Google Launches Gemini Agent: It Works for You Across All Company Software

Not a Chatbox — a Workhorse

On October 8, Google Cloud unveiled Gemini agent at its “Gemini at Work 2026” event. Google calls it “a general-purpose agent for work.”

The usage model has changed: no more step-by-step instructions — just give it a goal. It plans on its own, picks its own tools, connects to the company’s business systems, and hands back finished work. The agent runs in the cloud; close your laptop and the task keeps going.

This is the third big move in roughly a month: OpenAI launched its always-on agent “dots” in September, and Meta launched the personal agent Muse, which can shop and book travel. Google has chosen the enterprise office as its battlefield.

What It Can Do for You

Google showed two examples at the event.

Scheduling meetings: say “set up a meeting with my regular contacts,” and it finds the people, checks calendars, coordinates times — handling external attendees too.

Doing research: ask for a market study, and it builds the findings into a financial model in Sheets, then hands you a slide deck.

The capability list: answering questions, knowledge work, writing documents, writing code, running code, and coordinating tasks across business systems. A chatbox answers questions; this agent gets the job done.

Which Software It Can Reach

Reach is this launch’s biggest selling point.

Google’s own: Gmail, Drive, Docs, Sheets, Slides, Chat, Calendar. The agent shares the same business context and permissions across these apps, so cross-app work needs one instruction.

Third-party software is in too: Microsoft 365, Slack, Salesforce, ServiceNow, Jira, Shopify. Data platforms aren’t left out: Snowflake, Databricks, and Google’s own BigQuery.

Devices: web, iOS, Android, Windows, Mac, and the command line, plus a headless mode — it can be embedded into third-party apps to do the work without its own interface.

Close the Laptop, the Work Continues

The agent lives in the cloud. Tasks can run for hours or days; shut your computer and it keeps going.

It maintains unified memory and context: brief it on your phone, check in on your computer, no need to re-explain when you switch devices. Reports say Google gave it four kinds of memory — conversational, semantic, procedural, and episodic — with the goal of learning your work habits over time.

For complex tasks, it breaks things down on its own: spinning up temporary sub-agents, each owning a piece, running in parallel or in sequence, with no manual delegation from you.

The Headliner: Give It a Desk

“Coworker agents” were the most talked-about feature at the event.

You can create an agent with a fixed identity: its own email address (@agents.company.com), its own calendar, its own storage. You assign it a role — bookkeeper, inventory manager, marketing consultant. It can only access information you explicitly give it — the part enterprises care about most.

Google’s imagined scenario is specific: small and midsize businesses with lean teams and a pile of miscellaneous work, letting agents take on fixed roles like bookkeeping, inventory, and market analysis.

The Model Is Plug-and-Play

One notable design choice: the agent is separated from the underlying model.

Right now it automatically picks between Google’s own Gemini models and Anthropic’s Claude models depending on the task, balancing quality and cost. Google says more private and open models will be added later.

Every lab is racing on models; Google has built a “model router” at the agent layer first. That means swapping models later won’t require swapping agents — the agent layer becomes the new entry point.

Enterprise First

This product line is clearly aimed at enterprise customers.

Tailored versions for finance and legal are already in preview, with government, healthcare, and retail editions on the way. The customer cases at the event were all enterprise-flavored: Brazil’s Bradesco bank used it to cut document review time from one hour to five minutes; Orange Spain has deployed over 1,000 custom agents.

Google also rolled out cost-control tools (reported by techstrong.ai) — clearly worried about enterprise finance teams’ hearts skipping a beat when the bill arrives.

What Wasn’t Said

Google didn’t disclose full pricing or a timeline for general availability. Per Reuters, Alphabet shares edged up on the day of the announcement — the market, at least, bought it.

For regular users, two things are worth watching: when this makes it into personal Google Workspace subscriptions, and how tight the coworker agents’ permission boundaries turn out to be in practice. How strict those boundaries are determines whether it can truly work as a colleague.