From being invited back to being sent away: Chen Hang's 437 days at DingTalk

437 days.

This is the duration of Chen Hang's second tenure at DingTalk.

From March 31, 2025, when Alibaba announced the acquisition of HHO, and the founder returned to the battlefield after a four - year absence, to June 11 this year when he stepped down as CEO, it was a total of 437 days.

During these 437 days, DingTalk held two important product press conferences and launched "Wukong", the world's first enterprise - level AI - native working platform. There were also two widely circulated long resignation posts, DingTalk received rare and severe criticism from Alibaba partners in 27 years, and Chen Yusen, born in 1992, took over.

Three questions string together the whole story of these 437 days:

What did Chen Hang do wrong? What price did Alibaba pay for this comeback? What was left on DingTalk's chessboard when Chen Yusen, born in 1992, took over?

The story may start from an apartment called Lakeside Garden in 2014.

1. Back to Lakeside Garden

In 2014, in an old apartment at No. 176, West Wenyi Road, Xihu District, Hangzhou, six Alibaba employees gathered around a table for a weekly meeting.

They had just failed once. Their previous product was called "Laiwang" - Alibaba once bet 1 billion yuan, launched a mass - mobilization campaign, required each employee to recruit 100 external users in a month, Jack Ma personally endorsed it, and Liu Chuanzhi, Shi Yuzhu, and Jet Li were invited to join. However, it didn't shake WeChat at all.

After that failure, Chen Hang (with the nickname "Wuzhao") led a few people into this apartment - Lakeside Garden. This place had incubated Alipay, Tmall, and Cainiao successively and was a lucky place for Alibaba.

The interior of the replicated Lakeside Garden in Alibaba's Xixi office area. Photo source: Taken by "Jiazi Guangnian".

In January 2015, the first - generation DingTalk was launched. Its core insight was to reverse - engineer functions from the needs of enterprise bosses. For example, features like "Ding", read/unread status, enterprise address book, and approval - these features, which were later criticized by many as "strong management and strong control", answered the most simple and anxious questions in Chinese enterprises back then: Did the other person see what I said? Is the task I assigned progressing?

This insight was obviously a sharp breakthrough. DingTalk had over 100 million users in the first year and over 300 million in three years.

In DingTalk's recruitment announcements, Chen Hang referred to the team as a "madhouse", and the back of the T - shirt read "BE CRAZY". He himself worked more than 15 hours a day, often from 8 a.m. to the early morning.

His words at the mobilization meeting in 2018, "I don't know what you do when you go home before 10 o'clock", are still a frequently cited joke in the Internet industry.

In 2020, things changed. Alibaba announced the "Cloud - Ding Integration" strategy. DingTalk was upgraded to the "Large DingTalk Division" and incorporated into the Alibaba Cloud Intelligence Business Group, directly managed by Zhang Jianfeng, the then president of Alibaba Cloud Intelligence. Chen Hang's DingTalk was required to be deeply integrated with Alibaba Cloud's proprietary cloud architecture, and the originally independent standardized product route began to conflict with the customized needs of large customers.

In 2021, Chen Hang was removed from the position of DingTalk CEO and became an assistant to Zhang Yong, the CEO of Alibaba Group. Ye Jun took over DingTalk. In the same year, Chen Hang officially left and founded HHO to engage in cross - border business and launched some small intelligent hardware products.

During his years at HHO, the outside world's evaluation was not high. It was a startup with many stories but short of money and didn't achieve impressive results. He later told the media that during his time at HHO, he "learned to strictly control costs and stipulated that only one of the three elevators could be used". These trivial matters are not what executives of large companies need to consider.

But during the years when he left DingTalk, DingTalk's situation became increasingly awkward.

In 2024, it was still the largest enterprise office application in China, with 700 million users and 26 million enterprise organizations. However, its commercialization progress was overtaken by Feishu. During Ye Jun's tenure, 60% of DingTalk's R & D resources were invested in customized services for large customers, and the ISV ecosystem complained that "the platform had become a channel provider".

