Google announces Gemini agent: A universal AI for enterprise work

Google Cloud CEO Thomas Kurian announced the Gemini agent at the Gemini at Work 2026 keynote, positioning it as a single, universal agent for business tasks. Unlike standard interactions where users provide instructions, this agent accepts objectives and returns completed work products, such as code, reports, or media, by planning tasks and using connected tools.

Architecture and Capabilities

The agent is built on a unified interface that integrates chat, autonomous task execution, and code generation. It supports omnipresent access across web, iOS, Android, Windows, Mac, CLI, Google Workspace, Microsoft 365, and Slack. A key architectural feature is persistent execution, which allows long-running tasks to continue in the cloud even if the user's device is offline, maintaining a single memory and context graph.

The system employs multi-agent orchestration to create dynamic sub-agents for parallel or sequential workflows. It also introduces the Coworker agent, a persistent role with its own email address and repository access, designed to collaborate with human teams. The agent supports model choice flexibility, allowing users to orchestrate Gemini models alongside Anthropic's Claude models to balance accuracy and cost.

Integration and Context

The Gemini agent integrates deeply with enterprise tools, including Confluence, Microsoft Office, Teams, Slack, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, and Snowflake. It utilizes a skills library of modular prompts for reusable workflows and maintains four types of memory: session, semantic, procedural, and episodic. This allows the agent to learn from interactions and onboard new users by understanding personal and organizational context.

Market Adoption and Performance

Google reported that nearly 80% of Google Cloud customers use AI products, and nearly 90% of Fortune 100 companies use Gemini Enterprise. Over the past year, nearly 500 customers each processed more than 1 trillion tokens. Customer case studies highlight significant efficiency gains:

• Bradesco: Reduced document review time from 1 hour to 5 minutes and risk inconsistencies by 60%.
• SOMPO: Built over 10,000 custom AI agents for 34,000 employees, cutting new model development time from 1 week to 1 day.
• Ulta Beauty: Achieved a 3x increase in digital sales conversion using a Gemini Enterprise-based shopping assistant.
• Wesfarmers: Saved 500,000 hours of administrative work using internal agents at Bunnings.

Governance and Infrastructure

Security features include unique encrypted IDs for every agent, role-based access control (RBAC), and an Agent Gateway that acts as an AI network firewall. Cost controls allow for real-time spend caps and smart routing to optimize token usage. The infrastructure relies on the AI Hypercomputer, with the new TPU 8i offering an 80% improvement in price-performance ratio over the previous generation.