Google launches Gemini agent, and it runs Claude too

Google just pitched one agent to replace your pile of them.

Google Cloud CEO Thomas Kurian introduced the Gemini agent on Thursday at Gemini at Work 2026, calling it a single, universal agent for work, according to Google Cloud's post adapted from his keynote. Google says it answers questions, handles knowledge work, makes images and media, and writes and runs code, all from one prompt box and one API.

One agent, many models, Claude included

Google says the agent runs in the cloud with one set of memories across devices, so long jobs keep going after you close your laptop. It works inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. You can reach it from the web, iOS, Android, Windows, Mac, the command line, Microsoft 365 and Slack.

For bigger jobs it creates temporary sub-agents with their own identities. Google also describes "coworker agents" with persistent roles, their own Workspace accounts and @agents.company.com email addresses.

It connects to Jira, Salesforce, ServiceNow, Snowflake, Databricks, BigQuery, Postgres and local desktop files, plus any MCP server, per Google.

The model line is the surprise. Google says each job runs on the best-fit model across the Gemini family and Anthropic's Claude models, available today, with other private and open models planned.

On cost, Google lists Smart Routing to the lowest-cost model that meets the job, and real-time spend caps set in the Cloud Billing Console. When a cap trips, the project's agent pauses until someone resumes it.

Versions for financial services and legal are in preview. Government, healthcare and retail are listed as coming soon, with no dates.

Google is selling the controls around the model

Here's the thing. Google will route your work to a rival's model inside its own AI agent. That tells you where it expects to earn: in identity, permissions, audit logs and billing.

The governance list is long. Google says each agent gets a cryptographically attested identity, every action is logged to the agent rather than a person, agents run in a sandbox, and an Agent Gateway enforces policy on agent traffic in real time.

That's the part your IT department will read first. An agent that can open any document is a liability. One that obeys a rule like Google's example, "agents may not open documents classified Need to Know," is something a security team can approve.

Google also cites a customer at scale: SOMPO, it says, has built more than 10,000 custom agents across its 34,000 employees.

What you can do with it today

Not much, unless you're in a preview. Google's post gives no price for the Gemini agent and no general availability date.

If you're on Workspace, ask your Google account team three things:

• When does your plan get it, and at what price?
• Can your admins allow or block specific models, including Claude?
• Can spend caps be set per department?

Google also says per-token prices have fallen 98% since 2024. Agents spend tokens in loops, though, so the spend cap will matter more to your bill than that figure.

Watch for the price sheet

The launch that counts is the one with a price attached. Until Google publishes one, any comparison to the agents you already pay for is on paper only.

Get the preview terms. Then decide.