Google launches Gemini agent, a single AI agent for work tasks

What Google Cloud announced

Google Cloud introduced Gemini agent this Thursday (the 8th), described by the company as a single AI agent for work. According to Reuters, it can answer questions, execute tasks, create content, and write code. The news was published by India's Economic Times on October 9, 2026.

The move doesn't come out of nowhere. It follows the September launch of OpenAI's always-on agents called dots, which pursue user goals across applications autonomously. Last month, Meta launched Muse, a personal AI agent capable of shopping, booking travel, sending emails, and making payments on the user's behalf. Alphabet's shares rose slightly at the opening of trading after the announcement.

How Gemini agent works in practice

According to Google, the agent plans the work, uses tools, connects to a company's business systems, and returns finished work in documents, email, and development environments. That last part is what matters most to developers: the agent isn't limited to generating loose text, it aims to deliver a finished artifact right where the team already works.

Gemini agent operates inside Google Workspace apps, including Gmail, Docs, Sheets, and Calendar, and can also be accessed via Microsoft 365 and Slack. This means that bringing it into a company doesn't require abandoning whatever productivity suite the team already uses today.

In short: the agent is designed to live where work already happens, not to be yet another isolated app that requires switching context.

Model routing: Gemini and Claude in the same layer

One of the most relevant technical details of the announcement is per-task model selection. Gemini agent selects the best model for each job and, currently, that includes both Gemini models and Anthropic's Claude models, with other models to be added later.

For those building on top of this layer, the detail suggests that Google is treating the language model as an interchangeable piece within the product, not as the product itself. In practice, this means the conversation about "which model to use" stops being a manual decision made by the engineering team and becomes a routing decision made by the platform itself, task by task. It's a pattern that already appears in other agent stacks and tends to push teams toward model-agnostic architecture, instead of locking integrations to a single provider.

Coworker agents with their own email

The announcement includes a feature that changes how permissions are thought of within an organization: users can create "coworker agents," agents that act as team members, with their own email address and access only to the information they are granted.

This brings the agent's identity model closer to that of a human employee: it receives an inbox, participates in communication flows, and has a defined access scope, instead of automatically inheriting the permissions of whoever configured it. For platform and security teams, the design points to a problem already familiar from corporate integrations: how to audit and revoke access for an identity that isn't a person but acts as if it were.

Where it's already available and what's still missing

Versions built for financial services and legal work are already in preview. Versions for government, healthcare, and retail are on the way, according to Google. The source doesn't detail a timeline for general availability, pricing, usage limits, or how the connection to third-party business systems works beyond what has already been mentioned.

The released material also contains no technical detail on how developers will integrate or extend Gemini agent beyond use within Workspace, Microsoft 365, and Slack. That's the gap that matters most to those building enterprise software: whether an open API layer will exist to orchestrate these agents within their own products, or whether access will remain limited to the integrations already announced.

What changes for those integrating tools in Brazil

Teams that currently maintain integrations with Workspace, Microsoft 365, or Slack need to track how Gemini agent positions itself relative to bots and automations already present in these environments. The coexistence between an agent that reads email, creates documents, and writes code and a company's internal automations tends to create overlapping responsibilities that don't yet have a defined standard.

The competition between Google, OpenAI, and Meta over this productivity agent format also signals that the decision of which AI vendor to use is no longer just about which model answers better. It now involves which agent is already plugged into the tool suite the company uses day to day, and that's an adoption criterion different from what engineering teams used to prioritize until recently.