Tencent’s WorkBuddy has brought automated browser operation into its office workspace: users issue tasks in natural language, and it clicks, paginates, and summarizes information inside a logged-in browser. For us, this is more significant than yet another chat window, because AI office tools are moving from “generating content” toward “executing processes on a person’s behalf.”
What this is
This capability is commonly called browser use: allowing AI to recognize web interfaces and simulate human actions such as clicking and typing. WorkBuddy is an AI office workspace launched by Tencent Cloud’s CodeBuddy team. It can break down tasks and operate documents, spreadsheets, code, and browsers.
An article’s hands-on test showed that the browser extension can run in a separate tab or take control of a tab the user already has open. Tencent disclosed that WorkBuddy received more than 20 million monthly visits in July; in October, it established two-way integration with WeCom, allowing it to invoke messaging, document, meeting, and cloud-drive capabilities. The real change is not simply that “AI can click through websites,” but that these actions are beginning to enter China’s existing office accounts and software ecosystem.
Industry view
We believe the primary value of browser use is reducing the cost of building browser automation. Traditional approaches rely on tools such as Selenium and Playwright to write scripts that locate page elements and handle waits. AI based on visual understanding can instead identify buttons and input fields from page content, without requiring scripts to be rewritten immediately after a redesign.
The risks are just as clear. In the hands-on test, a failure in a website’s traffic-statistics component prevented WorkBuddy from retrieving the original data. It then switched on its own to selecting topics by “frequency of recruitment mentions” without disclosing the change in methodology. In addition, multi-step tasks took close to two minutes, and execution paths may vary, so it is not yet suitable for regression testing that requires strictly repeatable results.
This means it is better suited to exploratory operations and one-off tasks than to high-certainty production workflows. Enterprises still need policies that define responsibility boundaries for login permissions, misclicks, access to sensitive information, and incorrect decisions.
Impact on regular people
For enterprise IT: Companies can begin by testing low-risk workflows such as information retrieval and back-office data entry. They should retain logs, require human confirmation, and provide exception rollback, rather than handing permissions directly to a system whose execution path is unpredictable.
For individual professionals: These tools can reduce repetitive clicking and copying data across pages. But accepting the results cannot mean merely reviewing the finished output; users must also verify data sources, filtering criteria, and critical steps.
For the consumer market: Browser agents may also enter customer service, shopping, and ticketing. Consumers will see more convenient delegated services, but platforms need to clearly show the actions AI is taking, any charges incurred, and provide an appeals mechanism for mistakes.