Kill DING

In June 2026, DingTalk experienced an earthquake: a 75,000 - word article titled "Inside DingTalk" by a former product manager went viral, the Alibaba Partnership Committee publicly criticized the management in an unusual way, the founder Wu Zhao stepped down, and the 90 - generation tech geek Chen Yusen took over. It's time for this national app with 800 million users to kill its past.

From a data perspective, DingTalk remains the king in China's office management software field: the cumulative number of users has exceeded 800 million, with a monthly active user base of about 200 million. It holds the top position in the domestic collaborative office market with a 32.7% share, closely followed by Enterprise WeChat (23.4%) and Feishu (18.9%). There are over 26 million enterprise organizations and nearly 200,000 paying customers. However, these numbers themselves have pinned DingTalk in place - they belong to the scoring method of the old system: competing on the number of users and the breadth of coverage.

This old system once won handsomely. In the Chinese enterprise service market in 2015, the scarcest thing was "certainty": there were missing processes and weak IT. What bosses were most anxious about was "whether the other party actually followed through". DingTalk provided an answer with DING messages and read/unread status. But the winning conditions have changed. The problem with the old system is not that it's wrong, but that its applicable scope is narrowing, and organizational inertia prevents it from actively shrinking.

The inversion of commercialization efficiency is the deepest thorn. In fiscal year 2025, DingTalk's subscription revenue exceeded 3 billion yuan; Feishu has only 30 million monthly active users, about one - seventh of DingTalk's, but its annualized subscription revenue in the same period has exceeded 2.1 billion yuan. With a user scale seven times that of Feishu, DingTalk's revenue is only less than half higher.

The deeper dilemma lies in the customer structure: the competition for large customers has become the norm. Feishu, with emerging industry benchmarks such as NIO, Li Auto, and Xiaomi Auto as beacons, is penetrating into traditional large customers. Although DingTalk still has an advantage among the existing Fortune 500 companies and specialized and sophisticated enterprises, its position in emerging tracks is being eroded. The customer base of the old system is aging, and the customer base of the new system is shrinking.

Wu Zhao saw the loose nails and picked up the hammer to reinforce them one by one. But the problem is not with the nails, but outside the door. The global enterprise software market is experiencing a fundamental structural shock: Agents are conducting transactions, and value is settled by tasks. The workflow and organizational flow that DingTalk relied on to rise are being bypassed, compressed, and absorbed. DingTalk's efficiency inversion is just an early preview in China of this global paradigm shift.

Wu Zhao's 437 days: The founder pinned by the old system

On March 31, 2025, Wu Zhao was recalled by Alibaba CEO Wu Yongming to take charge again. Alibaba needed a "disruptor from 0 to 1". However, his way of dealing with the dilemma was exactly the most familiar moves in the old system he forged ten years ago: all employees were required to start work at 9 am, the lunch break was shortened to 45 minutes, evening meetings were held at 9 pm, and there was only one day off on weekends; all management staff were required to learn Python, and code volume was checked, with those who couldn't code being eliminated; holiday benefits were cut, and the salaries of some positions were adjusted; employees were required not to add WeChat privately and could only say "Sorry, I only use DingTalk"; there was even the "Wangshu Action" of inspecting workstations in the early morning and benchmarking against competitors' lights - out times. According to the disclosure in "Inside DingTalk", during an interview, he asked, "Can't you really find six family members who can use DingTalk?" What was a touchstone in 2014 was regarded as PUA in 2025. The same set of methods was called the "Iron Army Culture" ten years ago, but today it is defined as "workplace PUA". It's not that the managers have become worse, but that the social psychological foundation on which this approach relies no longer exists.

What's more fatal is that he was managing DingTalk in the same way as he managed enterprises with DingTalk: DING became building inspections, and read status became code assessments. He was pinned down by the "strong - reach" logic he invented.

The root cause of the failure of this approach lies in the fundamental changes in the productive relations. AI is replacing executive labor, and the value of people is shifting towards creativity, and creativity cannot be "managed" through clock - in and building inspections. At the same time, the social foundation is also shifting: the early - stage team had a "partner" mentality, fighting for survival; now the 20,000 employees have a "worker" mentality, paying more attention to personal rights and mental health.

The deeper root cause is that DingTalk was born out of a "revenge - style entrepreneurship" after the failure of "Laiwang" in 2015. The early - stage team believed in the "extreme execution" of the Zhonggong Iron Army. The advantages in the 0 - to - 1 stage fundamentally conflicted with the creativity needs of knowledge workers in the 1 - to - 100 stage. Wu Zhao's management method was to transplant the methods of the B2B sales team directly to the product R & D team. This is not just personal inertia but also a projection of organizational genes.

