Wu Zhao's Probationary Period Isn't Over Yet

In 1975, Kodak engineers invented the first digital camera, but company executives chose to lock it away in a safe, as film was the company's cash cow.

Thirty years later, Kodak was engulfed by the digital wave and went bankrupt. Meanwhile, Apple, in 2007, launched the iPhone, which completely killed off its own iPod business and built a new empire.

The rise and fall of companies can hinge on a single decision: whether to be a gentle reformer or a radical revolutionary. That is the question.

On June 4th, a 75,000-word essay titled 'Inside DingTalk' by Teng Yaxin appeared on DingTalk's internal network. Spanning 105 pages and eight chapters, with extensive literary metaphors, it provided a near-panoramic account of an AI project from its inception in April 2025 to its eventual contraction.

The author, an AI product manager in DingTalk's Wukong Business Unit, joined the core confidential project ONE after starting in 2025. Defined as 'DingTalk's Agent-driven work information flow for the AI era,' the product aimed to shift users from 'people seeking tasks' to 'tasks finding people,' becoming DingTalk's 'new homepage' in the AI era.

However, from its launch in August 2025 to its eventual contraction and integration into Wukong, ONE's core lifecycle lasted just around ten months.

At its peak, the project had about 3 million DAUs—a respectable figure for any AI startup product but fell short of expectations for a flagship project bearing the mission of 'DingTalk's new AI homepage.' Ultimately, ONE contracted, with its team and resources fully migrating to 'Wukong Agent.'

What exactly happened during those ten months may never be fully clarified, but the author summarized at the end: The problems cannot be reduced to 'the team not trying hard enough.' On the contrary, many issues arose from overexertion, which already says something.

01 Unavoidable Issues

From the outset, the ONE project was burdened with too many missions.

At the user level, it aimed to address the pain point of scattered work information within DingTalk, reducing omissions and searches. At the product level, it needed to create a new AI-era entry point for DingTalk, showcasing its AI capabilities externally. At the organizational level, it carried the function of rallying team morale and reshaping DingTalk's image. At the commercial level, it had to find scalable landing scenarios for AI capabilities, achieving a triple win for users, products, and the group.

A single product simultaneously bearing user value, AI narrative, organizational proof, and commercialization expectations—each objective had its legitimacy, but when bundled into the same delivery cycle, problems emerged.

The long essay mentions that during development and operations, the product positioning constantly wavered, struggling to balance managerial and ordinary employee needs, as well as between senders and recipients of information. This ambiguity directly affected design and iteration rhythms.

Driven by a high-pressure 'daily release' iteration mechanism, the team could only prioritize visible surface features, leaving underlying infrastructure chronically deficient. The product continuously adjusted direction, ultimately exhausting its initial momentum through constant pivots.

Another former DingTalk employee, 'Chen Ming,' offered a similar observation when commenting on the essay: 'Too often, there are too many intentions behind actions, with impure purposes, leading to completely distorted movements and rhythms. Sometimes, expectations for new things are too high, demanding an ALL-IN approach and disruption from the start. New things need time and space to grow. It's not about how bold one is; productivity doesn't scale with audacity.'

The ONE project also touched on a more fundamental proposition in product design logic: power distribution behind efficiency.

The essay's author expressed their judgment: 'Visibility' in work products is never neutral. Seeing a message may imply read receipts; seeing a to-do may imply responsibility; seeing a meeting may imply preparation.

ONE's core design philosophy was to have AI proactively aggregate work information, discovering, organizing, and advancing tasks for users. When AI presents a task to users in advance, it's not just about calculating reading efficiency but also responsibility costs.

This judgment strikes at the fundamental dilemma of AI office products.

From its inception, DingTalk has faced an unresolved structural contradiction: the paying party is corporate management, whose core need is control; the users are ordinary employees, whose core needs are autonomy and work flexibility. These two forces inherently clash.

This is a contradiction embedded in the product's DNA, not something any team or project phase can resolve.

Under this structure, which direction would AI's 'proactive service' logic lead? Does the system judge 'important information' and prioritize push (push) from the employee's perspective or the manager's? Does AI's good intention of reducing user burden inadvertently become a tool for managers to enhance reach?

These questions remained unanswered by the ONE project. However, the clues it provides suggest that when 'tasks finding people' becomes reality, it's unclear whether the to-dos pushed by algorithms are reminding employees or transmitting 'you must complete this' responsibility pressure.

02 Strategic Choices Behind Leadership Changes

From Ye Jun to Wu Zhao, DingTalk underwent two starkly different transformations.

