Google's Gemini agent: what the "universal agent for work" does, who gets it and what it costs

On October 8, 2026, at its Gemini at Work event, Google Cloud announced the Gemini agent, which it calls "your new single, universal agent for work." In Google Gemini's business product, one agent now answers questions, does knowledge work, makes images and media, and writes and runs code. Google says you hand it an outcome rather than instructions, and it can keep working for hours or days after you close your laptop. It is not broadly available yet. Google hasn't given a release date or detailed pricing.

This story is based on Google Cloud CEO Thomas Kurian's keynote post, Google's Gemini Enterprise product page, and reporting by VentureBeat, 9to5Google, TechRepublic and CBS News, checked October 12, 2026.

The short version

• What: one agent for chat, autonomous background work and code, inside Gemini Enterprise, Google Workspace and third-party apps.
• Status: private preview for "a select group of customers," per a Google Cloud spokesperson quoted by VentureBeat. Google says wider availability is coming "soon," with no date.
• Price: Google told VentureBeat the agent comes "at no additional charge" where Gemini Enterprise is available, including through Workspace business and enterprise subscriptions. Google will "share additional pricing details closer to the wider release."
• Models: Google says it routes jobs across Gemini models and Anthropic's Claude models "today," with other private and open models later.
• Controls: each agent gets its own identity, actions are logged to an audit trail, it runs inside a sandbox, and admins can set per-project spending caps.

What is the Gemini agent?

Kurian's post describes it as one agent that "plans the work, uses skills and tools, connects to your systems, and brings back something finished." It works "inside the documents, the inbox, and the developer environments you already work in." The pitch is a step beyond the chat assistants of the past two years: "You give it objectives, not instructions. You delegate an outcome and come back to finished work."

Google says it runs on web, iOS, Android, Windows and Mac and in "any channel (command line, Google Workspace, Microsoft 365, or Slack)," and developers can embed it through an API as a "headless agent." It runs in Google's cloud, so long jobs continue when your device is off. It spins up temporary sub-agents for big jobs, and it keeps four kinds of memory, including "skills it writes for itself."

If "agent" still feels like a slippery word, our explainer on what "agentic" really means breaks down the difference between an assistant that answers and one that acts.

What can it access?

According to Google, the agent can connect to:

• Workspace: Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. 9to5Google reports you can call it by typing @Gemini inside Gmail, Docs, Sheets, Slides and Chat.
• Other business software: Google names Confluence, Microsoft Office, Teams, Slack, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, plus "files on your own desktop."
• Any MCP server, "inside or outside your company network." Our explainer on MCP covers how that protocol works.

The most unusual feature is the coworker agent. Google says a team can create agents with "dedicated identities including their own @agents.company.com emails, their own persistent storage," that see only the context the team provides. VentureBeat's example is an "Event Planner Agent" with its own Workspace account, calendar and Drive storage.

Which models does it use?

Kurian writes that the agent orchestrates "across our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future." A feature Google calls Smart Routing picks the model for each job based on cost and performance. VentureBeat reports that admins can restrict which models agents may use.

There's an inconsistency to watch. As of October 12, the FAQ on Google's Gemini Enterprise page still says "Google Gemini models today, with additional models coming soon." The same page lists "third-party model choice" among features "currently in early access." Read together, Claude support appears to exist in preview rather than for every customer.

On data handling, VentureBeat reports Google says prompts, documents and chat histories stay inside Gemini Enterprise when Claude is used. It also reports that Google "did not specify whether information is transmitted to external model providers for processing." For Google's own frontier model, see our coverage of Gemini 4 Argon, which Kurian lists as the top of the Gemini lineup.

Who can get it, and how much does it cost?

Last verified: October 12, 2026

VentureBeat reports Google hasn't said whether coworker agents need extra licenses, how usage-based charges apply to long tasks and sub-agents, or how many coworkers an organization can create. Google's main cost control is a per-project spending cap: when it's hit, "that project's agent pauses" until someone resumes it.

Security and admin controls

Google lists these safeguards:

• Identity: "Every agent gets its own identity, cryptographically attested and governed like an employee, with least-privilege permissions."
• Audit: "every action Gemini takes is written to an audit trail and attributed to the agent rather than to a person."
• Sandbox: agents run "inside an Agent Sandbox with its own network boundary."
• Agent Gateway: described as "an AI network firewall" that enforces policy in real time. Google's example rule: "agents may not open documents classified Need to Know."

An agent that reads email and documents can be steered by instructions hidden in them, which is why our piece on the case against unattended agents argues for hard limits rather than trust. Google hasn't published independent evaluations of the agent on long, cross-app tasks.

How it fits with the competition

CBS News pointed to Meta's Muse personal agent, launched in September, and OpenAI's "Dots" helper for ChatGPT Pro and Business Premium users as recent rivals. The bigger fight is with Microsoft: Google's agent also works in Microsoft 365, Teams and Slack, the same surfaces as Microsoft's Copilot. Analyst Carmi Levy told Computerworld (as quoted by Reworked) that Google is trying to become "the gatekeeper of the new enterprise operating system."

On the consumer side, Google is replacing Gems with skills. The business agent goes further, writing its own skills and remembering how you work.

What's confirmed, reported and uncertain

• Confirmed by Google: the October 8 announcement; the design principles; Workspace, third-party and MCP connections; coworker agents with their own identities; Claude support through orchestration (per Kurian's post); spend caps, audit, sandbox and Agent Gateway; and the early-access list on the Gemini Enterprise page.
• Reported (Google spokesperson to VentureBeat): private-preview status; "no additional charge" where Gemini Enterprise is available; pricing details to come before wider release.
• Uncertain: the general availability date; whether coworker agents need extra licenses; usage-based charges; whether data leaves Google when Claude handles a job; and why the product FAQ still lists only Gemini models.