You use google's gemini agent for work, explained: what it does, who gets it, and the controls that matter by selecting the right tool, preparing your input (text, images, or videos), configuring settings, generating content, and refining the output through iteration.
What Google announced
At its Gemini at Work event on October 8, 2026, Google Cloud CEO Thomas Kurian introduced the Gemini agent, which Google's announcement calls "a single, universal agent for work." It answers questions, handles knowledge work, creates images and media, and writes and runs code, "all in a single agent and a single API." The pitch is in one line: "You give it objectives, not instructions. You delegate an outcome and come back to finished work."
It is not available to most companies yet. 9to5Google reports it is "currently in private preview" and "will see wide availability soon for Workspace customers with select Business and Enterprise plans." Google has not published a price or named which plans, and consumers come after businesses.
What it actually does
Strip away the keynote language and there are four ideas, each a step past a chatbot:
- It runs in the cloud, not on your laptop. Work that takes hours or days "keeps running after you close your laptop," and it keeps one set of memories whichever device or app you use it from: web, iOS, Android, Windows, Mac, the command line, Workspace, Microsoft 365 or Slack.
- It already knows your context. Google's example: ask it to set up a meeting with "the usual team of regional event leads next week" without naming anyone, and it works out who they are from your chat spaces and last event thread, then checks their calendars. It keeps four kinds of memory: for the task at hand, for what it has read, for how jobs get done, and for what it has done before.
- It splits work across sub-agents. For multi-step tasks it creates temporary, job-specific sub-agents, each with its own identity, and coordinates them in parallel or in sequence.
- It can become a coworker. Describe a role and it creates a permanent agent with "its own Workspace account, including an email address, calendar, Drive, and presence in your company directory." Colleagues @mention it in Chat, and its edits show under its own name in version history.
It connects to Confluence, Microsoft Office and Teams, Slack, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres, Snowflake, desktop files and any MCP server, and companies can publish their own tools and skills to a shared registry.
The model underneath is a separate choice
The most unusual line in the announcement is this one: "Gemini is the agent, and the model underneath it is a separate choice." The agent routes each job to the model that fits, "across our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future." Google names its own line-up as Argon for frontier reasoning (see Gemini 4 Argon), Flash for speed and volume, Omni for media and Gemma for edge work. It does not say which Claude models are offered, so check with Google before planning around Opus 5.5 or Sonnet 5.5. Google calls the routing Smart Routing and sells it as a cost control: simple loops go to cheap models, hard steps to expensive ones.
The controls IT teams should look at first
An agent with its own email address and access to Jira and Snowflake is a security question before it is a productivity one. Google frames it as four questions: who is this agent, what is it allowed to touch, what did it do, and what should it never touch. Its answers:
| Control | What Google says it does |
|---|---|
| Identity | Every agent gets its own identity, "cryptographically attested and governed like an employee," with least-privilege permissions |
| Audit trail | Every action is logged and attributed to the agent, not to the person who asked |
| Agent Sandbox | Every task runs inside a sandbox with its own network boundary |
| Agent Gateway | All traffic in, out and between agents passes a policy firewall: write a rule once, such as "agents may not open documents classified Need to Know," and it applies to every agent |
| Spend caps | A hard limit per project in the Cloud Billing Console; when it is hit, that project's agent pauses until someone resumes it |
The access model is the part to understand before rollout. A coworker agent "sees only what you share with it," and its access "follows the sharing and membership your team already uses." That is sensible, and it also means the agent inherits every overshared folder and every stale group membership you already have. Cleaning up sharing is the first real preparation step.
Where it fits against the other work agents
Every large vendor now sells the same idea, and they differ mainly in where they live. Microsoft's Copilot Autopilot is built around Microsoft 365. Meta's Muse is a personal agent for errands, not a company tool. Google's distinguishing bets are that the agent is not tied to one vendor's model, that it works inside Microsoft 365 and Slack as well as Workspace, and that the controls are one gateway and one billing console. Those are claims. Until the preview opens up, nobody outside Google's early customers has tested them.
What to do now
For how multi-agent systems like this are put together, see AI agent architecture patterns, and for agents that live in Chat and Slack, AI agents in team channels. The other assistants are in every AI assistant in the directory. The launch is in the October 8 briefing.
Explore curated tools related to this guide: