Google has taken a significant step toward agentic artificial intelligence with the introduction of a unified AI agent built into its Gemini platform, announced at a Google Cloud event on Thursday. Unlike traditional AI assistants that simply respond to questions, this new capability is designed to act on behalf of users - completing tasks, managing workflows, and interfacing with external systems, all from within a single interface.

The announcement reflects a broader shift happening across the AI industry, where the focus is moving away from purely conversational tools toward systems that can independently handle assigned responsibilities. These include generating code, scheduling meetings, booking travel, and managing appointments. The trend has gained momentum alongside the emergence of consumer-oriented agents from companies like Meta and others operating through messaging platforms, as well as OpenAI's recent introduction of its own agent-oriented features within ChatGPT.

Google enters this space with a notable advantage in reach. As CEO Sundar Pichai highlighted at the event's opening, Gemini currently serves more than one billion monthly active users. He also pointed out that close to 90 percent of Fortune 100 companies are already using Gemini in enterprise settings, giving the company a strong foothold from which to scale its agentic ambitions.

Business Users Come First

Rather than launching the new agent capabilities simultaneously to all users, Google has chosen to prioritize its business customers. According to Pichai, this phased approach will give the company time to work through the more demanding challenges related to security, performance, and scale before a broader consumer rollout takes place.

Thomas Kurian, CEO of Google Cloud, described the agent's design philosophy in practical terms. He explained that the system can be given objectives rather than step-by-step instructions, allowing it to independently plan its own approach, draw on specialized tools and skills, and tap into a company's internal infrastructure to see tasks through to completion.

Model Flexibility and Third-Party Integration

By default, the agent selects the most appropriate AI model for each task automatically. However, users retain the option to manually choose a model themselves, including models from outside Google's own ecosystem. At launch, third-party model support begins with Anthropic's Claude models, with plans to expand the model picker to include open source options and additional private models going forward.

Requests submitted to the agent can include a range of attachments - files, folders, or purpose-built project packages that combine specific files and skill sets relevant to particular workstreams. The agent is also capable of connecting directly to a company's data sources and enterprise platforms. Supported integrations at launch include Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake, among others.

The agent can additionally interface securely with any Model Context Protocol (MCP) server, whether hosted inside or outside a company's network. MCP is an emerging standard that allows AI models to connect with external tools and data sources in a structured, interoperable way.

Transparency Through a Tasks Inbox

To keep users informed about what the agent is doing at any given moment, Google has introduced a dedicated tasks inbox interface. Through this view, users can follow the agent's reasoning process in real time, observe how it delegates work to specialized subagents, monitor the loading of particular skills, and track progress across active tasks.

One of the more distinctive aspects of the new agent is that it operates with its own Google Workspace account, functioning much like a regular employee within an organization. It has a dedicated email address and its own context layer - meaning it is aware of company structures, team assignments, employee time zones, approval hierarchies, and calendar availability. Users can interact with the agent by tagging it directly, sending it an email, sharing documents with it, or adding it to a group chat. Any actions it takes are logged in an audit trail attributed to the agent itself, keeping a clear record that distinguishes its activity from that of human users.

Broad Platform Availability

Access to the agent will be available across a wide range of platforms and devices. Users will be able to work with it on iOS and Android mobile devices, Windows and Mac desktops, through the command line interface, and via Google Workspace, Microsoft 365, ServiceNow, and Slack.

Google noted that several high-profile companies participated in early testing of the agent, including sportswear brand On, e-commerce platform Shopify, and payments company PayPal. Among the existing Gemini Enterprise customer base, Google cited major organizations including BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers.

Cost Management for Enterprise AI

Alongside the new agent capabilities, Google announced updated spending options aimed at helping businesses manage the financial side of enterprise AI adoption more effectively. These include multi-model orchestration tools, intelligent routing that directs tasks to the most cost-efficient models, and real-time spending caps that give organizations greater visibility and control over their AI expenditure as usage scales.