On March 16, Alibaba officially announced the establishment of the Alibaba Token Hub (ATH) business group, to be led directly by CEO Eddie Wu, consolidating core AI businesses including Tongyi, Qwen, and Wukong under a single coordinating structure.
The following day, the AI DingTalk 2.0 launch event proceeded as scheduled. The centerpiece was Wukong — presented as the world’s first enterprise-grade AI-native work platform, built on eleven years of DingTalk’s accumulated infrastructure. This is not a product refresh. DingTalk has rewritten its underlying architecture from the ground up, transforming from office software operated by humans into what the company describes as an operating system that AI can operate — and in doing so, it is directing 800 million users and 30 million enterprise organizations toward a broad migration to AI-native working.
The Logic Behind Wukong: Alibaba’s Token Strategy
To understand Wukong, it is necessary to first understand the organizing logic behind Alibaba’s creation of the ATH business group: the Token. In the AI era, Tokens function as a universal unit of computational currency, and they form the core thread running through Alibaba’s AI architecture.
The division of responsibilities is clear. Tongyi Lab is responsible for generating Tokens. The MaaS business line is responsible for delivering them. Qwen and Wukong are responsible for applying them. Within that structure, Qwen addresses consumer and personal use cases — solving the problem of making AI accessible to individuals. Wukong addresses enterprise use cases on the business-to-business side, solving the problem of making AI genuinely useful to organizations.
Wukong’s capacity to carry Alibaba’s enterprise AI ambitions is inseparable from the foundation DingTalk has built over eleven years. Since Chen Hang — known by his nickname Wuzhao — led the Laiwang team to pivot and build DingTalk 1.0 in 2014, the platform has grown to 800 million users and 300,000 paying enterprise customers. In doing so, it has accumulated the organizational elements that enterprise-grade AI agents require most: organizational structures, permission systems, approval workflows, and proprietary data assets. These constitute what the company describes as the secure operating environment that agents need to function reliably within a corporate context.
At the launch event, Wuzhao framed the transition directly: “Over the past eleven years, we have accumulated enough confidence. Today, we smash DingTalk, rebuild it with AI, and forge Wukong.” The name itself carries deliberate resonance — Sun Wukong, the Monkey King of Chinese mythology, is defined by a disruptive origin, limitless adaptability, and the capacity to overturn established order. The company is drawing an explicit parallel to the disruption of traditional enterprise software.
What Wukong Actually Is: An Executor, Not an Assistant
The distinction Alibaba is drawing between Wukong and conventional AI assistants is a structural one. Wukong is not a conversational interface that provides answers within a chat window. It is, in the company’s framing, a system with the capacity to act — one that has been given, in effect, hands and feet.
To achieve this, DingTalk undertook a foundational technical overhaul: a complete rewrite of the underlying codebase and a full conversion of the system to a command-line interface (CLI) architecture. Previously, human users operated DingTalk through a graphical interface — clicking, navigating menus, filling out forms — a method that is inherently inefficient and that AI cannot directly invoke. The CLI conversion decomposes all DingTalk capabilities into tens of thousands of atomic commands that AI can call directly, bypassing manual operation and enabling autonomous task planning and execution across what the company calls its “lobster legion” of collaborative agents.
The foundation of the system is Agent Runtime, a cross-system, cross-model platform designed to operate across Windows, Mac, and Linux environments and to be compatible with Tongyi Qwen as well as other large models. Layered on top of this is a five-tier security framework and four AI defense lines, designed to address the safety and controllability concerns that have been the primary obstacle to enterprise adoption of autonomous agent systems.
In terms of deployment, Wukong is available both as a standalone application and as a plugin embedded within DingTalk, reaching more than 20 million enterprise organizations on the platform. Invitation-based testing launched at the time of the event. Cross-platform connectivity with Slack and WeChat is supported, with the intended user experience described as a single point of entry from which any task can be executed.
Industry Applications: One Person, One Team
The most consequential demonstration at the launch event was not a technical specification but a series of applied scenarios showing what Wukong can accomplish within specific industry contexts. Wuzhao presented ten industry solutions, each structured around a consistent proposition: one person, working with Wukong, can perform the work of a professional team.
In software development, a complete ticketing system was generated in fifteen minutes from a set of meeting recordings — work that previously required two or three engineers over two weeks. In finance and accounting, a single voice instruction queried fifteen bank accounts and produced an automated daily cash flow report, eliminating the manual process of consolidating data from multiple sources. A staff member in the testing phase reportedly broke down at the demonstration. In legal services, a single instruction produced a complete set of pre-trial materials, with the potential to scale one lawyer’s client capacity from ten to two hundred. In cross-border commerce, product selection, supplier sourcing, and primary-image optimization were fully automated; in internal testing, one merchant’s main-image click-through rate rose from 0.8 to 2.1.
Beyond these sector-specific applications, Wukong is designed to integrate into routine enterprise workflows: intervening in meetings to summarize discussion and flag digressions; building what the company describes as an intelligence system to monitor global industry trends and competitor activity and push relevant analysis to organizations of any size; generating sales analysis reports, supporting designers in client-facing proposal work, and handling the full production chain from topic selection to finished video for content teams.
Wuzhao offered a self-deprecating illustration of the shift: “I spent hours checking domain names. A colleague handed the same task to Wukong and it was done in two minutes. I suddenly realized I’d become an old fossil.” The comment points to a broader structural change — the progressive transfer of execution-layer work to AI, with humans retaining decision-making authority over outcomes.
From the rebranding of DingTalk’s identity, to the formation of the ATH business group, to Wukong’s official launch, Alibaba has articulated a clear and comprehensive claim on the enterprise AI market. Wukong is not an incremental update to DingTalk. It is a rearchitecting of the enterprise software category itself — a shift from a tool that humans use to a platform that AI can operate on behalf of humans.
The migration of 800 million users has begun. For the 43 million small and medium-sized enterprises that DingTalk serves, Wukong represents a materially lower barrier to meaningful AI adoption. When a single individual working with Wukong can execute the output of a full team, and when every core workflow of an organization becomes a candidate for AI optimization, the underlying economics of how work is organized begin to shift in ways that go well beyond any individual product launch.
As Wuzhao put it at the event: “Whoever can help AI connect to the physical world the fastest — understand it, and help humans solve problems within it — is doing the right thing.” The enterprise AI contest has entered a new phase. Wukong is Alibaba’s bid to define what that phase looks like.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.