Gemini Agent Enters Private Preview With Smart Routing

The Gemini agent can run a job for hours or days and pick a different model for each task. That sounds great if you own a delivery plan, until you read the availability line: private preview, no general release date.

Google unveiled it on October 8, 2026 at its Gemini at Work 2026 event. Below: what is confirmed, what is not, and what to ask before it shows up in a plan.

What Google announced about the Gemini agent

Google calls it a "new single, universal agent for work". According to Google Cloud's announcement, it answers questions, does knowledge work, creates images and media, and writes and runs code, all through one agent and one API. It works in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, and can also be reached from Microsoft 365, Slack and the command line.

For bigger jobs it creates temporary sub-agents, each with its own identity, that run parallel or sequential steps for hours or days. You can also assign a persistent "coworker" agent with its own @agents.company.com address and storage, which sees only what you share with it.

Smart Routing and multi-model orchestration

Google says the agent "runs each job on the model that fits best", orchestrating across Gemini models and Claude models from Anthropic today, with other private and open models planned. Google's post calls this a routing tool; press coverage calls it Smart Routing. The announcement does not name Claude versions. Some outlets do, but I could not trace those names to Google, so treat them as unconfirmed.

Spend caps per project

Admins can set a hard limit on a project's AI spend in the Cloud Billing Console. Google says Gemini tracks token usage and sandbox costs, and when the cap triggers, that project's agent pauses until someone resumes it with one click. Because tracking is per project, costs can be charged back to departments.

Gemini agent private preview: what is confirmed and what is not

Confirmed by more than one source:

• Private preview. 9to5Google reports it is in private preview, with Google promising "wide availability soon for Workspace customers with select Business and Enterprise plans". A Google Cloud spokesperson told VentureBeat that wider availability is forthcoming, with no date, as relayed by Let's Data Science.
• Multi-model routing, Claude included. Stated in Google's own post.
• Spend caps and per-agent identity, sandboxing and audit trails. Also in Google's post.

Not confirmed:

• Price. Google's post gives none and only mentions "flexible spending options". Secondhand reports say it will be bundled with Gemini Enterprise, and it is unclear whether usage will still be metered. Treat price as unknown.
• Details Let's Data Science flags as unanswered: general availability timing, regional coverage, supported connectors, usage limits, and administrator controls over routing.

Why the Gemini agent matters for delivery plans

Three things change your risk picture, even before you have access.

First, if a router picks the model per job, you cannot assume the same model handles the same task next week. Quality review becomes your job, not the vendor's.

Second, a spend cap turns cost into a schedule risk. When the cap triggers, the agent pauses. A stalled task two days before a milestone is a bigger problem than a billing email.

Third, agents with their own identities need an owner. Someone has to approve their permissions and read their audit trail.

What to do on Monday morning

Keep it off the critical path. Treat it as a dependency with an unknown date.

1. Ask your Google Cloud or Workspace account team whether your organization is in the preview, and for a date.
2. Ask about price and metering, regional availability, and whether admins can see and restrict which models handle which work.
3. Pick one low-stakes recurring task, such as a weekly status draft, as the first test once you have access.
4. Decide who owns the spend cap, who gets told when an agent pauses, and who reviews the agent's permissions.

Until Google publishes a date and a price, the Gemini agent belongs in your plan as a pilot with a named owner and a fallback, because an agent that can pause itself and switch models is only as dependable as the questions you asked before launch.