**February 2026**
Weeks after Claude Cowork set the bar for AI-powered desktop workflows in the West, China’s tech sector has responded. Alibaba has unveiled **QoderWork**—billed as the country’s first desktop agent—offering similar “zero-friction” automation: one instruction can trigger file organization, data analysis, report generation, and even travel planning, all running locally on the user’s machine.
From Chat to Desktop
For the past three years, tools like ChatGPT and Gemini stayed inside the chat box: you had to upload files, copy outputs, and move between apps. They couldn’t see your desktop or tidy your folders. With desktop agents, that boundary is breaking. AI is moving from “giving instructions” to “taking action” on the device.
Alibaba’s Qoder lead, Ding Yu, framed the shift clearly: the goal is to move AI from the “chat era” into the “desktop agent era.” QoderWork’s design is local-first, tool-oriented, and agent-driven. It runs on the user’s machine with controlled system-level file access, supports the Model Context Protocol (MCP), and allows custom Skills so teams can encode their own workflows.
What QoderWork Can Do
Early testers (via Chinese tech outlet XinZhiYuan) ran it through typical office scenarios. In one, a cluttered download folder with hundreds of PDFs, spreadsheets, and screenshots was handed to QoderWork for “file organization”; it turned the mess into a structured personal knowledge base. In another, a large travel-behavior dataset was summarized by attraction type, with male/female ratios and an Excel output—done in the time it takes to take a sip of water. The agent can also pull live data from the web to produce research reports, automate academic citation formatting from a set of papers, and generate full PPTs (including embedded interactive demos, e.g. for teaching optics). As a “private travel assistant,” it can pull options from Fliggy and Ctrip, compare them, and produce a custom PDF itinerary—and, with consent, sync the plan into the user’s calendar.
Why “Local” Matters in China
Many AI assistants follow an upload-to-cloud, process-in-cloud, download-result pattern. That raises speed and privacy concerns—especially for sensitive data like financials, code, or unreleased strategy. QoderWork is built for local execution: the agent runs on the user’s device and calls local apps, so data doesn’t have to leave the machine. The product also emphasizes step-by-step planning and clear feedback: complex requests are broken into tasks, executed in order, and the user can see and control the process; when instructions are ambiguous, the agent asks instead of guessing. Under the hood, MCP provides a standard way to connect to local databases, Notion, or custom scripts, while custom Skills let users package recurring workflows—e.g. a “weekly sales report” Skill that pulls Excel data, compares to last week, and drafts an email.
Domestic vs. Global
In the article, “domestic” (国内) refers to China: a homegrown alternative to Claude Cowork, built by a major Chinese tech player (Alibaba) and aimed at Chinese professionals and enterprises. QoderWork is in limited invite-only testing; Mac users can join first, with Windows support promised soon. The broader message is the same as in the West—2025 was the year of “vibe coding”; 2026 may be the year of “vibe working,” where describing the task matters more than mastering every tool.
Source: XinZhiYuan via Baidu (Feb 1, 2026). QoderWork: qoder.com/qoderwork
