Google’s New Work Assistant Can Run on a Rival’s AI Model

Google’s Gemini agent takes a whole task from one prompt and finishes it inside Gmail, Docs and Sheets. The private-preview agent runs each job on whichever model fits, including Anthropic’s Claude, and will come with Gemini Enterprise, which nearly 90% of the Fortune 100 already use.
artificial-intelligence
entrepreneurship
Author

Kabui, Charles

Published

2026-10-28

Keywords

google-gemini-agent, gemini-enterprise, enterprise-ai-agents, multi-model-routing, ai-spend-caps

Download as PDF

Google announced a single assistant for whole office tasks on October 8, at its Gemini at Work event. The Gemini agent takes an outcome you describe, plans the steps, calls your company’s tools, and returns a finished document, deck or piece of code inside Gmail, Docs, Sheets, Slides, Chat and Calendar. It will come built into Gemini Enterprise, Google’s platform for company AI, at no extra charge. It runs in Google’s cloud, so a job keeps going for hours or days after you close the laptop, and one memory follows you across devices. Each task runs on whichever model fits, from the small Flash to the flagship Argon. Anthropic’s Claude models are already on the list.

Google calls the model underneath a separate choice. Administrators choose which models staff may use, and routing picks among them by cost and difficulty, so simple tasks can skip the priciest model. Google also points to August cost controls, including hard project-level limits that pause an agent when the budget runs out. Nearly 90% of the Fortune 100 already use Gemini Enterprise, but that counts the existing product. The agent is in private preview, with no release date.

The same agent runs Anthropic’s Claude or Google’s own models, and Google says more vendors’ models will follow. It says your context, skills and data stay put when the model changes, and every action goes to an audit trail. VentureBeat notes Google has supplied no independent benchmarks of how reliably these long, multi-app jobs finish.

Read More: Anthropic’s enterprise agent stack: sandboxes, MCP tunnels and a PwC deal

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Google’s {New} {Work} {Assistant} {Can} {Run} on a {Rival’s}
    {AI} {Model}},
  date = {2026-10-28},
  url = {https://toknow.ai/posts/gemini-agent-universal-work-agent-claude-models/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Google’s New Work Assistant Can Run on a Rival’s AI Model.” https://toknow.ai/posts/gemini-agent-universal-work-agent-claude-models/.