Google launches Gemini enterprise agent that integrates Gmail, Docs and Slack to handle work and schedule meetings by itself

Google Cloud launched the Gemini enterprise agent at the Gemini at Work event on October 8, which allows employees to assign tasks through natural language. AI arranges the execution steps, calls the enterprise system, and sends the completed documents, presentations or emails back to the original work environment. This service integrates Google Workspace, Microsoft 365, Slack and multiple enterprise data platforms, and can also continue work across different devices. Thomas Kurian, CEO of Google Cloud, said that users only need to explain the goal and do not have to instruct the agent step by step on how to complete the task.

This time, Google will bring the proxy function directly into the office software that employees are familiar with. In Gmail, Docs, Sheets, Slides, and Chat, users can call agents via @Gemini. For example, when a supervisor requests to update the latest project briefing, Workspace can recognize that this is a task that can be delegated and provide a one-click assignment option. If you want to schedule a meeting with regional event leaders next week, agents can find relevant people from chat rooms and past discussions, check the calendar, and send an email to coordinate the time.

Please read on...

Gemini Enterprise Agent also provides a unified prompt input field for users to handle Q&A, document creation, program development and other tasks. By default, the system selects models based on task requirements. Flash can be used for simple tasks, while Argon can handle longer tasks. Users can also specify their own models. The first wave of third-party options includes Anthropic Claude, and will be expanded to open source and enterprise private models in the future.

Enterprise profile connections include Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake in addition to Google Workspace, Microsoft 365, and Slack. Users can directly attach files and folders, and can also organize commonly used files and skills into projects for agents to use when performing tasks. For companies that have scattered documents, communications, and databases in different systems, agents can obtain authorization information from the same work portal, reducing the need for employees to repeatedly switch applications.

Google also keeps the Gemini agent working across devices. Tasks are executed in the cloud, and memory and progress can be extended to web pages, mobile phones, desktops and command lines. Users can still continue operations after switching to another device. Enterprise developers can embed the agent into existing software through APIs to provide functions as a headless agent that does not require a separate operating interface.

The system can also create sub-agents with independent identities. Enterprises can configure a dedicated account and @agents.company.com email address for the agent so that it can participate in work communication, but it can only access data within the authorized scope. Memory functions are divided into two categories: procedural and episodic. The former saves the way work is performed, including the skills established by the agent itself; the latter records previously performed tasks. Agents can therefore continue to use existing processing flows without having to re-obtain complete instructions each time.

Google hopes to use this to integrate AI agents into the daily work of enterprises. Employees can assign tasks in existing files or chat rooms, while agents collect data, use tools, and produce results. If the company needs a more detailed division of labor, it can also create sub-agents with different identities to handle work according to their respective authority. This design is different from the way ordinary employees use chatbots directly. Enterprises must manage agent identities, data access scopes, and execution permissions at the same time.

A variety of products have entered the enterprise AI agent market. OpenAI's ChatGPT Work provides work delegation capabilities such as document analysis, presentation production, and report writing. Dots uses cloud computers, browsers, and plug-ins to continuously perform tasks. When necessary, program development work can be handed over to Codex. Dots announced a phased opening to Pro and eligible Business Premium users on September 29. Anthropic is developed with the Claude Code handler, and Cowork is aimed at non-engineering users who need to perform file work in authorized folders.

Self-deployed open source solutions include OpenClaw and Nous Research’s Hermes Agent. Both belong to the agency framework and can connect different models according to needs. OpenClaw Enterprise has begun to develop towards multi-tenancy and enterprise rights management; Hermes Agent connects multiple messaging platforms with a single gateway and provides functions such as memory, scheduling and isolation of sub-agents. This type of tool provides more space for deployment and model selection, and the management of related systems is also the responsibility of the adopter.

Google's entry point is the Workspace environment already used by enterprises, allowing employees to initiate tasks directly from Gmail, Docs or Chat, and then the Gemini agent connects to other data and application systems. For companies that already have extensive Workspace deployments, this integration reduces the need to build additional work portals. However, the access rights required to operate across systems and what data agents use to perform their work must still be integrated into the company's existing management processes.

Gemini Enterprise Agents are not yet fully available. Google has announced early test customers such as On, Shopify and PayPal, and is expected to expand to some Workspace business users in the future. There is no specific information on the launch date, applicable subscription plans and prices in the Taiwan market.