Ye Jun's era was 'reformist.' In March 2022, DingTalk officially commenced commercialization, shifting core metrics from DAU to revenue and active enterprise numbers, establishing an open ecosystem strategy of 'PaaS First, Partner First.'

By FY2025, DingTalk's subscription revenue exceeded 3 billion yuan, ARR surpassed $200 million, user scale grew from 300 million to 700 million, and enterprise organizations increased from 15 million to 25 million.

This path's logic was optimization within the existing framework, buying time with patience. However, 'reform' also faces ceilings. Functionality-driven models seemed slow in the AI era.

After Wu Zhao's return, he pivoted to a revolutionary path. He transformed DingTalk from a feature-oriented collaboration tool into an 'AI-native product,' building the Agent OS intelligent operating system, with AI agents as the core, capable of proactively understanding enterprise intent and autonomously executing tasks.

Simultaneously, he tightened management discipline. Ifeng Technology reported that DingTalk's lunch break was compressed to 45 minutes, with full attendance required by 9 AM, a 9 PM evening review, and the team shrinking from over 1,900 to around 1,600 within six months.

In December 2025, he launched over 20 AI products in one go, and in March 2026, introduced the enterprise-grade AI-native work platform 'Wukong,' officially renaming DingTalk's business group to the 'Wukong Business Unit.'

A detail from 'Inside DingTalk' about interviews is particularly thought-provoking: The author was told during their interview to complete a 'major assignment'—invite six or more family members to join DingTalk and build a family tree organization. After joining, one of the probationary assessment criteria was to serve an enterprise to V6 1000 points, a level achieved by less than 2% of all platforms.

These details collectively outline an organizational culture profile: using high-intensity, high-standard 'obedience tests' to screen and train team members, adapting them to the enterprise's work rhythm and cultural norms.

At a specific stage, this approach is efficient—it ensures organizational discipline and execution speed.

However, when this test-like pressure permeates every aspect of product development, it may come at a cost: employees' focus shifts from 'making a good product' to 'completing processes' and 'avoiding mistakes,' with creativity eroded by layers of approval and dense (frequent) reporting.

This raises a deeper strategic question: When a company faces pressure from technological cycle shifts, what kind of leader does it need? A 'revolutionary' capable of launch (mobilizing) comprehensive change, or a 'reformer' who understands how to gradually advance improvements within existing foundations?

The answer may not be as simple as an 'either-or' choice. However, the ONE project's experience suggests that rushing to prove a revolutionary stance without careful consideration of organizational resilience may incur high internal consumption costs.

03 A Moment of Self-Examination

Chinese business history lacks no contrasting examples of 'revolutionaries' and 'reformers.'

In 1993, when Lou Gerstner took over near-bankrupt IBM, he chose not to dismantle the company into independent units for sale but instead preserved IBM's overall structure against consensus, rebuilding foundational capabilities step by step from operational profits, ultimately pulling IBM out of the mire.

He didn't shout 'disrupt everything' upon taking office but spent a decade transforming IBM from a hardware company into a services and solutions giant. 'Reformers'' patience can also make history.

Similarly, in the 1990s, facing Microsoft Office's dimensionality reduction strike, Kingsoft founder Qiu Bojun locked himself in a hotel room in Caiwuwei, Shenzhen, for 14 months, writing 122,000 lines of code to single-handedly maintain WPS's lifeline.

However, Kingsoft truly emerged from predicament (predicament) only after Lei Jun took over and spent nearly two decades through steady product iterations and differentiated competition.

One company chose strategic resolve amid upheaval; the other held onto ideals in desperation. Looking back, it's hard to simply judge which path was superior.

Today, DingTalk stands at a similar crossroads. It has over 700 million users and approximately 26 million enterprise organizations, holding a 32.7% share in China's collaborative office market, ranking first in the industry—but this position is not secure.

Feishu achieved subscription revenue approaching DingTalk's 3 billion yuan level with around 12 million users, with monetization efficiency far exceeding DingTalk's. DingTalk's B-end AI transformation has yet to find a clear model, while C-end users' identification with DingTalk remains weak.

The emergence of 'Inside DingTalk' is less an accusation against Wu Zhao personally than a collective examination of DingTalk's transformation path. An internal 75,000-word reflection sparking such widespread resonance indicates it has voiced what many wanted to say but dared not.

As Lu Xun once said in 'What Happens After Nora Leaves?,' 'The most painful thing in life is waking up from a dream with no path forward. Dreamers are happy.'

For DingTalk, it has awakened to the urgency of transformation. The question is which path to take: comprehensive change through revolutionary self-amputation or gradual progress through reformist tinkering.

This is not just DingTalk's dilemma but a common proposition for the entire Chinese internet industry in the AI era.

- END -