Google Introduces Gemini Agent for Work Across Business Apps

Google Cloud introduced a Gemini agent for work on Oct. 8, describing a single interface that can answer questions, create content, write and run code, and carry out assignments across business applications. In its announcement, the company said the agent plans tasks, connects to business systems and returns completed work rather than only suggesting steps. Those capabilities are company descriptions, not independently measured results.

The launch puts the emphasis on continuity: Google says the agent runs in the cloud, retains context across devices and can keep working on assignments that take hours or days after a user closes a laptop. It can be accessed through the web, mobile and desktop environments, the command line, Google Workspace, Microsoft 365 and Slack, according to the company. That makes the announcement broader than a feature confined to one document editor.

One Agent, Multiple Models

Google distinguishes the agent from the underlying model that handles a particular job. The company says the system can orchestrate work across its Gemini models and Anthropic's Claude models, with other private and open models planned. Its stated goal is to match a task to an appropriate model rather than always use the largest one. Claims that this improves quality or reduces costs remain Google's assessment of the approach.

The agent can also create temporary, task-specific sub-agents, each with an identity, to coordinate parallel or sequential work, Google said. A separate coworker-agent concept provides a persistent role, dedicated storage and an email identity. According to the announcement, those coworker agents receive access only to context that users or their teammates provide. The distinction matters: a temporary helper and a standing software team member are different administrative responsibilities.

Google says the product connects to collaboration software, development tools, enterprise platforms and databases. The announcement names services including Salesforce, ServiceNow, Git, Jira, BigQuery and Postgres, alongside Workspace and Microsoft applications. Teams can publish tools and reusable skills into shared registries. In Google's description, those skills are modular instructions and workflows that teach the agent how an organization wants a particular kind of job completed.

Controls Are Part of the Product Pitch

The company's launch materials put identity and policy management, authorization, permission controls, sandboxing and network gateways alongside the agent's functional capabilities. They also describe model routing and spending controls. Those safeguards are central to evaluating the proposal: Google is not simply offering answers in a chat window, but a system it says can use connected tools and execute work within business environments.

In a separate announcement for smaller businesses, Google said Gemini can connect to services including Asana, Docusign, Dropbox, GitHub, Notion, Shopify and Slack. It presented potential roles such as project manager, inventory manager and marketing consultant. Those are suggested applications, not evidence that the product can replace the corresponding human positions or reliably perform every responsibility associated with them. The distinction is especially important when assessing a broad product launch.

The practical change Google is proposing is a move from separate prompts in separate applications toward delegated objectives with shared context. Its materials describe the architecture and intended controls, but they do not establish universal productivity gains for customers. For prospective users, the announcement supplies a product framework to evaluate: which systems an agent can reach, which models it uses, how spending is limited and what permissions govern its actions.