Google Cloud Launches Gemini Agent to Automate Enterprise Work Across Any Platform

Google LLC's Cloud division unveiled Gemini agent today, positioning it as a unified artificial intelligence assistant with the ability to work autonomously, write code and execute tasks across multiple devices and platforms. The system operates without requiring a dedicated interface, functioning instead through the command line, Google Workspace, Microsoft 365, Slack and third-party applications wherever users need it.

Thomas Kurian, chief executive of Google Cloud, described the assistant's operational model: "You give it objectives, not instructions. You delegate an outcome and come back to finished work." He emphasized that for agents to function effectively, they must integrate with personal workflows, systems of record and enterprise controls. The system can answer questions, manage knowledge work, generate images and media, and transform concepts into completed deliverables.

The platform embodies persistent execution, maintaining a single repository of memories, context and personalization across all communication channels. Gemini can instantiate smaller sub-agents with distinct identities and expertise, each with dedicated @agents.company.com email addresses and isolated persistent storage. These sub-agents access only the context explicitly provided by users or team members.

Google Cloud emphasized flexibility in model selection, with each task running on the most appropriate model. Straightforward assignments execute on Gemini Flash, a lightweight model designed for routine work, while complex long-horizon projects utilize Argon, a flagship frontier model. Anthropic PBC's Claude models are available immediately, with additional leading private and open-source models coming later.

Skills, Tools and Institutional Knowledge

Kurian emphasized Google's approach to providing comprehensive business context understanding, encompassing pricing structures, product portfolios and departmental conventions. Gemini absorbs and surfaces the breadth of institutional knowledge that distinguishes organizations, anchoring responses through company-specific data.

The agent securely integrates with collaboration platforms including Confluence, Microsoft Office, Teams, Slack and Workspace. It also connects to development tools such as Git and Jira, enterprise systems including Salesforce and ServiceNow, data warehouses such as BigQuery, Databricks, Postgres and Snowflake, and local desktop files.

Users can develop reusable skills functioning as instructions, knowledge bases and workflows that teach agents to execute specific multistep processes. While Gemini ships with a global library of prebuilt skills, organizations can create custom skills and publish them to a shared company registry. Users can also instruct agents to convert completed tasks into reusable templates.

The system learns continuously from every question posed and objective assigned. It maintains session memory across extended periods, conversations with multiple people and interactions with other agents while processing documents. Gemini continuously refines its understanding of user workflows and expected outputs, dedicating as much effort to observing user behavior as to tool utilization.

Google has developed domain-specific skills, beginning with data-focused capabilities. The company is extending Gemini with machine learning tools for data scientists and engineers. When users describe desired outcomes in natural language, Gemini generates PySpark code, supplies notebooks for editing and testing, trains models and resolves issues independently.

For operational teams, Gemini leverages BigQuery reporting skills and the Knowledge Catalog to construct and preserve queries. Organizations can generate real-time operational reports through simple requests. Once saved, teams execute reports without incurring additional token expenses, consistently delivering verified results.

Spending Controls and Cost Management

Google Cloud highlighted that despite per-token pricing declining approximately 98% since 2024, enterprise AI consumption has surged substantially. The company announced flexible spending options in August, which are now becoming available.

Internally, multimodel orchestration selects the optimal model for each task, distributing complex assignments across quick models for minor components and frontier models for challenging elements, thereby reducing expenses. Smart routing automatically directs enterprise workloads to the model delivering maximum performance at the lowest cost.

Users can establish hard spending limits through the Cloud Billing Console. Gemini monitors token consumption and pauses when a cap is reached. Users can resume operations through the console with approval, though doing so requires acknowledging the budget overage.

Since tracking operates at the project level, organizations can allocate AI expenses to specific departments. This capability enables businesses to audit and forecast budgets by team and department, facilitating more detailed cost allocation and planning.