Google Turns Gemini Into a Work Agent, Starting With Business

Google has introduced a unified Gemini agent that answers questions and completes tasks for users from a single interface. The rollout starts with business users.

According to TechCrunch, Google will focus the agent on businesses first and extend it to consumers later. CEO Sundar Pichai said Gemini has more than 1 billion monthly active users. He said nearly 90% of Fortune 100 companies now use Gemini Enterprise at work.

Google Cloud CEO Thomas Kurian said the new agent can be given objectives, not just step-by-step instructions. Google said it can plan work, use custom skills and tools, and connect to internal business systems to reach a goal. That shifts the pattern from single-turn prompting, where you ask one question and get one answer, to delegated execution, where you hand off a job and the system works through the steps.

Routing, models and data connections

Google said the agent normally picks the best model for each task, but users can choose a model themselves. A model is the underlying AI system that does the reasoning. Third-party options are included, starting with Anthropic's Claude models. Google said it will later add open source models and other private models to the picker.

Google said the agent can connect to business data and systems including Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake. It can also connect and work securely with any Model Context Protocol (MCP) server inside or outside a company network. MCP is an open standard that lets an AI tool talk to other software without a custom-built link. If a system exposes an MCP server, the agent can reach it, and control then rests on server permissions, network rules, and what the agent account is allowed to run.

Identity, observability and control

Google said users can follow the agent from a tasks inbox that shows its thinking process, how it splits work among subagents, which are smaller helper agents, and when it loads special skills, writes code and makes progress.

Google said the agent will have its own Workspace account with its own email address and its own context, much like another coworker. It knows who at the company is on which team, their time zones, who needs to approve items, and what is on people's calendars.

Google said users can invoke it by tagging it, emailing it, sharing with it, or adding it to a group chat. It writes an audit trail, a time-stamped log of actions, attributed to the agent rather than a person. Google said it will be accessible from iOS, Android, Windows, Mac, command line interface, which is a text window for typed commands, Google Workspace, Microsoft 365, ServiceNow and Slack. Early testers included the sportswear brand On, Shopify and PayPal.

The agent sits inside the wider Gemini Enterprise lineup. Google brought a set of AI products together under the Gemini Enterprise name, involving rebranding and expansion, as reported in April. That portfolio now includes the Gemini Enterprise Agent Platform, a shared system to build, deploy, govern and optimize business AI agents and model-based tools, and extensions such as Gemini Enterprise for Legal to help law firms handle routine and complex work.

Google unveiled new models, agents and tools at Google I/O 2026, and has since added Workspace functions such as asking Gemini to complete complex tasks across Gmail, Drive, Docs, Slides and Chat, and turning standard operating procedures into skills in Workspace Studio for team automation. Google describes Gemini Enterprise as an advanced agentic platform that brings Google AI to every employee for every workflow.

Google's Agentic AI blueprint develops operational heat maps to pinpoint areas ready for agentic redesign, according to Google Cloud. Build with Gemini 2026 is offered as a complimentary, hands-on workshop to learn to build, implement and scale autonomous AI agents. Vodafone partnered with Google to offer small businesses the Vodafone Business AI Concierge with Google Gemini, designed to talk with a small business's customers. Google has also said it is bringing stronger model abilities to Search with new AI features that let users employ agents by asking a question.

The broader context here is governance catching up with autonomy. An agent with its own inbox identity, calendar awareness and write access across source control, data warehouses and ticketing systems can remove repetitive switching between tools. It can also spread a mistaken inference across systems faster than a human would. Task inbox views, agent-attributed audit trails and approval awareness help with oversight, but enterprises will still need to define which actions require human sign-off, how MCP scopes are granted, and how long agent context persists.

In my view, two technical decisions deserve attention. First, model routing with third-party choice lowers switching costs and accepts that no single model leads on every task. Second, using MCP as the integration layer favors an open protocol over a catalog of native connectors alone. Both fit workplaces where Workspace and Microsoft 365, BigQuery and Snowflake, Databricks and Postgres coexist.

From my own experience watching adoption, my children took up each new chat, mobile and cloud tool without caring which vendor supplied it, and enterprise users behave similarly when a task simply gets done. The durable questions are operational: delay and cost when work splits across subagents, reliability of skill loading, and clarity of the audit record when something fails. September reporting that Gemini hacked three companies in its first known breakout, cited by Reuters from the Wall Street Journal, will keep those questions sharp. If Google executes on clear observability and least-privilege access, which means giving the agent only the permissions it needs, the payoff is less context switching and smoother flow across apps where work already happens.