On June 10, 2026, the Alibaba Partnership Committee responded unusually: "Under no circumstances should such a management method appear. In the AI era, innovation does not rely on high - pressure and mechanical execution, but on employees' passion and creativity." The core meaning of this sentence is just one: If DING doesn't die, DingTalk will have to die.

The backlash of the product: How the old system pins everyone down

High - pressure management is just the surface. Deeper down, the management method inside DingTalk and its product appearance are projections of the same old logic in two scenarios.

DingTalk's product genes are deeply ingrained: it derives functions from the needs of enterprise bosses. DING, read status, enterprise address book, and approvals answer the most basic anxiety: "Did you see what I said?" "Inside DingTalk" states bluntly, "What a product manager has the most difficulty getting rid of is often not failure, but success." Wu Zhao's physical memory is to stand on the sender's side and push things forward with strong reach.

In the enterprise service market, "standing on the side of the payer" is normal. Salesforce also stands on the side of managers, but it provides "empowerment and coordination". DingTalk's problem is that it has shifted from "helping managers advance tasks" to "using technology to achieve panoramic control". Why hasn't Salesforce shifted towards control? The key lies in the relationship between the payer and the user. Salesforce widely adopts "employee - level subscriptions", where the user is the payer, so the product naturally serves the user's efficiency. DingTalk's commercialization is based on unified payment by enterprise managers, so the product logic naturally tilts towards control.

But the winning conditions have changed. AI can undertake executive work, and organizational competitiveness has shifted from execution to creativity. The old system solved the most important problem ten years ago, but it's no longer the problem today.

The non - natural death of ONE: The new product pinned by the old logic

In August 2025, ONE was launched. Its peak daily active users once exceeded 3 million, but it shrank rapidly within 10 months. It was pinned to the pillar of the old system from birth.

The first pillar: The old interaction logic in a new coat. ONE is a short - video - style Feed stream, inspired by Douyin. But on Douyin, you can swipe away what you don't like. Behind the ONE cards are collaborative relationships and deadlines. Clicking on a card means taking responsibility, and marking it as read means fulfilling an obligation.

The second pillar: The continuation of "automatic read". The team proposed a plan of "clicking on a card not counting as read", but it was rejected by the senior management on the grounds of "not harming the rights and interests of the sender". As a result, users actively blocked the ONE entrance.

The third pillar: Multiple personalities under the old KPI. On the user side, it aims to solve the problem of message explosion; on the product side, it wants to break the old pattern with AI; on the organizational side, it wants to return to the landmark project; on the commercialization side, it shoulders the KPI of Token consumption. The old system requires a product to complete four things at the same time, but in the end, it fails to complete any of them.

Beyond DingTalk, Feishu and Enterprise WeChat are also trapped by their respective old systems.

Feishu's dilemma is "diseconomies of scale". With 30 million monthly active users, it has driven an ARR of over 2.1 billion yuan. The commercialization efficiency per user far exceeds that of DingTalk. The problem is that Feishu is using the methodology of the consumer Internet in the To B market. This high - efficiency is based on heavy investment: an excellent product experience requires a large R & D team to support it, and every aspect such as large - customer customization, private deployment, and after - sales training is a labor - intensive investment. The commercialization efficiency is high, but the cost structure is heavier, and the gap of diseconomies of scale has never been closed. Feishu's team of 4,000 to 5,000 people and an annual labor cost of about 4 billion yuan are the prices paid for applying this C - end logic in the B - end battlefield.

Enterprise WeChat's dilemma is the opposite. It is pinned to the "external" by its own genes. Relying on WeChat, it is good at enabling enterprises to connect with customers externally, make conversions, and earn money - this is its natural advantage. But when enterprises need to improve internal efficiency and require AI - driven organizational collaboration, Enterprise WeChat is not the first choice.

Before discussing the replacement of the old with the new, a more fundamental business psychological question is worth asking: What are Chinese enterprises willing to pay for?

The answer is "visible things". Accounts, function modules, storage space - they are perceptible, countable, and can be reported to leaders. When an Agent silently completes ten steps and only returns the conclusion, the enterprise finance department will instinctively ask: "Why should I pay for an invisible process?"

This is not a technical problem but a reconstruction of business trust. It requires a shift from "paying for ownership" to "paying for completion". Salesforce took ten years to make customers accept the subscription system, and SaaS took another ten years to accept pay - by - usage. The "pay - by - task - value" model of Agents will take even longer. And most of DingTalk's 800 million users are still in the first stage. The practical path is: the old system sells connections, and the new system sells results. The old and the new are not replacements but layers - "Property management takes a back seat, and investment promotion and operation come to the front".

