Google's universal Gemini agent to take on entire work tasks – with its own Workspace identity
On Thursday, October 8, 2026, Google Cloud launched a universal Gemini agent designed to take on entire work assignments, not just answer questions. Most striking of all: the agent is to have its own organizational identity in Workspace — with an account, email address, calendar, Drive storage, and a place in the company directory. But Google left several key questions about activation, licensing, and administration unanswered.
What was announced
At the Gemini at Work 2026 event, Google presented what the company itself describes as "the new Gemini agent, your single, universal agent for work." The wording comes from Google Cloud's own blog, signed by Sharon Prosser, Vice President for SMB and Scaled at Google Cloud, and confirms the launch on the same day as VentureBeat's coverage.
The point of the launch, according to VentureBeat, is to consolidate Google's enterprise AI offering — until now a collection of assistants and specialized tools — into a single platform that can take over entire missions. The agent can do research, create documents, write code, and coordinate other agents, and it is designed to work on tasks lasting hours or days while retaining memory, context, and identity across the applications it uses.
Google does not state a maximum duration for tasks in the primary sources; VentureBeat only refers to "hours or days."
The agent as a colleague: its own identity in Workspace
What distinguishes this platform from a chatbot is the identity model. VentureBeat, citing the announcement by Google Cloud CEO Thomas Kurian, describes the example of an "Event Planner Agent": a manager can create a Gemini agent that gets its own Workspace account, its own email address, its own calendar, its own Drive storage, and a visible place in the company's directory.
In practice, this means the agent is not merely a window an employee opens, but an entity that exists in the organization's systems — it can receive invitations, have documents shared with it, and appear as a participant in workflows in the same way as an employee. It is a fascinating architecture, but also one that raises practical questions: Who owns the agent's files? Who answers colleagues who write to the agent's email? Google has not yet answered such questions.
According to Google, the agent works across web, mobile, desktop, and Google Workspace — and also Microsoft 365 and Slack. The latter is worth noting: Google is positioning its agent as a work colleague in competitors' productivity tools as well, not just in its own ecosystem.
Multiple models, not just Gemini
Under the hood, the architecture is model-agnostic, according to Google. The company supplies its own Gemini models, but the agent can already use Anthropic's Claude models, and support for more proprietary and open-weight models is planned, VentureBeat writes.
Google's blog elaborates on the orchestration with three claims that must all be read as the company's own promises, not verified performance figures:
- Multi-model orchestration that is supposed to select the right model for each task.
- Smart routing that automatically triages workloads so each one runs on the model that delivers maximum performance at the lowest possible cost.
- Real-time project-level spend caps in the Cloud Billing Console.
Google also claims that context, skills, and data follow along if a customer switches models — so that you don't lose the work your agent has done because you switch from Gemini to Claude or vice versa. This is central for customers who want to avoid lock-in to a single model vendor, but the mechanisms behind it have so far only been described in marketing materials.
The security model — Google's claims
For IT departments, the security architecture may be the most interesting part. Google states, according to VentureBeat, that each agent receives a cryptographically attested identity, and that permissions are governed through the company's existing security systems. The agent's actions are logged in audit logs, including when it creates isolated environments to run code. In addition, an "Agent Gateway" is described, which is to enforce policies on the agent's communications.
This is a meaningful approach: by giving the agent an attested identity instead of letting it act as a real person, it becomes possible in principle to distinguish human from machine activity in the logs — and to roll back actions performed by an agent that does something wrong. But these are still claims from Google. There is no independent verification of how this works in practice, how granular it is, or how it interacts with existing tools such as conditional access and DLP in customers' environments.
The race for digital employees
The launch does not come in a vacuum. In recent weeks, a series of competitors have launched similar concepts:
- September 25: Microsoft introduced an updated Copilot that combines productivity work, software development, and persistent automation.
- September 29: OpenAI launched its always-on "dots" agents.
- October 6: Anthropic announced native Claude integration with Google Docs, Sheets, and Slides, allowing users to edit Workspace files directly through Claude or use Claude inside Google's productivity apps.
- Earlier this fall: SpaceXAI launched persistent Grok Bot agents in August, followed by Team Bots in September.
Anthropic's Workspace integration two days before Google's launch is particularly worth noting: Claude can already edit Google's own files, and Google is responding with an agent that can run Claude models. The vendors are becoming both competitors and mutually dependent on each other's platforms.
What Google has not answered
For an IT decision-maker, the most concrete part of this story may be what is missing. VentureBeat points out that Google has not specified:
- How customers activate the new universal agent. The existing Gemini Enterprise service is available via web and mobile, but the activation mechanism for the new agent is unclear.
- When all the features will be available. Google has confirmed early access with SMB customers, but not a general availability timeline.
- How licensing and provisioning work for agents with their own Workspace identity.
- How much control administrators have over the individual components of an agent's Workspace presence — for example, can you give an agent a calendar but not email, or hide it from the company directory?
As long as these questions remain open, interested parties should treat the launch as a directional signal rather than a product they can plan around in detail.
A note on the sources
One secondary source, Cryptobriefing, describes a platform said to be an evolution of Vertex AI, with developer tools such as Agent Development Kit, Agent Studio, Agent Runtime, and Memory Bank, a claimed maximum runtime of seven days, and access to over 200 models. These details are not confirmed by either Google's own blog or VentureBeat, and the source's dating (April 22, during Google Cloud Next) conflicts with the October 8 timeframe that both primary sources confirm. AIMag has therefore chosen not to use this information as facts about the product.
What remains
With this launch, Google has taken a clear stance: the AI assistant of the future is not a tool you use, but a colleague with its own identity who works alongside you for days. It is a concept with great impact — and with a host of security, legal, and organizational questions that none of the vendors in this race have fully answered. For now, the agent exists in early access with SMB customers; how it works in practice remains to be seen.