Google launches desktop AI agent to challenge Cowork's market lead

Google’s “Agent” tab first appeared in Gemini Enterprise’s UI on April 13, a subtle but telling addition that hints at a full‑blown desktop experience. The new pane sits beside the familiar chat window and splits into two entry points—“New Task” and “Inbox”—each opening a split‑screen view where a conversational thread runs on the left while a sidebar on the right displays a Goal field, a list of selectable agents, connected apps, attached files, and a “Require human review” toggle. According to Testingcatalog, the layout mirrors the workflow‑centric design of Claude Cowork, where a model receives a concrete objective, tool access, and relevant data before executing multi‑step actions (Testingcatalog, 13 Apr 2026).

The sidebar’s “Require human review” switch is the most striking clue that Google is gearing up for actions beyond the browser. Testingcatalog notes that the toggle suggests the system may launch tasks that need oversight—potentially anything from opening local files to manipulating desktop applications—before they run in the cloud. This implies a shift from a pure web‑based assistant toward a hybrid that can orchestrate work on the user’s machine, a capability that would directly challenge Cowork’s current market dominance.

Google’s broader Gemini revamp reinforces the desktop ambition. The same source reports that Gemini’s “Projects” and “Skills” sections are being refined in tandem with the Agent UI, pointing to an overarching strategy to turn Gemini into a persistent work platform rather than a one‑off chatbot. By unifying goal‑driven agents, project tracking, and skill libraries under a single interface, Google appears to be building a cohesive ecosystem that could eventually merge with its AI Studio desktop app—another product already in development, per the report.

If the Agent tab is anything to go by, Google is positioning the feature as a “task execution workspace” rather than a simple conversational tool. The design encourages users to define a clear objective, select the appropriate AI agent, attach relevant files, and optionally gate actions behind human approval. This mirrors the agentic workflow that Cowork popularized, but with Google’s massive cloud infrastructure and integration with its suite of services, the potential reach could be considerably larger. Testingcatalog emphasizes that the move is a direct response to the growing “desktop and agent‑focused products” from OpenAI and Anthropic, suggesting Google is intent on not just matching but overtaking the competition.

While Google has not confirmed a launch date, the timing of the UI leak—just weeks before its annual I/O conference—signals that a public demo may be on the horizon. The company’s history of rolling out desktop AI tools, combined with the Agent tab’s functional depth, makes a dedicated desktop app a realistic next step. As Testingcatalog concludes, the “bigger question is whether these efforts stay separate or eventually merge into a single product,” but the trajectory is clear: Google is laying the groundwork for a desktop AI assistant that could reshape the market and give Cowork a serious run for its money.