After ONE: The AI Organizational Experiment Legacy of DingTalk

```

Since June 2026, an article titled "Inside DingTalk" has been circulating in the product circle and enterprise service industry.

The author claims to have participated in the development of DingTalk’s core confidential project ONE, and as an insider, recalls the entire process from project initiation, launch, to gradual marginalization.

According to the timeline in the article, ONE was born in April 2025, officially debuted in August, and once reached about 3 million daily active users. It was also the most important AI product attempt after Wu Zhao returned to DingTalk.

According to the product definition, ONE's goal is to reorganize work scattered in messages, meetings, approvals, to-dos, documents, and calendars, so that users can see the most important matters of the day when they open DingTalk each day, achieving a shift from "people look for tasks" to "tasks look for people," like an AI office version of "Douyin" (TikTok).

If this concept succeeds, it would change not only a product feature but also the way corporate organizations acquire information, allocate attention, and advance work.

Many people take "Inside DingTalk" as evidence of DingTalk’s AI transformation setbacks, while others see it as a sketch of the organizational path after Wu Zhao’s return.

This article is more like an observational record written by a project participant after leaving, carrying personal perspectives and inevitable emotions and positions. Precisely because of this, it provides an observational angle rarely accessible from the outside.

Over the past year, major model companies have been talking about Agents, workflows, and AI native. What’s often seen at launch events are glamorous product demos, but few are willing to talk about what actually happens when they enter a corporate organization.

ONE happens to provide a sample.

On the surface, this is the rise and fall of an AI project; more deeply, it documents a real friction between corporate organizations and AI workflows.

What ONE Wants to Change

If ONE is understood merely as an AI feature in DingTalk, its ambition might be somewhat underestimated.

Over the past twenty-plus years, enterprise software has adhered to a tool logic. Messages are in the message page, documents in the document page, approvals in the approval center. Users actively search, organize, judge information, then decide their next action.

ONE wants to change this process. It hopes that AI can first understand work, then reorganize it. Users no longer need to browse dozens of group chats, nor switch between approval, to-do, and calendar back and forth. AI filters and sorts things in advance, presenting important information proactively.

From a product design perspective, this already involves changing the entry point of office software. The ONE team had envisioned several typical scenarios: In the morning, AI has already organized meetings and to-dos for the day; during work, AI proactively identifies key items and issues reminders; before ending the day, AI helps users check for omissions, avoiding missed important information. It would also recommend industry trends, competitor info, and research materials based on users' work content.

Behind this design lies a judgment: The competition in future enterprise software is shifting from the number of features to contextual capabilities.

With the rise of the Agent wave, users want AI to understand their work environment, master organizational context, and push things forward. DingTalk naturally has advantages here. It possesses corporate address books, organizational relationships, message links, approval systems, meeting systems, and document systems. Ordinary AI assistants hardly know who is connected to whom, or which messages affect project progress, but DingTalk does.

In the traditional internet era, entry meant traffic; in the AI era, entry starts to mean context. This is also an important reason why Wu Zhao is again betting on AI.

The Weight of Multiple Objectives

ONE’s later experience is hard to simply attribute to product design itself. "Inside DingTalk" mentions that from its birth, ONE carried multiple expectations.

First are expectations at the user level. Messages are ever increasing, group chats ever more chaotic; many people's first act on opening DingTalk daily is switching between dozens of windows. ONE hopes to help users reorganize these fragments so work no longer gets dragged by the message flow.

Next are expectations at the product level. The era of large models has arrived, DingTalk needs a new AI entry and new product narrative. For DingTalk after Wu Zhao’s return, ONE also carries organizational significance; it’s not just a functional project but hoped to be a representative of a new round of AI transformation.

Further, there are expectations at the commercialization level. AI needs computing power and real scenarios. DingTalk naturally has vast work scenarios, thus seen as an important AI pilot ground. The problem is, each goal is valid individually, but may not be achievable simultaneously by the same product.

ONE wants to become the entry that everyone opens daily, and also solve deep enough problems; needs to cover most users, and also prove commercial value; bears strategic significance while shouldering growth metrics. A product for all users usually addresses common needs; real paid scenarios often come from specific industries and businesses.

Thus, ONE kept switching among several directions: workflow, content flow, knowledge service, Agent entry, AI workbench… Each direction makes sense, but each is hard to go to extremes.

Looking back, ONE carried not just a product goal. It was expected to represent DingTalk’s AI transformation and also explore AI-native organizational forms.

For DingTalk, ONE is both a product and an experiment.

AI’s Organizational Experiment and Friction

The core discussion of "Inside DingTalk" is what happens when AI enters organizations.

From birth, DingTalk has been software with distinct management attributes. Whether it's DING, read/unread, approval flow or attendance, behind these are concrete management demands: Has the notification been seen? Has the task been advanced? How far has the project progressed?

These demands constitute DingTalk's earliest product genes. The author makes a comparison: WeChat is closer to the recipient’s perspective, carefully avoids "read/unread" and forced interference; DingTalk is closer to organizational collaboration, emphasizing reach rate, execution rate, and moving things forward.

In the AI era, this difference is further amplified. If AI helps employees summarize messages, filter key points, reduce information overload, it’s like a secretary; if AI helps organizations identify who hasn't replied, who hasn't followed up, which tasks are risky, it becomes part of the management system.

Both capabilities often come from the same set of data. The same "tasks look for people", different roles see different values. Employees see less group chats and red dots, managers see fewer omissions and delays, organizations see higher execution certainty.

This is also the difference between enterprise AI and consumer AI. Consumer AI faces individuals, enterprise AI faces organizations. Organizations have hierarchy, division of labor, and chains of responsibility. Once AI enters workflows, it participates in information distribution, priority sorting, and process advancement.

Previously, AI office discussions mostly focused on model ability, Agent ability, and automation. As AI begins to enter real organizations, a new issue is emerging: most organizations today are still built around people; management, collaboration, and responsibility chains are all designed around humans.

AI can help organizations gain stronger visibility and execution certainty, but it’s hard to immediately become the new organizational center. How information is distributed, responsibility divided, decisions formed, and organizations operated are questions without unified answers yet.

From this angle, the value of "Inside DingTalk" is not in proving why ONE didn't become DingTalk’s new entry point, but in recording the collision of AI workflow entering real organizations.

ONE didn't give answers, but it exposed the questions ahead of time.

The next competition in AI office software may not only happen at the model layer, nor only in the number of Agents. How to let AI integrate into organizations while keeping organizational operation stable is becoming a common issue for all enterprise software.

Risk Alert and Disclaimer ClauseThe market carries risks and investment needs caution. This article does not constitute personal investment advice, nor does it take into account the special investment goals, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Invest accordingly and take responsibility for your actions. ```