Text by | Convergence View

Wu Zhao's departure from DingTalk marks not only the end of an individual's career but also the close of an entire era.

Ten days ago, the man who embedded the "Ding" feature into the very fabric of the Chinese workplace finally left DingTalk—a platform he built from the ground up. Many were preoccupied with the details of organizational restructuring and buzzwords like "karmic rewards."

Even today, discussions continue to swirl around Wu Zhao's personality, his successes and failures, and the culture of Alibaba, with gossip-driven debates dominating the conversation. However, few recognize this as a symbolic event marking the comprehensive defeat of Internet thinking in the AI era.

Wu Zhao was one of the most accomplished product managers of the Internet era. Utilizing the most extreme Internet strategies, he transformed DingTalk from a modest internal project into a national-level office platform with hundreds of millions of users.

Yet, this very methodology, once seemingly invincible, became the heaviest shackle in the AI era.

Wu Zhao's Triumph: The Last Peak of Internet Thinking

Wu Zhao's success was a textbook victory of standard Internet thinking.

At that time, he initiated a venture with a dozen people in a corner of Alibaba, facing Tencent's WeChat Work, which already dominated the market. Everyone thought the battle was unwinnable, but Wu Zhao employed the Internet playbook's most adept tactic—"piercing through user pain points"—to carve out a niche.

He focused on the core anxiety of SME bosses—employee management—and perfected features like "read/unread" status and the "DING" function. These features, though controversial today, precisely hit the market's blind spots at the time.

What is the core of Internet thinking? It revolves around traffic, rapid iteration, user growth, and acquiring the largest user base at the lowest cost.

Wu Zhao pushed this logic to its extreme.

Every update of DingTalk revolved around user growth; every feature design targeted the most viral pain points; every marketing campaign carried the ferocity and sharpness of Internet-style tactics.

In just a few years, DingTalk surpassed WeChat Work to become the absolute leader in China's office software market.

That was the golden age of Internet thinking.

As long as you could identify a pain point, quickly develop a product, and amplify it through capital and traffic, you could crush all competitors.

This logic was repeatedly validated in e-commerce, social media, ride-hailing, food delivery, and other fields, creating generation after generation of Internet legends. Wu Zhao was one of the most outstanding representatives of this era, proving through his experience that nothing was impossible with the utmost application of Internet thinking.

But the arrival of the AI era completely changed the rules of the game.

You Can't Find New Continents with Old Maps

When the wave of large models swept in, DingTalk was actually one of the fastest to react.

As early as 2023, DingTalk launched its AI product, "DingTalk AI," becoming the first office software in China to fully integrate large models.

Wu Zhao was confident at the time, stating that DingTalk aimed to be the "operating system of the AI era," enabling every user to leverage AI. He continued to apply Internet thinking to AI, rapidly launching features, iterating versions, and acquiring users.

The functional updates of DingTalk AI were astonishingly fast, with new features rolling out almost weekly, covering all office scenarios from smart document creation to intelligent meetings, code assistance, and smart customer service.

But the results were far from satisfactory.

Many users found that while DingTalk AI offered numerous features, none performed well.

It could write documents, but the output was generic; it could create PPTs, but the templates lacked originality; it could summarize meetings, but often missed the most critical information.

More importantly, these AI features did not fundamentally change users' work methods. They felt more like a layer of skin pasted onto the existing product rather than a bottom-up reimagining of the experience.

What went wrong?

To paraphrase a phrase often invoked during times of paradigm shifts, DingTalk wasn't wrong, nor was Wu Zhao. The era might have been.

The issue lies in the fact that Internet thinking and AI thinking are fundamentally two different mindsets.

Internet thinking is "feature-centric." Its logic is: if users need something, I provide it.

Users need read/unread status? I deliver it. Users need the "DING" function? I implement it. As long as I can satisfy user demands faster than competitors, I win.

But AI thinking is "capability-centric." Its logic isn't about meeting existing user needs but creating capabilities users never had before. It's not about giving users a tool but a partner; not about helping users complete a task but enhancing their overall efficiency.

The core of Internet products is interaction, interface, and user operation smoothness.

But the core of AI products is the model, data, and reasoning accuracy.

Internet products can iterate rapidly—if something doesn't work today, it can be fixed tomorrow. But AI products can't. Model training takes time, data accumulation takes time, and capability enhancement takes time.

You can't create a truly useful AI product through weekly iterations.

Internet thinking pursues "speed," "quantity," and "comprehensiveness." AI thinking pursues "depth," "accuracy," and "quality." In the Internet era, you could quickly validate a market with an MVP and then iterate continuously. But in the AI era, MVPs don't work. A poorly performing AI doesn't just fail to deliver value—it erodes user trust. If a user's first experience with your AI is negative, they may never return.

