On Thursday, October 8, 2026, at its Gemini at Work 2026 event, Google Cloud introduced the “Gemini agent,” which it describes as a “new single, universal agent for work.” The tool is built into the Gemini Enterprise app, Workspace and third-party services, and according to the company it lets users hand it “objectives, not instructions.” The launch targets businesses first and, for now, is in private preview for selected customers.
The announcement comes in the middle of a race to build AI agents, systems that do not just chat but take on entire tasks: generating code, scheduling meetings, booking travel or preparing documents. TechCrunch noted that the move follows the rise of consumer agents such as Meta’s Muse, those that operate over messaging such as Instinct, and the recent debut of Dots in ChatGPT. Bloomberg framed it as an escalation of the competition with OpenAI, Anthropic and Microsoft for corporate customers.
Google CEO Sundar Pichai opened the event with figures that explain why the company is starting with the corporate world: Gemini has more than 1 billion monthly active users, and nearly 90% of Fortune 100 companies use Gemini Enterprise at work. According to Pichai, starting with businesses will let Google solve the “harder problems around security, scale, and performance” that come with agents of this calibre before bringing them to the general public.
Thomas Kurian, CEO of Google Cloud, explained that the agent can plan work, use custom skills and tools, and connect to each company’s internal systems to reach its goals. 9to5Google reported examples from Google itself: if a manager emails asking for the latest project update as a slide deck, Workspace recognizes it as a delegatable task and offers a one-click option to pass it to Gemini.
Another example cited by the company is asking Gemini to set up a meeting with “the usual team of regional event leads” without supplying a single name or email address. The agent works out who those people are from the membership of a chat space and the thread from the last event, checks their calendars and starts an email thread to coordinate a time, even with external participants. Google also pitches it as a team member that can act as a project manager or as an analyst in a finance department.
One of the most striking details is that the agent is not tied to Google’s models. By default it picks the best model for each task, but users can take over and choose another one, including third-party models, starting with Anthropic’s Claude family. Google says it will expand the model picker to open models and other private models in the future, with the stated goal of delivering the best possible quality while lowering the cost of each job.
On connections, TechCrunch listed Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake, among other services. The agent can also work securely with any server compatible with the Model Context Protocol (MCP), inside or outside the company’s network. Requests can include files, folders or projects that combine documents and skills designed for specific workstreams.
Perhaps the most symbolic change is that the agent will have its own Workspace account, as if it were just another colleague, with its own email address and context. It knows who is on which team, each person’s time zone, who needs to approve each item and what is on people’s calendars. Users can call on it by tagging it, emailing it, sharing a document with it or adding it to a group chat, and every action is written to an audit trail attributed to the agent rather than to a person.
Under the hood, Gemini coordinates subagents to tackle multi-step tasks, in parallel or in sequence, that can run for hours or days. According to 9to5Google, the agent runs in the cloud with a single set of memories and context, and Google distinguishes four kinds: session memory for the task at hand, semantic memory as a structured knowledge base, procedural memory for how a job gets done, including skills it writes for itself, and episodic memory of everything it has done before.
To supervise it, Google offers a “tasks inbox” where users can see the agent’s thinking, its delegation to subagents, the skills it loads, its code and the progress of each assignment. The agent will be available on iOS and Android, on Windows and Mac desktops, on the command line, in Workspace, Microsoft 365, ServiceNow and Slack. In Gmail, Docs, Sheets, Slides and Chat, mentioning @Gemini will be enough, and an API will let it be integrated into third-party applications as a “headless” agent without its own interface.
Early testers include sportswear brand On, Shopify and PayPal, while Gemini Enterprise’s customer base includes BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers. Google also announced flexible spending options, such as multi-model orchestration, smart routing and real-time spend caps, to help businesses keep the cost of artificial intelligence under control.
The significance of the announcement goes beyond a new feature. Giving an agent its own identity, persistent memory and access to internal systems shifts the question from what AI can answer to what it can do on behalf of an organization. The audit trail and the tasks inbox point to an obvious concern: if an agent acts like an employee, companies will need to know what it did, when it did it and with which permissions.
Open questions remain. Google has not given a general availability date; 9to5Google says it will arrive “soon” for Workspace customers on select Business and Enterprise plans. Per-task pricing details and independent evaluations of the agent’s reliability on long jobs are not yet known either. The opening to Anthropic models, for its part, raises questions about how data, responsibilities and costs will be shared between providers.
In short, Google has chosen to compete in the agent era with an offering that combines scale, integration with office tools and a multi-model approach that is unusual for a major lab. If the private preview confirms the promised reliability, the “Gemini agent” could become one of the first digital colleagues deployed at scale in large companies. For now, the promise is ambitious, and the real test begins in the offices of its first customers.
