Google is officially taking Gemini past the chat window and putting it to work. At its Cloud event on Thursday, the tech giant launched a unified autonomous agent capable of executing complex jobs across corporate software on an employee’s behalf, from writing code and scheduling meetings to pulling records across disparate databases.
The rollout marks a pragmatic evolution in enterprise software: instead of having staff copy and paste answers back and forth between apps, the agent directly interacts with company tools to see projects through to completion.
Enterprise-First Deployment Strategy
Google plans to establish the technology within corporate environments before considering a broader consumer release. During the presentation, CEO Sundar Pichai noted that while Gemini has grown to over 1 billion monthly active users, nearly 90% of Fortune 100 companies currently run Gemini Enterprise.
Deploying to businesses first provides Google with a controlled environment to tackle the substantial security, scalability, and performance hurdles that autonomous software introduces. Early access testing is already underway at companies including Shopify, PayPal, and On, alongside active enterprise accounts like Merck, BNP Paribas, and Ulta Beauty.
Working Toward Objectives, Not Prompts
Google Cloud CEO Thomas Kurian emphasized that the agent is designed to run on high-level operational targets rather than micromanaged instructions. Once handed a goal, the system independently outlines the required steps, selects appropriate skills or tools, and accesses connected corporate platforms to complete the task.
Native integration covers environments like Google Workspace, Microsoft 365, Slack, Jira, Confluence, GitHub, BigQuery, Databricks, Postgres, and Snowflake. Employees track active operations via a “tasks inbox” that provides direct visibility into the agent’s reasoning, code execution, and delegation to specialized subagents.
Embracing Anthropic’s Claude and Open Protocols
A central highlight of the announcement is Google’s move toward open model orchestration. Rather than locking businesses into the Gemini ecosystem, the agent features a built-in model picker. While Gemini handles automated routing by default, administrators and users can manually choose external third-party models to execute tasks, beginning with Anthropic’s Claude family. Google confirmed that this model selector will soon expand further to include private enterprise models and open-source architectures.
Alongside third-party model support, Google integrated compatibility with the open-standard Model Context Protocol (MCP). By supporting MCP, the agent can communicate directly and securely with external tools and proprietary corporate servers without requiring custom, vendor-specific middleware.
Coworker Infrastructure and Financial Controls
Structurally, Google is embedding the agent directly into organizational workflows. Each agent receives a functional Workspace identity, including a company email address, calendar awareness, and access to org charts and team time zones. Staff can engage the agent by tagging it in team chats, sending an email, or sharing documents directly with its account. Every action generates a verifiable audit trail tied to the agent’s credentials, not individual user accounts.
Cross-platform access includes iOS, Android, desktop applications, the command-line interface, and enterprise services like ServiceNow. To maintain operational control and model diversity, Google integrated multi-model routing and administrative guardrails:
| Capability | Technical Specification | Operational Use |
| Model Selection | Automatic optimal model routing, with manual override to Anthropic’s Claude; open-source models planned | Gives teams access to specialized reasoning while eliminating single-vendor lock-in |
| Protocol Support | Native Model Context Protocol (MCP) compatibility across corporate firewalls | Facilitates secure, standardized data exchange with existing tools and servers |
| Workplace Identity | Dedicated corporate email address, calendar access, and directory context | Ensures every automated change is accountable through distinct audit logging |
| Spend Controls | Multi-model orchestration, dynamic routing, and real-time spend limits | Prevents unexpected cloud and API billing spikes during large automation tasks |
Google stated that these multi-model capabilities will be paired with flexible spend caps and smart routing, ensuring companies can deploy autonomous agents across Claude, Gemini, and open models without losing control over operational costs.