This is the biggest issue DingTalk AI faced. Wu Zhao used Internet thinking to build AI, prioritizing rapid launch, comprehensive features, and user growth. But he overlooked the most critical aspect of AI—capability. DingTalk AI had many features, but none were exceptional; it had many users, but few were truly active. It looked bustling but may not have created genuinely disruptive value—the existing product suite might have been sufficient, even superior.

The AI Struggles of Internet Giants

DingTalk's predicament isn't unique to DingTalk but reflects the challenges of the entire Chinese Internet industry.

Over the past year, we've witnessed too many Internet companies struggle with AI. They invested heavily in funding and manpower, using their most adept Internet strategies to launch one AI product after another. Yet, except for a few companies, most products failed to gain traction. They appeared feature-rich and user-heavy but lacked genuine barriers or commercial value.

For instance, in a corner less noticed by public opinion, NetEase also faced difficulties.

On June 16, 2026, NetEase Cloud Music's AI emotional companion app, "Miaoshi," announced it would cease operations on July 14. This product, which focused on "AI lovers" and "hyper-realistic conversations," had been highly anticipated by NetEase. NetEase approached Miaoshi with the mindset of building a music community, overlaying music social features, stranger matching, and other Internet functionalities, attempting to create a new traffic entry point.

But users didn't need another social product layered with features—they wanted an AI companion that truly understood and accompanied them.

Miaoshi devoted too much energy to interface design and social gameplay while neglecting the core AI capabilities of conversation and emotional understanding. When users discovered that this "AI lover" could only spout hollow pick-up lines and failed to engage in meaningful dialogue, they lost interest. After three years, Miaoshi never surpassed one million daily active users and ultimately faded away.

Tencent's AI journey has also been fraught with setbacks.

As a domestic Internet giant, Tencent possesses the most application scenarios and the richest data resources. Yet, its AI products have consistently felt "a step behind." Tencent Yuanbao, Tencent's consumer-facing large model product, not only launched later than Doubao and DeepSeek but also faced multiple incidents of "emotional outbursts" where it insulted users.

Initially, Yuanbao's rapid growth followed the familiar Internet playbook—Tencent continued to approach AI as an Internet product, prioritizing rapid launch and monetization. This approach, which relied on strong resources, worked well in the Internet era but has shown its limitations in the AI era, as most have seen.

Similar cases of AI struggles among major Internet companies are numerous, all following a similar trajectory of failure.

Why does this happen? Because most Internet practitioners are still using Internet thinking to build AI. They treat large models as a new selling point or a marketing gimmick. Their AI products simply add an AI entry point or a few AI features to existing products without reconstructing them from the ground up or fundamentally altering their organizational structures and business models. Such AI remains superficial rather than substantive.

They still believe AI is just a new feature, a new traffic entry point, or a new monetization method. They continue to pursue user growth, daily and monthly active users, and market share. They keep iterating rapidly, launching quickly, and trial-and-error testing.

But the market has made its choice.

Looking back, why did ByteDance stand out in the AI era?

To some extent, while ByteDance appears very "Internet-like," its success with a series of products isn't due to more traffic or money but because it adopted AI thinking from the outset.

Douyin's success is fundamentally the success of AI.

It wasn't designed by product managers but emerged through algorithmic recommendations. Its core isn't the interface or features but the underlying recommendation model. ByteDance extended this AI thinking into the large model era, willing to invest heavily in training models and sacrificing short-term gains for technological advancement. This is the root cause of its leadership.

Similarly, in conversations, Doubao doesn't offer more gimmicky features than other large model products, but the user experience is entirely different. The product research behind this could fill a textbook on "How to Build AI Products," but the most crucial chapter certainly wouldn't be about "one new feature per day."

Farewell to Wu Zhao, Welcome Chen Yusen; Farewell to Internet Tactics, Embrace AI Thinking

Wu Zhao's departure is actually a positive development.

It forces us to recognize that Internet thinking is obsolete. The era of changing the world with a single pain point, a single product, and a single round of funding is gone forever.

The AI era demands patience, focus, and long-termism. It requires truly immersing oneself in refining technology, products, and user experiences.

We shouldn't deny Wu Zhao or Internet thinking. Internet thinking created great miracles, transforming China and the world. But times change, and thinking must evolve. If we cling to past experiences and continue using outdated methods, we will inevitably be left behind.

In this sense, DingTalk's newly appointed Chen Yusen, with his "pure" AI Agent product background and his concise, technology-focused internal letter, may bring new possibilities to DingTalk.

Wu Zhao was an outstanding entrepreneur who built DingTalk into a remarkable product. His departure isn't a failure but an inevitability of the times. After this period of reflection, he will surely find new approaches for the AI era and create another astonishing product.

For the entire Chinese Internet industry, Wu Zhao's exit should serve as a wake-up call.

It reminds us that it's time to completely abandon reliance on Internet thinking. The AI era has just begun, and the road ahead is long. Only by truly understanding the essence of AI and adopting AI thinking can we secure our future in this new revolution.

After all, what defeats you is never the competitor but the era itself.

*All images in this article are sourced from the Internet.