Google Cloud launches the Gemini agent -- one agent for all work, routing across Gemini and Anthropic models, with per-project spend caps
Announced by Thomas Kurian at Gemini at Work 2026, the agent takes objectives instead of instructions, spawns sub-agents with their own identities, and pauses itself when a project hits a hard AI-spend limit.
Google Cloud CEO Thomas Kurian introduced the Gemini agent on October 8 at Gemini at Work 2026, calling it "your new single, universal agent for work." The pitch is delegation rather than prompting: "You give it objectives, not instructions." The agent runs in the cloud so long jobs keep going after a laptop closes, can spin up temporary sub-agents each with their own identity, and can act as a "coworker agent" with a dedicated email address, its own storage, and access only to the context people share with it. It routes each job across Google's Gemini models and "Claude models from Anthropic," and connects to systems including Salesforce, ServiceNow, BigQuery, Snowflake and MCP servers.
The numbers Google put on the table
Google says that in the last year "nearly 500 Google Cloud customers each processed more than one trillion tokens," that nearly 80% of Google Cloud customers use its AI products, and that "per-token prices have dropped 98% since 2024" while enterprise AI volume has grown. The post lists dozens of customer results -- these are Google's and its customers' own figures, not independently verified. Industry versions are in preview for financial services and legal, with government, healthcare and retail listed as coming soon. The post doesn't give general-availability timing or pricing for the agent itself.
The feature that matters for spend
The most concrete control in the launch is the budget one: you can "set a hard limit on a project's AI spend in the Cloud Billing Console, and Gemini enforces it," pausing that project's agent when the cap is reached, with costs tracked per project so they can be charged back to teams. A falling per-token price paired with exploding volume is exactly how AI bills grow without anyone deciding they should. A hard cap stops the overrun, but it doesn't say whether the work done under the cap was worth it -- that still takes measuring outcomes, not tokens.