In February 2025, Alibaba announced an investment of 380 billion yuan to start a three - year AI journey. In the same month, Wu Yongming, the CEO of the group, clearly positioned DingTalk as "Alibaba's most important AI application in the toB field" during an earnings conference call.

A detail is worth pondering: The relationship between Wu Yongming and Chen Hang can be traced back to 1999. That year, Chen Hang joined Alibaba as an intern for the first time, and his mentor was Wu Yongming. Later, he left Alibaba twice and returned twice at Wu Yongming's invitation.

On the evening of March 31, 2025, it was reported that Alibaba had acquired the shares of HHO's investors. After the transaction was completed, Chen Hang became the CEO of DingTalk.

The story of the Bole and the Qianlima has been told many times in the Alibaba system. But few people ask: When a founder is invited back to the company he founded, is he here to put out a fire or to rewrite the script?

2. The "Tightening Spell" in April

Last April, just a few days after Chen Hang's return, there was an uproar within DingTalk.

A series of measures were gradually revealed: Clock - in at 9 a.m., shortened lunch breaks with employees required to be back at work by 1:15 p.m., evening summaries, bans on social media apps like WeChat, Weibo, and Xiaohongshu, and unified external communication phrases to "Sorry, I only use DingTalk".

Even more strict measures included: The technical team was required to check the code volume, and programmers with no code written in the past three months would be eliminated; all management positions were required to learn Python to reduce pure management positions; product managers were required to visit three enterprises per week for co - creation.

Chen Hang himself would patrol the building at 10 p.m. every night and give thumbs - up to the overtime workers.

On social platforms at that time, "DingTalk refugees" became a new label. Some employees changed their profiles to "DingTalk is getting competitive", and asked if other companies were hiring. Someone joked in the anonymous area: "I was rejected when adding someone on WeChat for a blind date. The other person left directly after hearing that I could only use DingTalk."

The author of the 75,000 - word long post "Inside DingTalk", which went viral on June 4, Teng Yaxin (with the nickname "Yousu"), a core product manager of the DingTalk ONE project, recounted in the post her experience of being interrogated like giving a "pledge of loyalty" during the DingTalk interview.

Chen Hang repeatedly asked: "Why can't it be done? Are there still people in your father's family? Are there still people in your mother's family? Are your grandparents still alive? Can't you really find them? Can't you really gather six family members who can use DingTalk?"

Although these questions seemed absurd, when Chen Hang incubated DingTalk in 2014, he used this logic to select like - minded people; in 2025, candidates' first reaction to this interview culture was that it was PUA.

The same action has a different meaning when the target changes.

At that time, Chen Hang also launched a "Ground - Level Movement": He required members of the product, R & D, and operation teams to serve as customer service representatives for two hours each day.

This action was later proven to be the key for Chen Hang to discover DingTalk's "data illusion". Previously, the customer service team reported that "the rate of transferring to human service was only 15%, and all were five - star reviews". However, through on - site visits, Chen Hang found that many users complained about "irrelevant answers", "no response to requests for more than a year", and "unable to find the human service entrance" when consulting.

Chen Hang quickly updated this data: DingTalk's real customer satisfaction was only 30%.

3. AI DingTalk 1.0, DingTalk ONE is Born

The tough - minded Chen Hang quickly achieved results.

On August 25 last year, DingTalk's tenth - anniversary press conference was also the AI DingTalk 1.0 press conference.

On - site, Chen Hang launched five products in one go:

DingTalk ONE: A new - generation interaction entrance that can automatically identify "the most important thing at the moment".

DingTalk A1: An AI recording card 3.8 mm thick, magnetically attached to the back of the phone, equipped with a Hengxuan 6nm chip.

AI Spreadsheet: Allows enterprises to generate AI applications without coding.

AI Recording: Supports cross - language meeting records in 72 languages.

AI Search and Query: A search engine that can search, query, and perform tasks.

