Guides / Gemini agent · By Phillip Hanasz · Published 25 October 2026

Gemini agent: Google’s agent for work

Gemini agent is the single “universal agent for work” Google Cloud announced on 8 October 2026 at Gemini at Work. It answers questions, does knowledge work, writes and runs code, runs tasks for hours or days, and can connect to any MCP server. It is in private preview for selected enterprise customers, and Google has not published a separate price.

What is Gemini agent?

It is one agent for all kinds of work, inside Gemini Enterprise and Google Workspace. Google’s description: the agent is Gemini, and the model underneath is a separate choice. What it does, by Google’s announcement and early coverage:

Long-running work in the cloud. Tasks can run for hours or days, start from schedules or events, and continue on another device.
Sub-agents. It splits complex jobs across temporary teams of sub-agents.
Coworker agents with their own account. Google’s example, an event planner agent, gets its own Workspace account with an email address, calendar, Drive storage and a place in the company directory. It sees only files shared with it.
Everywhere you work. Web, iOS, Android, Windows, Mac, the command line, Workspace (@Gemini in Gmail, Docs, Sheets, Slides and Chat), Microsoft 365 and Slack, or headless through an API.
Several model families. Gemini models and Anthropic’s Claude today, with others planned. Smart Routing picks the cheapest model that does the job; admins can restrict which ones are allowed.

Several write-ups point out that it is a different product from the agent mode in the consumer Gemini app.

Who can get Gemini agent and what does it cost?

Today only selected enterprise customers, in private preview. Google says wide availability will follow for Workspace customers on select Business and Enterprise plans, without a date. VentureBeat reports it will be included at no extra charge where Gemini Enterprise is available, with more pricing detail promised closer to release. Open questions as of 25 October 2026:

whether coworker agents with their own Workspace account need their own license,
whether long tasks and sub-agents are billed by usage,
availability outside North America and data residency.

What is known: project-level spend caps in the Cloud Billing Console pause an agent when its budget runs out, and work resumes with one click.

Does Gemini agent support MCP?

Yes. Google says Gemini can connect to any Model Context Protocol server inside or outside the company network, and teams can publish tools to an internal registry for everyone else. Named connectors include Salesforce, ServiceNow, Jira, Git, BigQuery, Snowflake, Databricks, SAP, Workday and Microsoft Teams.

Google has not published the setup steps yet. For a remote MCP server such as Next Everything, which gives agents live JLCPCB, DigiKey and Mouser stock and prices, you need the endpoint https://mcp.nexteverything.io/mcp and the header Authorization: Bearer <your key> from your dashboard. If a client only takes a URL, https://mcp.nexteverything.io/mcp?key=<your key> works too. We have not tested Gemini agent ourselves, because it is in private preview; the same values work in every client on our docs page.

Security and control

Own identity: each agent has a cryptographically attested identity with least-privilege permissions, set by your security admins.
Audit trail: every action is logged and attributed to the agent, not to a person.
Agent Sandbox and Agent Gateway: tasks run in a sandbox with its own network boundary, and all traffic to, from and between agents passes a policy firewall.

Google’s announcement does not describe a general “ask before you send or pay” step like Claude Duet’s. For any agent that can order parts, keep the payment step with a person: Next Everything never charges on its own, it emails a payment link.

Comparison grid: Gemini agent, Copilot Autopilot and Claude Duet are in preview and Grok Bot in early beta; price published for Claude Duet and Grok Bot, credits only for Copilot, none for Gemini agent; own MCP server yes for Gemini agent, Duet and Grok Bot, through Copilot Studio for Microsoft; own identity or mailbox yes for Gemini agent and Copilot Autopilot — Gemini agent and Copilot Autopilot are built for company IT; Duet and Grok Bot for individuals.

Gemini agent vs Copilot Autopilot vs Claude Duet

	Gemini agent	Copilot Autopilot	Claude Duet
Announced	8 Oct 2026	25 Sep 2026	21 Oct 2026
Status	Private preview	Private preview	Research preview, US waitlist
For	Companies on Gemini Enterprise / Workspace	Companies on Microsoft 365 Copilot	Claude Pro and Max subscribers
Price	Not published	Usage, in Copilot Credits	From $20 a month (Pro)
Your own MCP server	Yes, any	Through Copilot Studio	Yes, custom connector

More on the others: Microsoft Copilot Autopilot, what is Claude Duet and personal AI agents in 2026.

FAQ

Is Gemini agent available now?

Only in private preview for selected enterprise customers, as of 25 October 2026.

Does Gemini agent cost extra?

Google has not published a price. Reports say it will be included where Gemini Enterprise is available; usage-based charges for long tasks are not yet clear.

Can Gemini agent use Claude?

Yes. Google says it runs on Gemini and Claude model families today.

Can Gemini agent order electronic parts?

Only through a tool that can, such as a parts MCP server, and with a person approving payment. Which agents can do it today: can an AI agent order electronic parts.

Sources

About this guide

We build Next Everything, an MCP server that lets AI assistants and agents check a BOM against the JLCPCB/LCSC catalog and live Mouser and DigiKey prices, and order the parts you approve. Features, prices and availability here are as announced by each company and reported in the press on the date above. These products are in beta or preview and change often.

Next Everything is not affiliated with Anthropic, Google, Meta, Microsoft, OpenAI, xAI, Hark or the other companies named here. Product names belong to their owners.