For DingTalk founder Chen Hang, the key insight of the artificial intelligence era did not come when OpenAI’s ChatGPT appeared. It came earlier, in 2020, after his first departure from DingTalk, when he traveled across China, studied cities of different sizes, and thought through what was changing.
“What I realized was that the most disruptive change in the AI era is not anything else, but that the user has changed,” he said. In China’s second- and third-tier cities, and in places farther from the center, he saw a shift in the core base of internet users. That, he said, changed the kinds of applications being built. The rise of apps such as Pinduoduo followed the same logic.
With the arrival of large language models, that view has only sharpened. Chen’s argument is that, in the AI era, the role of carrying out tasks online is shifting from humans to AI.
That idea ran through DingTalk’s product launch on March 17.
Around the same time, Alibaba announced on March 16 what could become a major organizational change since the start of the AI boom: the formal establishment of the Alibaba Token Hub, or ATH, business group, which includes the Qwen business unit.
Alibaba said the Wukong business unit, also part of ATH, would build an entry point for B2B AI applications integrated into enterprise workflows.
The launch came as Alibaba was completing a reorganization, and it marked the first formal public debut of the ATH business group.
For DingTalk, the main update was the launch of Wukong, an AI-native work platform. The business unit that includes DingTalk has been renamed the Wukong business unit.
Wukong operates alongside DingTalk but is tightly integrated with it. Its agent capabilities power features within DingTalk, and enterprise users are expected to access Wukong directly through DingTalk.
From Laiwang to DingTalk and now Wukong, is this another attempt at reinvention?
There was a time when Chen devoted himself fully to promoting DingTalk. He did not even use WeChat, relying on DingTalk alone to stay in touch with friends and family. Now, he said, he has “let go of the ego.”
“In the new era, whether it’s called DingTalk or Wukong no longer matters,” he said.
He said that is because an AI-driven work platform no longer depends on DingTalk itself.
“It doesn’t matter whether you use DingTalk or not, whether you use Slack, Teams, Lark, or WeCom,” he said. “Wukong can connect to all of them.”
Wukong had been in development for months. At the end of December, when DingTalk launched its Mulan version, Chen had already pointed to the direction of travel: DingTalk’s new positioning was as an agentic operating system built for AI-based work. At the same time, it introduced the enterprise AI hardware product DingTalk Real. The software installed in that hardware was the prototype of what has now become Wukong.
In terms of user experience, Wukong appears to be a more assertive product than DingTalk. DingTalk has completed a full CLI (command-line interface) overhaul, enabling Wukong agents to operate thousands of DingTalk functions natively instead of simulating human clicks in a graphical user interface.
Alibaba said this means tasks that once required people to click through DingTalk can now be called directly by AI. For example, if a user receives a client email while on the road, Wukong could remotely operate an office computer, download the file, format it using a previously used customs declaration template, and send it back. In Alibaba’s description, the process can be completed without further human input.
Enterprise security first
Most AI agents on the market today still handle relatively narrow tasks, such as writing copy or searching for information. Once they are plugged into enterprise operations, harder questions emerge:
- Who has permission to access which data?
- Can the AI’s actions be traced?
- How much is being spent on tokens each day, and how should those costs be accounted for?
Wukong follows a different design logic. It inherits an enterprise’s permission system: if an employee can view certain data, the AI can view it too. Its operations run inside a sandbox and leave auditable records. Token consumption is tracked in real time, allowing AI spending to be managed like any other budget item.
“This is the fundamental difference between Wukong and other AI agents: others solve the problem of getting AI to do work, while Wukong solves the problem of getting AI to work inside enterprises safely, controllably and with a clear accounting of costs,” Chen said at the launch event.
Chen said that distinction reflects nearly ten months of technical work by the DingTalk team. In a post-event interview, he said building enterprise-grade security was far from easy.
“Just for file reading and writing alone, AI can do thousands of operations per second,” he said. “To make version management and rollback possible is not something many companies can pull off.”
As OpenClaw has gained popularity, security concerns have also grown. In March, China issued a risk alert through a nonprofit cybersecurity technical body, warning that the tool’s default security configuration was weak and could allow attackers to take control of a system.
DingTalk has also taken a cautious stance. After OpenClaw drew attention, many Chinese companies asked whether it could be used. DingTalk’s customer service response, Chen said, was that without trained professionals putting security measures in place, it is not recommended for direct use in office environments.
Connecting capabilities across Alibaba
Under the plan, Alibaba’s B2B capabilities across Taobao, Tmall, 1688, Alipay, and Alibaba Cloud will be connected to Wukong over time in the form of “skills.”
Take sourcing on 1688 as an example. In the past, merchants had to review supplier credentials one by one, compare prices, and negotiate terms themselves. Alibaba said Wukong can conduct background checks, filter out unreliable suppliers, and recommend options after comparing prices across the web. If it performs as described, it could improve procurement efficiency.
Wukong has also launched an AI capability marketplace spanning development, review, listing, and distribution. It is compatible with open-source “skills” and aims to become “the world’s largest B2B skill marketplace.”
In addition, Wukong has rolled out a solution for one-person teams (OPT), with ten industry versions in its first batch: e-commerce, cross-border e-commerce, knowledge creators, development, retail stores, design, manufacturing, legal, finance and tax, and recruitment.
Unlike most general-purpose agent frameworks, Wukong’s OPT solution does not offer users a bare framework. Instead, it packages “skills,” workflows, and industry data. Users only need to make decisions and approve results, while AI agents handle execution.
Cross-border e-commerce is one example. Previously, users had to scan Amazon bestseller charts, compare prices on 1688, confirm inventory and logistics with suppliers, optimize product titles, and create multilingual videos on their own. Alibaba said Wukong could compress that process from a week to a single afternoon.
Chen told 36Kr that the future advantage of Chinese-made AI agents will not come from head-on competition at the foundation model layer.
“At the model layer, China cannot yet compete, at least for now. It still needs time to catch up,” he said. “The real strength lies in combining models with industries, and in using industry data to generate models and experience.”
“If you can get AI to understand a refurbishment drawing in manufacturing,” he said, “and if it really can understand it, then the exponential transformation of Chinese production begins. That is what we should be paying attention to.”
Wukong’s ambitions extend beyond China. Under the roadmap, it will eventually support access through WeChat, Slack, and other mainstream global IM (instant messaging) platforms across desktop and mobile.
For DingTalk’s enterprise users, Wukong is available immediately upon upgrading to the AI 2.0 version, and it remains free to existing paid DingTalk customers for now.
“Over the past 11 years, DingTalk changed the way we work,” Chen said. “Today, Wukong is trying to define an entirely new way of working for the AI era.”
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Deng Yongyi for 36Kr.