Google introduces Gemini agent as a universal AI agent for enterprise work

Google Cloud has introduced the Gemini agent at Gemini at Work 2026 as a single universal agent for enterprise work. It is designed to work across business applications, data and other AI agents while retaining context across tasks and devices.

Google said nearly 500 Cloud customers processed more than 1 trillion tokens each over the past year, while nearly 80% of its customers now use AI products. Nearly 90% of Fortune 100 companies use Gemini Enterprise.

Gemini agent architecture

The Gemini agent is built around six principles:

Coworker agents can have @agents.company.com identities, persistent storage and access only to the context shared with them.

Tools, skills and memory

Gemini connects with enterprise systems including Confluence, Microsoft Office, Teams, Slack, Google Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake. It can also work with desktop files and MCP servers.

A tools registry manages available tools, while a skills registry lets organizations create reusable capabilities and personal skills.

Gemini in Workspace

Gemini extends into Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar through three modes:

For example, a user can ask an agent in Chat to schedule a meeting based on a conversation, while a coworker agent can coordinate an event and update documents through comments.

Data and AI workflows

Gemini can work with enterprise data and development environments, including PySpark, notebooks and model training. Agents can also troubleshoot systems and make fixes autonomously.

The Borderless Lakehouse supports sources including Amazon S3, Azure Data Lake, Salesforce Data 360, SAP, ServiceNow, Workday and Apache Iceberg across Databricks Unity, Snowflake Polaris and AWS Glue. Google said about 90% of enterprise data is unstructured.

Industry capabilities

Google is also adding industry-specific capabilities with specialized data sources, skills and workflows.

Financial Services and Legal capabilities are in preview. CME Group, Deutsche Bank and Cooley are among organizations using related capabilities.

Security, governance and AI costs

Gemini Enterprise includes controls for managing agent access, data usage, security and AI spending across organizations.

• Security and governance: Controls cover identity, authorization, permissions, auditing and policy management. Agent Sandbox, Agent Gateway and OAuth help manage how agents interact with systems and data.
• Need to Know: Agents can be limited to the information required to complete a specific task.
• AI costs: Google said the price of AI processing per token has fallen 98% since 2024. Multi-model orchestration and Smart Routing can select models based on different tasks.
• Spending controls: Organizations can set real-time spending limits, monitor usage through the Cloud Billing Console and apply project-level chargeback.

Models and infrastructure

Gemini Enterprise can orchestrate different AI models depending on the task, including Google’s Gemini family and models such as Argon, Flash, Omni and Gemma. Google is also continuing to improve the infrastructure that supports these workloads, with its TPU 8i delivering up to 80% better price-performance than the previous generation.

• Model flexibility: Supports multiple models and multi-model orchestration for different workloads.
• TPU 8i: Delivers up to 80% better price-performance than the previous generation.
• Gemma on satellites: NASA’s Jet Propulsion Laboratory is running Gemma on a satellite, where its NAVI system analyzes Earth imagery and sends summaries.

Gemini in India

Google highlighted Indian organizations as part of its Asia Pacific customer deployments:

• Tata Steel: Deployed more than 300 specialized AI agents within nine months for manufacturing safety, predictive supply chains and customer complaints, reducing complaint turnaround time by 50%.
• Tata Consultancy Services (TCS): Named among Google’s Asia Pacific customers using Gemini Enterprise.
• Marubeni: Also named among Google’s Asia Pacific customers using Gemini Enterprise.

Customer deployments

Google also highlighted deployments across other regions:

• BNP Paribas: Uses LLM@CIB, an internal generative AI assistant supporting more than 65,000 employees with credit memo preparation and customer advisory tasks.
• Bradesco: Uses Gemini for fiscal, accounting and contractual analysis, reducing document review time from one hour to five minutes and risk inconsistencies by 60%, while delivering more than 10% financial efficiencies.
• Orange Spain: Has deployed more than 1,000 custom agents across HR, IT, sales and customer service, including incident triage, network analysis and knowledge retrieval.
• SOMPO: Has deployed more than 10,000 custom agents for 34,000 employees, reducing model development time from one week to one day.
• Ulta Beauty: Uses a shopping assistant covering 30,000 products, resulting in 3x higher digital sales conversion.
• Wesfarmers: Uses shopping and business agents across Bunnings, Kmart and Officeworks, with the Bunnings agent saving 500,000 administrative hours and shopping agents delivering up to 3x higher conversion.

Gemini ecosystem and availability

Google is expanding Gemini Enterprise through third-party agents, connectors and the AI Solution Finder, while also adding consulting and implementation partners.

• Partner ecosystem: More than 100,000 consulting developers and consultants participated in Gemini hackathons over the past month. Accenture has also formed a Gemini Enterprise Business Group with Google Cloud.
• Industry capabilities: Financial Services and Legal are currently in preview, while Government, Healthcare and Retail capabilities are planned.
• Model expansion: Google plans to add support for private and open models alongside its existing model choices.
• Enterprise rollout: The Gemini agent was introduced as part of Google’s Gemini at Work 2026 enterprise updates.