WorkBuddy Puts a Browser in the Agent's Hands: Independent Tabs, Its Own Cursor, Mobile Dispatch
Agent-driven browser work now runs without occupying the user's screen or mouse, and mobile dispatch turns it into an unattended background worker for tasks like form filling, site testing, and comment analysis.
WorkBuddy now offers browser control that runs in a dedicated tab group with its own mouse cursor, so the agent can open pages, click, and fill forms without taking over the user's screen or input. Users can send tasks from the mobile app or mini-program, pick which device runs the job, and check results later. The agent reports progress as it works, so no silent long-run.
Concrete uses from the field: it read the official WorkBuddy docs and organized a learning map in under ten minutes, classified cases by topic, and produced a study order for beginners. It filled an enterprise OA expense report, checked the fields against instructions, and listed the filled form for user confirmation before submitting. It also acted as a novice tester on a website, walked every navigation path, noted problems, ranked them by severity, and even proposed a design change. On mobile, it reviewed a few hundred Bilibili comments and summarized recurring topics, engagement, and sentiment into a report with suggestions for the next video.
Setup requires the latest client, installing the extension in Chrome, Edge, or QQ Browser, and enabling the 'Allow WorkBuddy to control browser' switch. Known limits are captchas, slider verification, slow page loads, and site redesigns, which may force human takeover or a retry.
Running the agent in its own tab group with a separate cursor means browser automation no longer blocks the user's own input, which is the difference between an annoying demo and a usable background tool.
The explicit confirmation step before a submission shows that agent vendors still need to earn trust for high-stakes actions; listing what will be submitted is a simple but effective control.
Using the agent to walk a website like a first-time visitor and rank issues by severity turns UX review into a repeatable, automated QA pass, though it still can't judge design quality.