By Bishan

Source: Bowang Finance

On June 4, Teng Yaxin (alias Yousu), former product manager of DingTalk, published a 75,000-word resignation essay titled 'Inside DingTalk' on Alibaba's intranet. This lengthy piece was not just an emotional outburst but a firsthand testimony with timestamps, project milestones, and meeting records. The essay named DingTalk CEO Wu Zhao (Chen Hang) 73 times, laying bare the high-pressure management during the closed development of the ONE project.

On June 10, Alibaba's Partnership Committee released 'Growth with Empathy and Integrity Defines Alibaba's Culture,' directly criticizing the management culture of a single business line—a rare move. On June 11, Wu Zhao stepped down, and Chen Yusen, born in 1992, took over as CEO. From the essay going viral to the top-level decision and the leadership change, the entire process took less than seven days.

On the surface, this was a high-pressure management failure. Digging deeper, Wu Zhao's management and product philosophies had both become obsolete in the AI Agent era. He was not a villain but an outdated figure—a founder with remarkable achievements from 0 to 1 who, upon his return, pushed the organization to the brink of collapse.

01

70,000 Words, 73 Mentions

Teng Yaxin's essay quickly surpassed 10,000 reads on the day of its intranet release. No one expected it to trigger an organizational earthquake.

The essay's core focused on the ONE project—DingTalk's most critical product iteration in 2025. During closed development, the team averaged 15-hour workdays, with the most extreme month allowing only 1.5 days off. Yousu fainted twice during the project, with the second incident requiring emergency transport to Zhejiang University First Hospital, where she was diagnosed with respiratory alkalosis and hypotension. Another product manager who joined the ONE project resigned due to cardiovascular issues immediately after the product launch.

Wu Zhao's management details were meticulously reconstructed in the essay. The entire team started work at 9 AM, with lunch breaks reduced to 45 minutes and evening wrap-up meetings at 9 PM. Management conducted lunchtime patrols, requiring employees caught using WeChat to write public self-criticisms in group chats. The technical team faced code volume audits, with 'programmers writing zero code facing elimination.' All management staff learned Python and took exams, while product managers were required to visit enterprises weekly for co-creation.

The most poignant passage occurred one evening at 11 PM. A product manager prepared to leave work when Wu Zhao publicly stated the next day, 'As a product manager, leaving at 11 PM? Should I buy you a cot for the project room?' Thereafter, Wu Zhao and HR's midnight patrols of the project room became routine during the ONE project.

The interview process was also detailed in the essay. Candidates were required to complete the 'Family Tree on DingTalk' assignment—adding family members to DingTalk and establishing a family tree organization with at least six people. Yousu faced repeated questioning from Wu Zhao for failing to meet the target: 'Are there more people on your father's side? Your mother's side? Are your grandparents still alive? Can you really not find six family members to join DingTalk?'

Former DingTalk vice president Ma Ruila subsequently published 'Outside DingTalk' in support. He had experienced the same phase: working seven days a week, returning home at 2 AM daily, and sleeping only five hours. He wrote, 'A young product with such innovative ideas ultimately needed 70,000 words to liberate herself from a system.'

On June 10, Alibaba's Partnership Committee responded with unusually direct language. The internal article titled 'Growth with Empathy and Integrity Defines Alibaba's Culture' rarely pointed fingers at a single business line's management style, labeling it 'not representative of Alibaba's culture.'

The next day, Wu Zhao stepped down, and Chen Yusen took over. The entire process unfolded rapidly, like a system patch update.

But this is not just a story of 'high-pressure management failure.' The 70,000-word essay hides a more pressing question—why did a founder with remarkable achievements from 0 to 1 push the organization to the brink of collapse upon his return? The answer may lie not in management manuals but in technological paradigm shifts. High pressure can force deliveries but cannot cultivate the organizational DNA needed in the Agent era.

02

Wu Zhao Was Not a Villain but an Outdated Figure

In 2014, after 'Laiwang' failed against WeChat, Wu Zhao led a team of fewer than ten people into enterprise office solutions. No one favored this sector—B2B was slow, dirty, and tiring, far less glamorous than B2C. Yet he created features like 'Ding,' read receipts, and approval workflows, carving out a market share from WeChat. By 2025, DingTalk had surpassed 700 million registered users, served 26 million enterprises, had approximately 200 million monthly active users, and held a 32.7% market share—firmly leading the industry. These achievements are undeniable.

High-pressure management was effective in specific stages. When directions were clear, teams were small, goals were singular, and products were built from scratch, strong goals, execution, and discipline could maximize combat effectiveness. Early-stage DingTalk epitomized this: everyone knew what to do and only needed to sprint forward. High pressure acted as a whip, driving horses with clear directions to run faster.

The premise was that execution was needed after directions were set, not creativity when directions were uncertain.

By 2025, DingTalk had nearly 2,000 employees and 700 million users, forming a mature, massive organization. The startup-era filter logic—996 workweeks, forced attrition, and overtime rankings—was imported intact. The result? Insiders revealed severe core team attrition after Wu Zhao's return: 'In the past, overtime could genuinely add 1,000 to 2,000 lines of code. Today, in the AI era, those extra lines hold no value.'

Wu Zhao's return was driven by intense 'proof-of-self' desires. After leaving Alibaba, he founded HHO (Two Hydrogen One Oxygen), focusing on cross-border e-commerce and digital light earphones (GPods), which struggled. After burning through RMB 15 million in registered capital, he relied on Alibaba's financial support to buy out investor shares and return. He needed to reprove himself—this anxiety manifested in more aggressive management: higher targets, stricter attendance, and stronger attrition policies. He set DingTalk an aggressive goal of 'achieving break-even within 12 months.'

