Google Cloud Launches Enterprise Gemini Agent That Can Use Anthropic's Claude

The announcement came in a long post titled “Welcome to Gemini at Work 2026: Introducing the Gemini agent,” signed by Google Cloud CEO Thomas Kurian. The post focused on updates to Gemini in the enterprise agent field, with the newly launched Gemini Agent as the centerpiece. Google describes it as a general-purpose office agent that covers many kinds of work tasks and can operate over long periods.

According to QbitAI, Gemini Agent can help research information, write emails, make presentations, analyze data, write and run code, call tools across applications, plan tasks on its own and bring in multiple sub-agents when needed. Its stated working principle is “You give it objectives, not instructions.” Google listed unified entry, persistent cloud operation, cross-platform calling, multi-agent collaboration, event triggers and scheduled tasks among its core abilities.

Two features drew particular attention. First, Gemini Agent can have its own enterprise identity, including an email account, calendar and independent credentials, functioning as an AI colleague. Second, it can automatically choose an underlying model according to the task and can even call Anthropic's Claude. Google said the underlying model can be selected separately. At this stage, Gemini Agent can dynamically choose between Google's Gemini family and Anthropic's Claude family, and Google plans to support more closed-source and open-source models in the future. The choice is based on effectiveness, speed and cost: simpler work can use lower-cost models, while complex tasks can call stronger models. Multiple models can also be combined within one project. Google calls this multi-model orchestration and provides Smart Routing for automatic routing. The report said accumulated context, Skills and enterprise data can be retained even if the underlying model changes, allowing companies to adjust models without rebuilding established workflows.

Google also proposed a Coworker Agent under the Gemini Agent framework. Enterprises can create a long-term identity with clear job responsibilities for Gemini, letting it participate in daily collaboration like another team member. The AI colleague can have an independent Google Workspace account, corporate email, calendar, Google Drive storage and a place in the company directory. Team members can invite it to Google Chat groups or mention it in documents. Its operations leave their own records so enterprises can manage and track what it has done. Gemini Agent has four types of memory. Session memory preserves the context of a current task, even if a task lasts several days, so the agent remembers completed steps and remaining work. Semantic memory accumulates structured knowledge from documents it reads, conversations with people and collaboration with other agents, such as company products, team responsibilities and internal meanings of business terms. Procedural memory records how a job should be done, including what data a report needs, what analysis process to follow and what format to deliver. The agent can write its own Skills to save methods learned during work for later tasks. Episodic memory records past work and execution experience. Google compared Gemini to a new employee who gradually becomes familiar with the user, the user's tools and the team before formally starting work.

In the example given in the post, Gemini Agent can complete a market analysis report by searching and organizing materials, analyzing relevant data, building a financial model in Google Sheets and using Google Slides to generate a presentation. It can choose tools and call different applications as needed, then return the finished work.

The launch came after OpenAI released Dots on September 29, according to QbitAI. Dots can work around the clock, has an independent cloud computer, remembers users' long-term goals and work habits and can connect to more than 4,000 applications. For developers, Dots can analyze new user feedback, modify code and prepare a pull request for review. For content creators, it can organize new interview records and produce content drafts. OpenAI is also exploring professional Dots for enterprise roles, giving agents independent identities and responsibilities, though that part is mainly in early enterprise pilots. Earlier, Meta introduced Muse, a personal AI agent for shopping, booking travel, sending emails and making payments. It can help find goods, compare options, operate websites and complete purchases. The report said Meta's Muse focuses more on personal life, while Google's Gemini Agent focuses on entering existing enterprise organizations and office systems. Gemini Agent can use business context from Gmail, Docs, Sheets and Calendar, and companies can create Coworker agents with independent email, calendar, account and permissions. Google also places model selection, enterprise data connections, security audits and cost control inside this system. Claude can be used as an underlying model when Gemini Agent completes tasks.

The post also announced that Gemini is entering Google Workspace comprehensively to support cross-application work; opening Skills, Tools and enterprise connectors; launching professional agents for data analysis, finance and law; and enhancing agent security governance and cost control.

Editor's Summary Google Cloud's Gemini Agent extends enterprise office automation by giving an AI worker a persistent corporate identity, four types of memory, cross-application execution and multi-model orchestration that can include Anthropic's Claude. The announcement follows OpenAI's Dots and Meta's Muse, placing Google's emphasis on integration with existing enterprise workflows, data and governance. The post also introduced broader Gemini updates for Workspace, professional agents, connectors, security and cost controls.