On 8 October 2026, at its Gemini at Work 2026 event, Google Cloud announced the Gemini agent. Thomas Kurian, Google Cloud's CEO, described it in a blog post as "your new single, universal agent for work". Everything below about what the agent does is Google's description.
What Google says it does
According to Kurian's post, the agent answers questions, handles knowledge work, creates images and media, and writes and runs code, all from a single prompt box. Google's summary of how it operates: "It plans the work, uses skills and tools, connects to your systems, and brings back something finished". The post says that finished work appears inside the documents, the inbox and the developer environments people already use.
The post says the agent works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. It can also be reached from web, iOS, Android, Windows and Mac, and through the command line, Google Workspace, Microsoft 365 or Slack. Google also says it can run as a headless agent, meaning it needs no dedicated user interface. Reuters, in a story carried by ARY News, reported the Workspace, Microsoft 365 and Slack access in similar terms.
Kurian's post says work can be assigned, scheduled or triggered by events, and that tasks lasting hours or days keep running in the cloud after a user closes their laptop.
One agent, more than one model
The post separates the agent from the model beneath it: "Gemini is the agent, and the model underneath it is a separate choice." Google says it orchestrates across its own Gemini models and Claude models from Anthropic today, and that other leading private and open models are planned for the future. Constellation Research and Reuters both reported the same arrangement, with Reuters noting that other models would be added later.
Google presents the routing as a cost measure. Its post lists multi-model orchestration, smart routing and real-time spend caps. It says each job runs on the model that fits best, to deliver optimal quality and lower costs, and that the best model for a task is not always the largest one. Constellation Research reported that admins can choose models and set project-level spend caps, with individual caps to follow in the coming weeks. The post says premium sportswear brand On tested the model-selection capability.
Coworker agents
A feature that drew attention in the coverage is the "coworker agent". Google's post says a user can describe a role and Gemini creates an agent for it. "The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory." Colleagues can add it to a Chat space or @mention it, and the post says it can appear under its own name in a document's version history.
On access, the post says a coworker agent acts under its own identity rather than the user's, and "it sees only what you share with it". Reuters described these agents as acting as team members, with their own email address and access only to the information they are given.
Security, governance and industry versions
Google's post says it secures and governs agents through identity and policy management, authorisation and permission controls, secure sandboxing and network gateways. Constellation Research reported that Google described distinct agent identities, permissions, audit trails and policy enforcement.
Google's post describes industry-specific versions for financial services and legal teams. Kurian said at the event that these are now in preview, with government, healthcare and retail coming soon, according to Constellation Research, and Reuters reported the same.
Announced, and in private beta as of 8 October
Google's post announces the agent. Constellation Research, reporting on the event and a follow-up session with analysts, wrote: "General availability for Gemini agent, currently in private beta, is expected around the end of October or early November." That is the outlet's report as of 8 October.
Constellation Research also reported that pricing will be consumption-based, with no extra SKU, and that in Workspace current plans will continue, with the exact consumption model for the agent coming at a later date.
The wider race
Reuters framed the launch as part of a race among tech companies to sell AI that can work autonomously. It pointed to OpenAI's September launch of "always-on agents called dots that chase user goals across apps on their own", and to Meta's release of Muse, "a personal AI agent that can shop, book travel, send emails and make payments for users".
My reading is that Google's pitch is aimed at the company rather than the individual. The post gives much of its space to identity, permissions, spend caps and model choice, alongside its account of what the agent can do for one person. Those controls are Google's account of its own product.
Sources
Dignan, L. (2026) 'Google Cloud launches Gemini agent to work across enterprise systems', Constellation Research, 8 October 2026, available at: https://www.constellationr.com/insights/news/google-cloud-launches-gemini-agent-work-across-enterprise-systems (Accessed: 22 October 2026).
Kurian, T. (2026) 'Welcome to Gemini at Work 2026: Introducing the Gemini agent', Google Cloud Blog, 8 October 2026, available at: https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026 (Accessed: 22 October 2026).
Reuters (2026) 'Google Cloud introduces Gemini agent for work as AI race heats up', ARY News, 9 October 2026, available at: https://arynews.tv/google-cloud-introduces-gemini-agent-for-work-as-ai-race-heats-up (Accessed: 22 October 2026).
Picture credits
- Top picture: Thomas Kurian, CEO of Google Cloud, wrote the blog post announcing the Gemini agent. Photo: Eric Dietrich, U.S. Air Force (Public Domain) (source)
- Google's headquarters campus. Photo: Mutante, CC BY-SA 3.0, via Wikimedia Commons
