October 8, 2026 — Google Cloud introduced the Gemini agent, a single AI agent for work that can answer questions, handle tasks, create content, and write code. Why it matters: this is Google’s move from AI copilots to AI coworkers — and it can tap rival models like Anthropic’s Claude to do it.
What happened
Google Cloud on Thursday announced the Gemini agent, a unified AI agent for work, per a Reuters report. It plans work, uses tools, connects to a company’s business systems, and hands back finished work in documents, email, and developer environments. It works inside Google Workspace apps (Gmail, Docs, Sheets, Calendar) — and, notably, through Microsoft 365 and Slack too.
Details: model-agnostic, with “coworker agents”
Three things stand out:
- It picks the best model for the job — running today on Google’s Gemini models and Anthropic’s Claude models, with more to be added later. A Google product routing work to a rival’s model is a notable admission that no single lab’s models win everywhere.
- “Coworker agents” can be created as team members with their own email address and access limited to the information they’re given — essentially synthetic employees with scoped permissions.
- Industry versions for financial services and legal work are in preview now; government, healthcare, and retail versions are coming soon.
Why it matters
The context here is a fast-moving agent arms race. OpenAI launched always-on Dots agents in September that chase goals across apps on their own, and Meta released its Muse personal agent last month. Google’s pitch is different in one key way: it’s not asking companies to come to its tools — it’s embedding the agent where work already happens, even in Microsoft’s apps. That matters because enterprise adoption of AI has been throttled by fragmentation: a dozen pilots, no single workflow. If one agent can live inside Gmail, Slack, and Teams at once, it has a real shot at becoming the default.
The subscriber take: everything AI, tested. The “coworker agent” idea sounds great in a press release and terrifying in an IT audit — scoped permissions and email addresses for software agents will be where this lives or dies. Watch for how admins actually control what these agents can touch.
FAQ
Which AI models does the Gemini agent use?
It runs on Google’s Gemini models and Anthropic’s Claude models, selecting per task; Google says more models will be added later.
Does it work outside Google Workspace?
Yes — Google says the agent can also be used through Microsoft 365 and Slack.
What are “coworker agents”?
User-created agents that act as team members, with their own email address and access restricted to the information they’re given.
When is it available?
Google announced the agent on October 8, 2026, with industry-specific versions for financial services and legal already in preview; government, healthcare, and retail versions are coming soon.
Sources: Reuters