, and writes and runs code. The most notable element is coworker agents, which have their own identity in Google Workspace.

One agent instead of many

According to Google Cloud's post, work is meant to start in a single prompt window, with the agent planning tasks, using skills and tools, and handing back a finished result. It supports chat, autonomous tasks, scheduled and event-triggered tasks, and code generation. The company lists web, iOS, Android, Windows, Mac, the command line, Workspace, Microsoft 365 and Slack among its surfaces.

The agent runs in the cloud and keeps a single set of memories and context. Google says tasks that take hours or days keep running after you close your laptop. The agent picks a model for each job; today these are Gemini models and Claude from Anthropic, with other private and open models to be added in the future.

An agent with its own Workspace account

Google highlights three modes of working across Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar: personal assistance, proactive delegation, and the role of a team member. In the last mode, a coworker agent gets its own Workspace account with an email address, calendar, Drive and a place in the company directory.

Colleagues add it to a Chat space or call it with an @mention. In a document's version history, the agent appears under its own name rather than the user's. Persistent agents have addresses on a domain such as @agents.company.com and see only the context shared with them. For multi-step tasks, Gemini can create temporary sub-agents, each with its own identity.

Tools, memory and costs

The agent connects to tools such as Confluence, Jira, Slack, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, as well as to any MCP server. Companies can publish their own tools and skills in registries, and skills are available at the global, department and personal levels.

On costs, Google promises routing between models and real-time spend caps. When a project's cap is reached, the agent pauses, and it can be resumed with one click. The post says per-token prices have fallen 98 percent since 2024, but it gives no pricing for the agent itself.

Security and industry versions

According to Google, every agent has a cryptographically attested identity with least-privilege permissions, and its actions are logged and attributed to the agent. The company also announced an Agent Sandbox with its own network boundary and Agent Gateway, a firewall for traffic to, from and between agents. These are the vendor's claims, and the post contains no independent testing.

Specialized versions for financial services and legal are in preview, while versions for government, healthcare and retail are coming soon. Google names CME Group and Deutsche Bank among users of the financial version. According to Google, Bloomberg Media raised its SQL query accuracy by 63 percent during initial development by grounding its data agents in the Knowledge Catalog.

What it means for companies

For Polish companies that use Workspace, the key point is that the agent is meant to appear in the system as a separate user, with an audit trail of its actions and a spend cap. At the same time, Google Cloud's post gives no general availability dates or prices, so it is hard to judge for now what such a deployment would cost. The agent currently uses Gemini and Anthropic's Claude models, and many of the adoption figures in the material come from Google itself.