A lengthy article titled "Inside DingTalk" has been circulating within product circles and the enterprise services industry since June 2026.
The author, claiming to have been involved in the development of DingTalk's core confidential Project ONE, provides a first-hand account of the project's journey from inception and launch to its gradual marginalization.
According to the article's timeline, Project ONE was born in April 2025 and officially debuted in August, once reaching approximately 3 million daily active users. It was the most significant AI product initiative following the return of the executive known as "Wu Zhao" to DingTalk.
By product definition, ONE aimed to reorganize work scattered across messages, meetings, approvals, tasks, documents, and schedules. The goal was for users to see their most important daily tasks immediately upon opening DingTalk, shifting from "people finding work" to "work finding people," akin to an AI-powered office version of "TikTok."
If successful, this vision would have altered more than just a product feature; it would have impacted how organizations access information, allocate attention, and advance work.
Many view "Inside DingTalk" as evidence of DingTalk's struggles in its AI transformation, while others see it as a profile of the organizational direction after Wu Zhao's return.
The article reads more like an observational record written by a participant after leaving the project, carrying personal perspective and inevitable emotions and biases. Precisely for this reason, it offers a viewpoint rarely accessible from the outside.
Over the past year, major AI model companies have been discussing Agents, workflows, and AI-native applications. Product launches often showcase polished demos, but few are willing to detail what truly happens when these technologies enter real enterprise organizations.
Project ONE provides a case study.
On the surface, it's the story of an AI project's rise and fall. On a deeper level, it documents the real friction between an enterprise organization and an AI workflow.
The Ambition of Project ONE
To view ONE merely as an AI feature within DingTalk would be to underestimate its ambition.
For over two decades, enterprise software has followed a tool-based logic. Messages are on the messages page, documents on the documents page, and approvals in the approval center. Users actively search for, organize, and judge information before deciding on the next action.
ONE sought to change this process. It aimed for AI to first understand the work and then reorganize it for presentation. Users wouldn't need to sift through dozens of group chats or constantly switch between approvals, tasks, and schedules. The AI would pre-filter and prioritize, proactively surfacing important information.
From a design perspective, this involved changing the entry point for office software. The ONE team internally envisioned several typical scenarios: upon starting work, AI would have already organized the day's meetings and tasks; during work, AI would actively identify and remind about critical items; before finishing, AI would help users check for oversights to avoid missing important information. It would also recommend industry trends, competitor information, and research materials based on the user's work content.
This design was based on a judgment: the future competition in enterprise software is shifting from feature quantity to contextual capability.
With the rise of the Agent wave, users expect AI to understand their work environment, grasp organizational context, and drive tasks to completion. In this regard, DingTalk possesses a natural advantage. It holds the corporate directory, organizational relationships, message chains, approval systems, meeting systems, and document systems. A typical AI assistant might not know who is connected to whom or which messages impact project progress, but DingTalk does.
In the traditional internet era, the entry point meant traffic. In the AI era, the entry point is beginning to mean context. This was also a key reason for Wu Zhao's renewed focus on AI.
The Burden of Multiple Objectives
Project ONE's subsequent trajectory cannot be simply attributed to its product design alone. "Inside DingTalk" mentions that from its inception, ONE carried multiple expectations.
First were user expectations. With an increasing flood of messages and increasingly cluttered group chats, many people's first daily action upon opening DingTalk is switching between dozens of windows. ONE hoped to help users reorganize these fragments, freeing work from being driven by message flows.
Second were product expectations. The era of large models had arrived. DingTalk needed a new AI entry point and a new product narrative. For DingTalk after Wu Zhao's return, ONE also carried organizational significance. It wasn't just a feature project; it was hoped to become the representative work of a new round of AI transformation.
Furthermore, there were commercialization expectations. AI requires computing power and real-world scenarios. DingTalk naturally possesses a massive volume of work scenarios, making it an important testing ground for AI implementation. The problem was that while each of these goals was valid individually, they might not be achievable simultaneously through a single product.
ONE aimed to be the entry point everyone opened daily while solving deeply complex problems; it needed to cover most users while proving commercial value; it bore strategic significance while carrying growth metrics. A product for a general user base typically addresses common needs. Scenarios willing to pay continuously often come from specific industries and specific business functions.
Consequently, ONE kept shifting between several directions: workflow, content stream, knowledge service, Agent entry point, AI workbench... Each direction was justifiable, yet each was difficult to perfect.
In retrospect, ONE was burdened with more than just a product goal. It was hoped to be DingTalk's AI transformation flagship while also being tasked with exploring new AI-native organizational forms.
For DingTalk, ONE was both a product and an experiment.
The Organizational Experiment and Friction of AI
The core discussion in "Inside DingTalk" is what happens when AI enters an organization.
Since its inception, DingTalk has been software with distinct management attributes. Whether it's DING messages, read/unread receipts, approval flows, or attendance tracking, these correspond to the deterministic needs of organizational management: ensuring notifications are seen, assigned tasks are progressed, and project status is known.
These needs formed DingTalk's earliest product DNA. The author draws a comparison: WeChat is closer to a recipient's perspective, deliberately avoiding read receipts and minimizing intrusive notifications; DingTalk aligns more with an organizational collaboration perspective, emphasizing reach rates, execution rates, and driving tasks forward.
In the AI era, this difference is further amplified. If AI helps employees summarize messages, filter key points, and reduce information overload, it acts more like a personal assistant. If AI helps the organization identify who hasn't responded, who hasn't followed up, and which tasks are at risk, it becomes part of the organizational management system.
Both capabilities often stem from the same dataset. The same concept of "work finding people" holds different value for different roles. Employees see fewer group chats and notification badges; managers see fewer oversights and delays; the organization sees greater execution certainty.
This also highlights the difference between enterprise AI and consumer AI. Consumer AI serves the individual; enterprise AI serves the organization. Organizations have hierarchies, divisions of labor, and chains of responsibility. Once AI enters the workflow, it participates in information distribution, priority setting, and process advancement.
Past discussions on AI in the office focused more on model capabilities, Agent capabilities, and automation. As AI begins entering real organizations, a new question emerges: most organizations today are still built around people, with management systems, collaboration mechanisms, and responsibility chains designed for humans.
AI can help organizations gain greater visibility and execution certainty, but it cannot immediately become the new organizational center. Questions about how information is distributed, responsibility is assigned, decisions are formed, and organizations operate currently lack unified answers.
From this perspective, the value of "Inside DingTalk" lies not in proving why ONE failed to become DingTalk's new entry point. It is more a record of the collision process when an AI workflow entered a real organization.
Project ONE did not provide the answers, but it surfaced the problems early.
The next phase of competition in AI office software may not be confined to the model layer or the number of Agents. The challenge of integrating AI into organizations while maintaining their stable operation is becoming a common issue for all enterprise software.
Comments