Google’s new Gemini agent is one assistant that answers questions, does knowledge work, makes media and writes code from a single prompt box. Google announced it on October 8 at its Gemini at Work 2026 event. Right now it’s a private preview, so most of us can’t try it yet.
Here’s what it actually does, what’s still unknown, and why IT teams should care.
- Announced: October 8, 2026, at Gemini at Work 2026.
- Status: private preview, with wider access promised “soon”.
- Who it’s for: Workspace customers on select Business and Enterprise plans.
- Where it runs: Gemini Enterprise app, Gmail, Docs, Sheets, Slides, Chat, plus web, mobile, desktop and command line.
- Price and exact dates: not published yet.
That last line matters. Anything you read about cost right now is a guess, so treat any price you see in the wild with suspicion.
What the Gemini agent does that a chatbot doesn’t
Most assistants wait for your next message. This one is built to take an objective and keep working. Google’s own announcement says it runs in the cloud, so long jobs continue after you close your laptop.
It can also spin up temporary sub-agents for bigger jobs, running steps in parallel or in order for hours or days. Think of it as a project lead that hands out pieces of work, not a single reply box.
It also reaches beyond Google’s own apps. It can connect to tools like Slack, Jira, Salesforce, Microsoft Office and any MCP server, so it can pull context from wherever your team already works.
Google gives two simple examples. Ask for a project update and it can turn that into a slide deck. Ask it to schedule a meeting with your usual team and it can work out the attendees from a chat space and check their calendars, including external guests.
Coworker agents and the memory behind them
The most interesting idea is the “coworker agent.” You describe a role, like a finance analyst, and the agent gets its own Workspace account with an email address, calendar and Drive storage. It only sees what’s been shared with it.
To keep context, the agent uses four kinds of memory. Here’s how Google describes them:
| Memory type | What it holds |
|---|---|
| Session | What’s happening in the current task |
| Semantic | A knowledge base the agent builds over time |
| Procedural | Skills, including ones the agent writes for itself |
| Episodic | What it did in past runs |
Reusable skills are the practical part. Teams can publish a skill once, like “build the weekly sales report,” and everyone’s agent can use it. Saved reports can then run on demand without extra token costs, according to Google.
Which models power the Gemini agent
The agent picks a model for each job. Today that includes the Gemini family and Anthropic’s Claude models, with other private and open models planned. Google’s keynote also named Argon for frontier reasoning, Flash for speed, and Omni for media.
If Argon sounds new, we covered why only Google’s cyber partners can use Gemini 4 Argon for now. That limit is worth remembering when you ask which models you will really get.
Google also says “Smart Routing” sends each task to the cheapest model that can do it well. That’s a vendor claim, so test it on your own workloads.
Admin controls: what IT teams get
An agent that can act on your files needs tight limits. Google lists these controls:
| Control | How it works |
|---|---|
| Agent identity | Each agent has its own attested identity, shown in logs |
| Permissions | Role-based, least-privilege, approved by security admins |
| Auditing | Every action is logged and attributed to the agent |
| Agent Gateway | An AI network firewall; one policy applies to all agents |
| Spend caps | Per-project limits; the agent pauses when it hits the cap |
The spend cap is the one finance teams will like. Agents that run for days can quietly burn budget, and a hard pause beats a surprise bill.
How it compares to rival work agents
Microsoft is pushing hard in the same space. Our breakdown of Microsoft Copilot Autopilot, Home and Code shows a similar bet on agents that act, not just chat.
The big difference so far is Google’s pitch of one agent across Workspace, Microsoft 365 and Slack. If that works as promised, it matters for mixed-tool companies. According to 9to5Google, it’s still in private preview, and Google hasn’t said how it relates to the existing Gemini app agent.
Early customer numbers, like a bank cutting document review from an hour to five minutes, come from Google and aren’t independently verified.
Frequently Asked Questions
Is the Gemini agent available now?
Only in private preview. Google says wide availability is coming soon for Workspace customers on select Business and Enterprise plans.
How much does the Gemini agent cost?
Google hasn’t published pricing. Check Google Cloud’s pricing pages or ask your account rep before you plan a budget.
Can I use it on my personal Gmail account?
Google hasn’t said. The launch focused on business customers, so assume consumer access is not part of this preview.
Does the Gemini agent only use Google models?
No. It currently uses Gemini and Claude models, and Google plans to add other private and open models.
Should you care yet?
If you run IT for a Workspace company, yes, start planning. Ask Google for preview access, decide who owns agent permissions, and set spend caps before anyone builds a long-running workflow. If you’re a regular user, there’s nothing to do but wait. This is a promising but unproven launch, and the missing price and dates are the real story until they show up.
Leave a Reply