Google is expanding Gemini from a question-and-answer AI into agentic AI that performs real work. TechCrunch reported on Oct. 8 that Google unveiled an integrated agent at a Google Cloud event that not only provides answers but also handles tasks for users. Google plans to apply it to enterprises first.

Google CEO Sundar Pichai (순다르 피차이) said Gemini has more than 1 billion monthly active users and that about 90 percent of Fortune 100 companies use Gemini Enterprise for work. He added that ahead of launching powerful agents, the company must resolve tougher issues related to security, scalability and performance.

Google Cloud CEO Thomas Kurian (토마스 쿠리안) said the new agent can receive goals as well as instructions. He said Gemini can draw up work plans, use tailored skills and tools, and connect with internal corporate systems to carry out tasks needed to achieve those goals.

The Gemini agent generally has AI choose the model best suited to a task, but users can also select a model themselves. It will initially support Anthropic's Claude model. Google plans to add open-source models and other proprietary models later.

The agent can also handle requests with attached files, folders and projects for specific workflows. It connects with Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake. It also supports secure connections to Model Context Protocol (MCP) servers on networks inside and outside a company.

Users can check Gemini task status in a 'task inbox'. The screen shows the reasoning process, tasks assigned to sub-agents, calls for specialized skills and code, and progress updates. Gemini has a separate Workspace account and email address, and operates based on information such as team composition, time zone, approvers and schedules. Users can call an agent by tagging, emailing, sharing materials or adding it to group chats, and work records remain as an audit trail under the agent's name rather than an individual's.

Gemini agents can be used on iOS and Android devices, Windows and macOS desktops, command-line interfaces, Google Workspace, Microsoft 365, ServiceNow and Slack. Google also plans to provide multi-model orchestration, smart routing and real-time spend cap features so enterprises can control AI costs more flexibly.