Google launches unified agentic Gemini that plans and completes work for businesses

Gemini gains an agent

At a Google Cloud event on Thursday, Google announced that Gemini is moving beyond conversation: a single unified agent can now take on assignments and complete them from one interface, according to TechCrunch. The debut lands as AI products across the industry shift from chat toward software that owns a task end to end — writing code, scheduling meetings, arranging travel — and it follows the arrival of consumer agents such as Meta's Muse, messaging-based assistants like Instinct, and ChatGPT's recent Dots.

Google is bringing the agent to business customers first, with a consumer rollout to follow. CEO Sundar Pichai told the event that Gemini has more than 1 billion monthly active users and that nearly 90% of Fortune 100 companies already run Gemini Enterprise at work. He framed the enterprise-first sequencing as a way to work through what he called the "harder problems around security, scale, and performance" that capable agents raise.

Objectives, not instructions

Google Cloud chief Thomas Kurian said the agent accepts "objectives, not just instructions." In practice, it plans its own work, invokes custom skills and tools, and reaches into a company's internal systems to finish the job. Requests can carry attachments such as files, folders, or project bundles that pair documents with skills. Out of the box it connects to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake and other systems, and it can work securely with any Model Context Protocol (MCP) server inside or outside the corporate network.

Treated like another employee

The agent gets its own Workspace account, complete with a dedicated email address and its own working context. Google says it knows who sits on which team, their time zones, whose sign-off a decision needs and what colleagues' calendars hold. Users can reach it by tagging it, emailing it, sharing files with it or adding it to a group chat. Every action it takes is written into an audit trail attributed to the agent itself rather than to a person — a detail aimed squarely at corporate governance.

A window into the machine

A "tasks inbox" lets users follow what Gemini is actually doing: its reasoning, how it delegates work to subagents, which skills it loads, the code it produces and its overall progress. The agent will be available on iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365, ServiceNow and Slack.

An open model picker

By default the agent picks whichever model best fits the task, but users can step in and choose — including third-party models, starting with Anthropic's Claude. Google said open source and other private models will join the picker later. On the cost side, the company is introducing flexible spending options such as multi-model orchestration, smart routing and real-time spend caps so enterprises can keep AI bills under control.

Early testers and customers

Sportswear maker On, Shopify and PayPal were among the early testers. Google also pointed to existing Gemini Enterprise customers including BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers.

Why it matters

This is one of the clearest signals yet that the big platforms are repositioning their flagship assistants from chat windows into autonomous workers, and Google is starting where budgets and compliance demands are highest: the enterprise. Three details stand out. Giving the agent its own account and audit trail treats accountability as a core feature — something regulated organisations will insist on before letting software act with minimal supervision. Supporting MCP servers and third-party models, with Claude first and open source promised, positions Gemini as an orchestration layer rather than a locked path to Google's own models. And with a reported billion-plus users and most of the Fortune 100 already on Gemini Enterprise, Google has distribution few rivals can match if the agent performs as advertised. The open question is reliability: executing objectives across business-critical systems is a far higher bar than answering questions, and the business-first rollout suggests Google knows it.