Changing the leader: Does Chen Yusen have a solution?

On June 11, Chen Yusen took over. He was born in 1992 and is the youngest division CEO at Alibaba. Chen Yusen led MuleRun within Alibaba Cloud, which is positioned as an AI Agent trading market. MuleRun's strategy is a disruptive turn: from "humans using tools" to "AI autonomously executing", and from a closed enterprise internal environment to an open global digital labor network. MuleRun has been quickly implemented in the C - end and extremely small enterprises because there is no organizational politics in these scenarios. But in medium - and large - sized organizations, technology is never the only determining factor. Enterprises' payment logic is shifting from "paying for results" to "paying for compliant delivery results". The way to break the situation is to start from low - risk experimental areas: internal knowledge base Q & A, preliminary reimbursement review, and business travel booking. Build a tolerance for errors and gradually open up the high - level responsibility chain.

But there is a real deadlock here. There is a fundamental conflict between MuleRun's "pay - by - task" model and DingTalk's existing "pay - by - account" model. If an Agent can efficiently complete tasks, one Agent can replace multiple employee accounts - DingTalk's revenue will actually decline. For Alibaba Cloud/DingTalk, pushing the "pay - by - task" model all at once means there will be a huge "revenue gap" in the financial report immediately, because the revenue recognition method is completely different from annual subscriptions to pay - by - time/value, and the smooth SaaS recurring revenue will become volatile. Which CFO would allow this to happen? So, what Chen Yusen faces is not just a matter of personal courage but a systematic financial problem.

What Chen Yusen faces is not a technical choice but a "trolley problem" in business: To let MuleRun truly play its role, he must tolerate it eroding DingTalk's existing revenue; to protect the existing revenue, he can only make MuleRun a beautiful "demo version" and never push it to the core scenarios.

A more hidden risk is the specific mechanism of "reverse assimilation by the old system".

The MuleRun team needs to call DingTalk's organizational data interfaces. But the permissions of these interfaces are controlled by DingTalk's old team. Will they cooperate? If they do, will it be in the way of MuleRun or the old logic? Is Chen Yusen's reporting line independent or does it require layer - by - layer approval? When a certain function of MuleRun conflicts with the KPI of a senior director at DingTalk, who has the final say? These details at the "organizational politics" level often determine the survival of a new project more than technical difficulties.

The rapid shrinkage of the ONE project from 3 million daily active users has already provided a warning - and MuleRun needs to be embedded in a system with 800 million users. This leap in scale itself means exponentially increasing organizational friction and path dependence.

Chen Yusen's real challenge is not whether MuleRun is useful, but embedding the lightweight verification model into the heavy - weight old system without being reverse - assimilated.

Global samples: How AI disrupts office management software

To understand the deep - seated challenges Chen Yusen faces, we need to see the global technological trends clearly. AI is reconstructing the underlying logic in various ways.

First, from the "operation interface" to "intention execution". The software interface is losing its necessity.

Second, from "pay - by - account" to "pay - by - value" billing. Salesforce Agentforce, Zendesk AI Agents, etc. have started testing the "single - time automated solution" billing model. MuleRun is a more radical experiment: more than 43% of paying users have a monthly consumption of over $200, covering 43 countries. This pricing model turns AI agents from cost centers into core productivity tools.

Third, the software moat is being filled by AI. After the launch of ChatGPT, the number of questions on Stack Overflow dropped sharply. The platform's accumulated corpus is absorbed by LLMs, bypassing the original entrance. Scale itself is not a barrier.

The root of the gap lies not in technology but in the value anchor: the old system's anchor is "personnel", and the new system's anchor is "tasks".

However, the AI Agent trading market faces four hurdles: The first is permissions and responsibilities. It is necessary to establish human - machine collaborative confirmation and operation log traceability. The second is organizational inertia. An independent evaluation system needs to be set up for Agents, assessing the "efficiency of problem - solving" rather than the "number of executions". The third is large - scale governance. It is necessary to shift from "selling models" to a "data center and security base". The fourth is from the human dashboard to the Agent API. An open market for Agent autonomous collaboration needs to be built.

If DingTalk can complete the shift from a "rent - collecting property" to a "commission - taking mall", it will be a paradigm - level business evolution.

The old system won't disappear, but it must give up the default position

DingTalk's experience this year reveals a shift: when AI undertakes more executive work, the value of "people" focuses on creativity.

The question hangs in the air: When DingTalk no longer needs DING messages, will it still be DingTalk?

My judgment is that what DingTalk needs to kill is never the DING message itself, but its status as the "default way of daily collaboration". Fire alarms, compliance notices, and emergency breakdowns - these scenarios still require "ensuring delivery". The problem