Gemini agent: what it is and how coworker agents work

Gemini agent is Google Cloud's new work assistant. It answers questions, does knowledge work, makes images and writes code, and it keeps running in the cloud after you close your laptop. You can also set it up as a "coworker agent" with its own Google Workspace account: an email address, a calendar, Drive storage and a listing in your company directory.

Google Cloud CEO Thomas Kurian announced it on October 8 at Gemini at Work 2026. His launch post on X came with a short video and a list of design principles, and the one most owners will notice is that Gemini "can also act as a coworker agent with a defined role and its own dedicated identity." Later that day @sundarpichai posted from NASA's Hangar One, where Google showed the agent to more than 700 Google Cloud customers. He called it "a single, universal agent for work."

What Gemini agent does

Kurian's pitch is that you give the agent an objective and come back later to finished work. Here is what that covers:

• Questions, research, documents, media and code all go through one prompt box, which you can reach from the web, phones, Windows, Mac, the command line, Workspace, Microsoft 365 and Slack.
• It runs in Google's cloud with one shared memory, so a job that takes hours or days keeps going while you do something else.
• For multi-step work it starts temporary helper agents, and Google says each one gets its own identity.
• It picks the model per job. Kurian says it "orchestrates across multiple models to deliver optimal quality and lower your costs."

The model part was the surprise of the day. Kurian's keynote write-up on Google's blog says the agent runs on Gemini models and on Anthropic's Claude models today, with other private and open models planned. @testingcatalog spotted a model menu in the launch video listing Gemini 4 Argon and Gemini 3.8 Flash next to Claude Opus 5 and Claude Sonnet 5.5, and noted it's the first time Claude models would show up on Gemini alongside Google's own. Google has sold Claude to developers through Vertex AI for a while. Putting a rival's model inside its main work product for business users goes a step further.

How coworker agents work

You describe a role, say a project coordinator for one team or a finance analyst, and Gemini creates an agent to fill it. Google says the agent gets its own Workspace account, with an address on an @agents.company.com style domain, plus a calendar, Drive storage and a spot in the company directory. People can add it to a Chat space, @mention it, or tag it in a document comment. Its edits show up under its own name in version history.

It only sees what your team shares with it. Security admins approve its permissions by role, and every action it takes lands in an audit trail under the agent's name, not a person's. All of its traffic goes through what Google calls an Agent Gateway, a firewall that enforces company policy. Admins can also set a hard spending cap in the Cloud Billing Console. When a project hits the cap, the agent pauses until someone resumes it.

Who can use it, and what it costs

For now, almost nobody outside the preview. 9to5Google reports the agent is in private preview, with wide availability coming "soon" for Workspace customers on select Business and Enterprise plans. Versions for financial services and legal are in preview, and Google says government, healthcare and retail versions are on the way. Google hasn't published a price or a general availability date.

My read

I think the identity setup matters more than the model menu. When an agent sends email as you and edits files as you, nobody can tell later which changes were yours and which were the software's. An agent with its own account, permissions and audit log can be managed more like a new hire, and most businesses will want that before they trust one with real work. OpenAI moved the same way last month with always-on dots. Google's version lives inside the email, calendar and Drive that a lot of small companies already use every day, which gives it an easier way in.

There's a catch, though. An agent that only sees what you share is only as safe as your sharing habits. If your Drive is one big folder anyone can open, a coworker agent can read all of it.

What this means for a small business

You don't need preview access to start preparing. Write down the recurring jobs a coworker agent could own, like quote follow-ups, the weekly report or sorting the shared inbox, and note what it should be allowed to touch for each. Then tidy up file sharing so those limits are easy to enforce. If you'd like help figuring out which of your processes are ready to hand off, New Face Design's free process audit starts with exactly that list.