Chen Hang said that "DingTalk in the AI era should serve real - world work scenarios". This was one of his few public appearances. But the night before the press conference, he was photographed patrolling the DingTalk campus in the early morning and made it onto the "Most Competitive CEO" trending list.

On the day of the press conference, DingTalk also disclosed a set of key data:

The number of enterprise organizations on DingTalk exceeded 26 million. Among them, the number of paying organizations exceeded 190,000. 79% of the 5,191 A - share listed companies were using DingTalk. The number of AI applications on DingTalk had reached 1.41 million.

DingTalk's customer satisfaction, which was 30% when Chen Hang launched the "Ground - Level Movement" in April, had been increased to 80% after he reorganized the customer service team and established three core teams for data engineering, model training, and effect evaluation. The cost was reduced by 90% simultaneously.

Besides these data, the most notable one was DingTalk ONE.

DingTalk ONE was positioned as "the new entrance of DingTalk in the AI era" and was regarded as the core of AI DingTalk 1.0. At that time, DingTalk ONE was a flagship product with high hopes.

This was a project that Chen Hang started to conceive in April. From April to August, DingTalk ONE went from project establishment to release in less than half a year.

However, this glorious moment didn't last long.

4. The Summer and Late Autumn of ONE

ONE was not a product that was carefully polished.

There is a review of ONE's life cycle in "Inside DingTalk": ONE's life cycle started to be conceived in April 2025, was first publicly introduced at the press conference on August 25, and its peak DAU was stable at around 3 million. It was the first "AI - native" project that Chen Hang promoted after his return.

But ONE was also a project with high fluidity. The design leader left in the second week, and the senior who contacted and recommended Yousu to join the project was transferred to another department in the fourth week. Only three people had been in the ONE project for more than three months, and Yousu was one of them.

ONE had a strong design gene, and the first person in charge was the head of the design center. But during the operation period, the original "card" form gradually evolved into "displaying all important content on one screen".

Obviously, this was a failed attempt.

After the DAU reached 3 million, the retention rate dropped sharply.

At the beginning of 2026, ONE was split.

It's worth thinking about: The rise and fall of ONE is not just a simple story of a product failure.

Its "summer" corresponded to the urgency of Alibaba's AI toB strategy - Alibaba needed a new entrance to undertake the narrative of the Agent era. But its "late autumn" also stemmed from this. From the very beginning, it was destined to meet the dual standards set by the group: it had to be quickly launched to prove the determination of the AI strategy and also had to accept the physical law of the retention rate of AI - native products.

When these two curves intersected, the product always lost.

There is a sentence in "Inside DingTalk" that accurately describes the cost of this efficiency: It was in a hurry to become a new entrance and to prove that DingTalk was not old.

DingTalk is not old. With 800 million users, 26 million enterprises, 1.41 million AI applications, and 190,000 paying organizations... it is still the number one in the domestic office collaboration track.

But what Chen Hang wanted was not just to be number one. He wanted to be "number one in the AI era".

5. The Day DingTalk was Shattered

On March 16, 2026.

Wu Yongming sent an internal letter to all employees, announcing the establishment of the Alibaba Token Hub (ATH) Business Group. This business group takes "creating tokens, delivering tokens, and applying tokens" as its core goals and is led by Wu Yongming himself.

It integrates five major divisions: Tongyi Laboratory (Qwen large - model), MaaS business line (Bailian), Qianwen Division (C - end AI assistant), Wukong Division (B - end AI - native working platform), and AI Innovation Division.

Among them, the "Wukong Division" made its debut for the first time.

24 hours later. On the morning of March 17, the AI DingTalk 2.0 annual new product press conference was held.

On this day, Chen Hang launched "Wukong", the world's first enterprise - level AI - native working platform. Its promotional poster features a cartoon - style Wukong holding a stick standing among a group of shrimp soldiers, and the picture is full of metaphors.

Chen Hang said on stage: "Today, we shatter DingTalk and rebuild it with AI to forge 'Wukong'. In the past, people used DingTalk to work. In the future, AI