Google Launches Gemini Agent for Businesses

Google deploys a unified agent in Gemini, initially for enterprise customers. The company emphasizes governance, integrations with existing systems, the freedom to choose third-party models, and new budgetary safeguards.

Internal Governance and Traceability Announced by Google

The agent will have its own Workspace account, complete with a distinct email address and context, similar to a colleague. It is aware of team compositions, time zones, individuals responsible for approvals, and the schedules of collaborators. Users will be able to engage the agent by mentioning it, sending it an email, sharing with it, or adding it to a group discussion. Each action taken by the agent will generate an audit trail attributed to it, rather than to an individual. A mailbox-style interface will allow users to track Gemini's activity, including its reasoning, delegation to sub-agents, loading of skills or code, and task progress.

Connections to Enterprise Tools and MCP Servers

The agent can connect to the company's data and systems, including Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake. Queries directed at the agent can include attachments such as files, folders, or projects tailored to specific workflows, including those that combine files and skills. It can also securely work with any Model Context Protocol server, whether inside or outside the organization's network. Google anticipates that the agent will be accessible from iOS and Android devices, Windows and Mac computers, the command line interface, as well as through Google Workspace, Microsoft 365, ServiceNow, and Slack.

Adoption and Use Cases for Enterprises

Sundar Pichai notes that Gemini has over 1 billion monthly active users and that nearly 90% of Fortune 100 companies use Gemini Enterprise. On, Shopify, and PayPal are among the first companies to have tested the agent, while BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers are mentioned as clients of Gemini Enterprise. AI tools are now evolving towards functions that go beyond simple conversation, with agents capable of writing code, organizing meetings, scheduling appointments, and booking travel. On the consumer side, new agents like Meta's Muse, Instinct via messaging, and ChatGPT's Dots have been recently launched.

Choice of Models, Business Objectives, and Budget Framework

Thomas Kurian specifies that the new agent is capable of receiving objectives, organizing tasks, employing custom skills and tools, as well as connecting to the company's internal systems to accomplish its missions. By default, the AI opts for the model deemed most relevant, but users can still select another one, including external models like Claude from Anthropic. Google intends to add the option to use open-source models and other proprietary models in the future. The company also presents new mechanisms to adjust spending, such as multi-model orchestration, intelligent routing, or real-time spending caps, to help businesses better manage their AI-related costs. Google plans to offer the Gemini agent first to enterprises, then later open it to the general public, a move that Sundar Pichai associates with the need to address security, scalability, and performance issues. The agent is designed to answer questions and perform tasks on behalf of the user from a unified interface, as announced during a Google Cloud event.