Objectively, his post-return achievements were undeniable. Within four months, the team analyzed 1,850 user demands, resolved 574 long-standing issues, and overhauled over 20 product lines. This iteration density was extremely rare in DingTalk's recent three-year history. Wu Zhao demonstrated in his own way that he remained capable of hard battles.

The problem was that the tactics from 2014 had become obsolete by 2025.

High-pressure-driven approaches suitable for a 10-person team caused systemic collapse in a 2,000-person organization. Startups relied on manual code writing, but in the AI era, tools like Cursor and Codex could generate equally high-quality programs in minutes. When Agents could automate data organization, report generation, and approval workflows, continuing to drive innovation through overtime hours represented a fundamental shift in underlying logic. Wu Zhao did not lose to a competitor but to the era itself.

His product philosophy faced even more fundamental challenges.

03

The Agent Era Doesn't Need 'DingTalk'

After OpenClaw's explosion in popularity, Alibaba internally advanced three core AI Agent products.

DingTalk's 'Wukong' positioned itself as the world's first enterprise-grade AI-native work platform, natively integrating thousands of DingTalk's enterprise capabilities and enjoying significant internal traffic and resource allocation. 'QoderWork' entered through AI programming tools, attempting bottom-up enterprise penetration. Chen Yusen's 'MuleRun' (internal codename 'Run, Mule!') launched latest but received the best business feedback. Insiders said, 'Wukong enjoyed all internal advantages but faced the most internal criticism.' 'MuleRun received the best feedback but launched latest.'

Behind these three products lay a clash of two product philosophies.

Wu Zhao's 'Wukong' followed a clear logic: DingTalk as the gateway, AI upgrading DingTalk. The core assumption was that enterprise office gateways remained DingTalk, with AI intelligently upgrading this gateway. Users first opened DingTalk, then let AI assist them. This top-down approach emphasized platform and integration.

Chen Yusen's 'MuleRun' took the opposite approach: Agent as the gateway, with DingTalk merely one of many tools invoked. AI agents themselves became new work gateways, with traditional enterprise software just one of many tools called upon by agents. Users no longer needed to open any software—they simply told agents what they needed. Chen Yusen described it as 'the world's largest labor outsourcing company,' enabling anyone without programming skills to create their own AI agents by clearly describing daily workflows.

One represented 'AI-enhanced DingTalk,' the other 'Agent-replaced DingTalk.' The 24-hour lightning leadership change essentially reflected Alibaba's choice between these two paradigms.

More critical than management collapse was the depreciation of DingTalk's core assets. In the past, DingTalk's most valuable features were messaging, approvals, collaboration, and knowledge bases—users needed to open DingTalk to work. But in the Token economy, AI agents could directly invoke these capabilities, completing tasks for users. People no longer needed to manually open software, switch pages, send messages, or fill out approvals. Agents called APIs, completed tasks, and humans only reviewed results. Software value shifted from 'enabling humans to operate more efficiently' to 'being callable by agents and completing tasks for humans.'

DingTalk's surface-level data looked impressive: 700 million registered users, 26 million enterprises. Yet only about 190,000 enterprises paid, with a conversion rate below 1% and an ARR of approximately $200 million.

In comparison, Feishu had 30 million monthly active users (less than one-sixth of DingTalk's) but an ARR exceeding $300 million (nearly three-fourths of DingTalk's) and a payment rate of 18-22%—18 times higher than DingTalk's. Feishu proved product value through high payment rates, while DingTalk proved gateway value through free scale.

In the Agent era, gateways themselves might lose value.

Alibaba had already made strategic adjustments. In March 2026, it established the Alibaba Token Hub (ATH) business group, incorporating DingTalk. In June, it founded the Token Foundry business unit, with CEO Wu Yongming directly in charge. The strategic direction was clear: advance AI commercialization with Tokens as the core, transforming DingTalk into the underlying infrastructure for AI agents to invoke enterprise capabilities. Chen Yusen's core argument was: 'The efficiency gap between AI-native teams and non-AI-native organizations can reach tenfold.'

Wu Zhao's departure marked not just a manager's exit but the conclusion of an era. When collaboration office competition shifted from 'whose IM is better' to 'whose agents can truly work for humans,' high-pressure-driven human resource stacking was destined to give way to lighter, more intelligent organizational forms.

DingTalk will not disappear. But its future form may bear little resemblance to most people's memories.

Shifting focus beyond DingTalk, this is not merely an internal reshuffle at one company.

All traditional SaaS providers still teaching users how to click menus and configure workflows face the same question: If agents can directly complete tasks, why should humans learn to use your software? DingTalk's reality—supporting 700 million users yet failing to surpass a 1% payment rate—lays bare this question's cruelty across the industry. Organizational competitiveness is shifting from 'how tightly one manages people' to 'how capable agents are'—a fundamental power transfer.

Chen Yusen inherits not just a product but an experimental field for an era. MuleRun proved his agent-building capabilities, but the complexity of 700 million users will not follow startup scripts. Alibaba is betting on a paradigm, not an individual.

The true variable in this experiment may never reside at the product level but within people themselves.

Teng Yaxin wrote 75,000 words not to vent but to liberate herself from a system. Ma Ruila used resignation to achieve the same. Wu Zhao was not a villain; he truly built DingTalk's empire in 2014—but that logic's cost was reducing humans to manageable code, requiring 70,000 words to prove one's humanity.

This signals something more fundamental than management collapse. When agents can complete data organization, report generation, and approval workflows for humans, people should transform from 'overmanaged code workers' to 'agents' masters.' The next generation may not need to write 70,000 words to liberate themselves—a single command like 'Build me an approval system' could reconstruct entire workflows in minutes through agents.

Chen Yusen's real challenge is not making DingTalk more usable but eliminating the need for humans to 'be managed by DingTalk.'