悟空

What Wukong is

Wukong is Alibaba's enterprise AI work platform delivered through DingTalk, positioned so the assistant performs work rather than only answering questions. Its slogan is a plain instruction to speak and let it act, and it is aimed squarely at the one-person business.

What it does

The assistant operates the platform's own applications such as documents, calendars and approvals, turning an instruction into an action rather than a suggestion. The loop is speak, execute, deliver.

It is deeply integrated with the platform and synchronised across phone, tablet and desktop. A skill centre offers preset capabilities for documents, data analysis and presentations alongside custom skills, and long-term memory of preferences and context means recommendations sharpen with use. Nine one-person scenarios are showcased, spanning retail, manufacturing, recruiting, e-commerce, design, development, legal, content and finance, and the plan is distributed as a desktop client with stated operating system requirements.

Who it is for

It is aimed at solo entrepreneurs, small businesses and teams already living inside the platform, where the integration is the reason to use it rather than a separate assistant. Organisations without that platform dependency get less from it.

What to keep in mind

The compliance gap is notable for an enterprise tool. The landing page publishes no privacy policy, service terms, data-security statement or filing details, even though the product's entire point is acting on company documents, calendars and approvals.

Before enabling it, ask what it can read and write, whether its actions are logged and attributable to the assistant rather than to a person, and how that maps to your internal controls. An assistant that can approve things on your behalf touches exactly the workflows that exist to prevent unauthorised action, so the audit trail is not a nice-to-have.

Because memory persists, review what it accumulates. Memory that sharpens recommendations also accumulates context about how your business runs, and that store is worth understanding before it grows for a year.

No price or plan information appears on the landing page, so check the platform's own commercial pages rather than assuming it is included.

Confirm the price and the policy, and settle how actions are logged before letting it near approval workflows. It is also worth clarifying what happens to the work it produces: actions taken inside documents and approvals create records, and whether those records distinguish an assistant's action from a person's is the difference between an audit trail and a gap in one. Ask about the desktop client's operating requirements too, since a work platform deployed this way becomes software on machines somebody else administers. Finally, pick one of the nine scenarios and test it end to end rather than enabling everything at once.