Google Cloud has unveiled Gemini agent, an autonomous AI assistant created to complete complex business tasks across enterprise applications, with support for multiple AI models and new tools to control spending.
The launch expands Google’s agentic AI strategy, moving far beyond chatbots that respond to prompts toward unified systems that can independently execute workflows. Gemini agent can write software, analyze corporate data, produce documents and perform other advanced business tasks.
A central feature is Gemini agent’s ability to operate across competing software platforms. It supports Google Workspace, Microsoft 365 and Slack, along with enterprise applications including Salesforce, ServiceNow and Jira.
The system also connects to corporate data platforms like Snowflake, Databricks and, of course, Google’s BigQuery. This broad compatibility is important because large enterprises typically rely on applications from numerous vendors, creating challenges for AI systems that need access to company information.
Gemini agent maintains a continuing record of user preferences, previous tasks and business context. This enables their work to proceed across devices and applications without requiring users to repeatedly explain their requirements.
One significant Gemini capability enables users to deploy smaller agents with specialized roles. These digital coworkers can receive individual email addresses, maintain separate storage and collaborate on multistep assignments. Access permissions limit each agent to authorized information, which is essential for enterprises managing sensitive corporate data.
Google has also designed Gemini agent to support competing AI models rather than relying exclusively on its own technology. The platform currently supports Google’s Gemini models and Anthropic’s Claude, with additional proprietary and open models planned. This flexibility allows organizations to select models based on price, performance requirements, and workloads.
Google is also providing reusable skills that enable agents to perform specialized workflows. For example, data scientists can request machine learning operations using natural language, prompting Gemini to generate PySpark code, train models and troubleshoot problems. Business users can generate operational reports through BigQuery without manually writing database queries.
Gemini agent is currently available in private preview for enterprise customers.
Containing Costs
A major focus of the launch is containing AI expenses as businesses expand deployments. Google reports that token prices have declined approximately 98% since 2024, although greater usage continues to create pressure on enterprise budgets.
To address this, Gemini agent uses automated model selection to assign less demanding tasks to lower-cost models while reserving more powerful systems for complex work. Admins can also establish spending limits through Google Cloud’s billing console. When an agent reaches its assigned budget, processing stops unless additional spending gets approved.
Furthermore, project-level tracking allows companies to allocate AI expenses to individual departments. The platform also includes identity management, auditing, authorization and policy controls created to support corporate governance requirements.
Early customers include Shopify, PayPal and sportswear company On. Google is also previewing specialized versions for financial services and legal applications, with healthcare, government and retail offerings planned.
