Google used its Gemini at Work event on Thursday to announce a "universal" Gemini AI agent that it says can operate across apps and devices in the background, according to The Verge. The agent is not a standalone product: it lives inside the Gemini Enterprise app, where users chat with it and hand it tasks from a single interface.
The scope Google describes is deliberately broad. The agent works directly inside Workspace apps including Gmail, Drive, Docs, Sheets and Calendar, and users can also reach it from a mobile device, a desktop computer, the web, and third-party apps such as Slack or Microsoft 365. Because it runs in the cloud, Google says the agent keeps the same context across every connected device.
Two structural features beyond a chat window are also described by Google. Tasks can be finished by the agent working alongside job-specific sub-agents, and it may serve in the role of a "coworker agent," possessing a dedicated identity plus its own "@agents.company.com" email address. According to Google, whichever AI model best suits the task at hand is automatically selected by the agent.
Availability is the most important caveat in the announcement. The Gemini agent is currently open only to enterprise customers in private preview, so freelancers, small studios and independent developers cannot sign up for it today. The Verge frames the release as a possible glimpse of Google's wider vision for an assistant that both answers questions and performs work from one window, rather than as a finished general-release product.
Why that framing matters: something adjacent is already shipped by Google. In May, a consumer-facing Gemini Spark agent was previously launched by the company, and tasks across Workspace and on the web can be performed by subscribers using it. Because Spark is reached through a separate interface inside the Gemini app, a rebrand of an existing consumer feature is not what the enterprise agent simply is; different surfaces with different access rules are where the two sit.
For this audience, the practical question is less about the demo and more about where the work lands. A cloud agent that holds context across devices and reaches into Slack and Microsoft 365 implies that task state, not just chat history, becomes portable between tools. Freelancers and small teams who already juggle a client's Workspace account alongside their own tooling would feel that shift first, because the agent's value depends on how many of those surfaces it can actually touch.
The sub-agent and coworker-agent details are the parts most likely to change how work is assigned. A dedicated identity with its own email address suggests the agent could appear as a participant in shared systems rather than as a tool one person operates. Google has not published how permissions, billing or accountability work when an agent with its own address acts inside a client's environment, and none of that is addressed in the supplied reporting.
There is also a gap between what Google says the agent does and what has been demonstrated publicly. The announcement describes automatic model selection and background operation across apps, but the evidence here contains no independent testing, no benchmark, and no named customer describing results. Those claims should be read as vendor statements until enterprise preview users report on them.
The device list is worth reading carefully. Mobile, desktop, web and third-party apps are named as entry points, and cloud execution is given as the reason context persists. That is a description of intended architecture, not a measurement of reliability. Whether context actually survives a handoff between, say, a phone and a Slack thread is exactly the kind of claim that private previews exist to test.
Timing is another unresolved point. The Verge reports the announcement as part of a Thursday event and notes the May launch of the consumer Spark agent, but the supplied material does not give a general-availability date, a price, or a list of which enterprise tiers qualify for the private preview. Anyone planning around this should treat the schedule as unknown.
For freelancers and small studios, the honest read is that this is a signal rather than a tool they can adopt now. The direction — one agent spanning the apps a client already pays for — is the same direction competitors are moving, and it raises the odds that client-side workflows will eventually include agent identities that freelancers have to work alongside or around.
The tradeoff is control. A single agent holding context across Workspace, Slack and Microsoft 365 concentrates a lot of a project's state in one vendor's cloud, and the announcement does not describe data boundaries, retention or what happens when a client and a contractor share an agent identity. Those are the questions a contractor would need answered before letting such an agent near client data.
What remains unknown is substantial: no pricing, no general-availability date, no published limits on which third-party apps are supported beyond the named examples, and no independent account of the agent completing real work. The only firm facts are the private-preview status, the enterprise-only audience, the named surfaces, and the two structural features Google described.
The reasonable conclusion for now is to watch the enterprise preview rather than plan around it. If Google later opens the agent beyond private preview and documents permissions and pricing, the cross-app context claim becomes testable; until then, freelancers and developers should treat the announcement as a statement of intent from a vendor with a consumer agent already in market.