Key Notes
- Google Cloud’s Gemini agent can dynamically create subagents for complex work.
- The agent supports Gemini and Claude models, with third-party model choice still in early access.
- Background delegation and mobile and desktop capabilities are also in early access.
Google Cloud has announced Gemini agent, a universal work agent that can create specialized subagents and use both Google’s Gemini models and Anthropic’s Claude. The system is designed to take responsibility for assignments across business software, including work that continues after a user closes their laptop.
The announcement comes with a rollout qualification. Google’s product page says background task delegation, mobile and desktop access, and third-party model choice are still in early access. Businesses should distinguish the capabilities being introduced from those already enabled in their accounts.
Gemini Is the Agent, Claude Can Be the Model
Google separates the agent employees interact with from the model performing a particular job. Its business announcement explicitly names Gemini and Claude as supported model families, with other private and open models planned. The company says routing work between models can balance quality and operating costs while preserving the surrounding business context.
That makes the product’s competitive position more interesting than another Gemini chat interface. Google is seeking to provide the place where work is assigned, connected to company systems and managed, even when a rival supplies the underlying intelligence. Model choice could give customers flexibility, although the actual value will depend on which models, tools and routing options their deployment makes available.
An Agent That Can Assemble Its Own Team
In Thomas Kurian’s launch presentation, Google describes temporary subagents created for specific parts of a larger assignment. Each has its own identity, and the main agent coordinates parallel or sequential steps. Tasks can run for hours or days, giving the system a way to split a complicated objective into smaller pieces of work.
Google also describes persistent coworker agents with defined organizational roles. These differ from temporary helpers: they can have their own email addresses and storage, continue across sessions and operate on context shared with them. The distinction is between assembling a team for one assignment and maintaining an agent with an ongoing responsibility.
Connecting Work Across Business Applications
The product page lists connections spanning Google Workspace, Microsoft 365, Slack, GitHub, Jira, Salesforce and other business software. Google says selected sources can remain where they already live, with access following existing permissions. It also describes memory for active tasks, business knowledge, workflows and previous actions across devices and channels.
For a business, the useful test is whether those connections reduce the handoffs between applications. A team might want an agent to gather project updates, draft a customer response and prepare supporting material without repeatedly rebuilding the context. That is an example of the intended workflow, rather than a task we have independently tested on the new agent.
Cost Controls Matter When Agents Keep Running
Google’s existing billing guidance describes project-level monthly spend caps that pause agent API calls when a limit is reached. Administrators can resume work or enable overages, and alerts track progress toward the cap. Its consumption options also distinguish per-user subscriptions from pay-as-you-go usage, with some offerings still rolling out.
Those controls address a practical problem: an agent coordinating several helpers can generate costs through repeated attempts and longer execution, not just the first prompt. The same need for clear boundaries appears in Personal Agent Protocol, which concerns how businesses recognize an agent and understand its user’s authorization. Permission to inspect a record and permission to change it remain different decisions.
What Businesses Can Try Today
Google directs prospective customers to its sales team or a 30-day Gemini Enterprise Plus trial, while warning that not every advertised feature is available immediately. Its separate small-business announcement also places Gemini agent in early access and promises broader availability without supplying a firm date. A trial should therefore be evaluated against the features actually provisioned, rather than the entire launch demonstration.
The approach joins the broader push toward persistent assistants, including OpenAI’s Dots. Google’s differentiator is the combination of workplace connections, agent coordination and model choice. For customers, the meaningful outcome will be completed, reviewable work at a predictable cost; the number of agents created is only one part of that assessment.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
