Google Cloud launches Gemini agent as a single, universal agent for work

One agent across devices and channels

The announcement was made at Gemini at Work 2026 and adapted from a keynote by Thomas Kurian. Google Cloud says the Gemini agent plans the work, uses skills and tools, connects to company systems and returns finished results inside documents, the inbox and developer environments. The company says it also has built-in cost controls and the security, administration and governance a company requires.

Google describes the model as delegating objectives rather than instructions. Gemini can act as a personal assistant or as a team member, for example a project manager within a team or an analyst in a finance department. Users can assign work, schedule tasks or have it respond to events.

Access is meant to be broad. According to Google Cloud, Gemini can be reached from web, iOS and Android, Windows and Mac desktops, and through the command line, Google Workspace, Microsoft 365 or Slack. It can also be built into third-party applications and run as a headless agent without a dedicated user interface. The company says Gemini works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar.

Because it runs in the cloud, Google says the agent keeps one set of memories, context and one personalization graph regardless of device. Work that takes hours or days continues after a laptop is closed.

Sub-agents, coworker agents and model choice

Google says Gemini can dynamically create a roster of sub-agents, described as temporary, job-specific agents, each with its own identity. They can coordinate parallel and sequential steps that run for hours or days. Gemini can also act as a coworker agent with a persistent, defined role. According to the company, coworker agents have dedicated identities, including their own agent email addresses and persistent storage, and only access context provided by users or team members.

The agent is separate from the model underneath it. Google says Gemini runs each job on the model that fits best, orchestrating across its Gemini family and Claude models from Anthropic today, with other leading private and open models planned for the future. The company argues that the best model is not always the largest, and that matching model to task raises accuracy on difficult jobs and lowers cost on simple ones.

Google says the sportswear brand On tested this dynamic selection capability to speed up time to market. It also pointed to Shopify, which blends frontier models for millions of merchants, and PayPal, which routes 10 million multi-model requests every week.

Tools, skills and memory

Google lists three building blocks: tools, skills and context. Gemini can connect to software including Confluence, Microsoft Office, Teams, Slack, Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, as well as files on a user's desktop. It can work with any Model Context Protocol (MCP) server inside or outside a company network. An enterprise tools registry lets teams build and publish tools for the rest of the company.

Skills are reusable sets of instructions, knowledge or workflows stored as modular prompts. Gemini ships with a global library, and teams or departments can publish custom skills to a shared registry. Google says Gemini chooses the right skills and tools for a task and learns from each execution, which improves consistency and saves token costs.

The agent keeps four kinds of memory, according to Google: session memory for the current task, semantic memory as a structured knowledge base, procedural memory for how a job is done, including skills it writes for itself, and episodic memory of past actions.

Scale, governance and what it means for businesses

Google Cloud says nearly 500 of its customers each processed more than one trillion tokens in the last year, nearly 80% of its customers use its AI products, and nearly 90% of the Fortune 100 use Gemini Enterprise. From this, the company concludes that organizations have moved past experimentation.

The announcement also covers new data and analytics skills that let technical and business users ask plain-language questions to get operational insights in minutes, industry-specific tools and connectors for financial services and legal teams, and security and governance features including identity and policy management, permission controls, secure sandboxing and network gateways. Cost controls include multi-model orchestration, Smart Routing and real-time spend caps.

As a customer example, Google cites BNP Paribas, which is deploying Gemini Enterprise into LLM@CIB, an internal generative AI assistant available to more than 65,000 employees, for tasks such as preparing credit memos. For enterprises, the pitch is a single agent that keeps context across tools and lets them switch underlying models without losing their data and skills.