People from the old era struggle to make things happen in the new one

Chen Hang (alias: Wu Zhao), who once made a high-profile return to DingTalk, has exited again after 437 days. His successor is Chen Yusen, a tech geek born in 1992.

On June 11, following two lengthy resignation posts that sparked public debate, an internal Alibaba announcement confirmed Chen Hang's official departure as CEO, with Chen Yusen taking over. He became the youngest business unit leader within Alibaba's ecosystem.

The two individuals represent distinct era-defining styles: one adheres to the "high pressure yields miracles" management philosophy of the mobile internet era, while the other embodies AI-era thinking driven by Agents.

Chen Yusen succeeding Chen Hang signifies a pivotal shift in Alibaba's AI strategy.

DingTalk serves as one of Alibaba's core ToB entry points, but AI-era product iteration methods differ fundamentally from traditional mobile internet approaches.

What's needed is a player who truly understands AI Native products and can lead DingTalk's evolution from a management tool into a productivity operating system.

According to exclusive information from Guangzhui Intelligence, after Chen Yusen assumes his role as DingTalk CEO, he plans to integrate DingTalk with MuleRun ("Mule Runs Fast"), an AI Agent product developed through his internal Alibaba startup.

This implies DingTalk's future product architecture will undergo complete transformation.

After all, DingTalk under Chen Hang's era was designed for the mobile internet era, constantly adding features. However, AI-era products require simplification—a single entry point satisfying all user needs.

Alibaba's Partnership Committee post titled "Emotional, Ethical, and Growth-Oriented: This is Alibaba's Culture" essentially serves as a declaration of era transition. AI-era innovation relies not on high pressure and mechanical execution but on employees' passion and creativity.

Thus, the divergence between Chen Hang and Chen Yusen reflects not just differing CEO styles but the handover of organizational paradigms between two eras.

Chen Hang's exit marks the end of manpower-intensive tactics and high-pressure sprints within Alibaba's ToB domain. Chen Yusen's appointment signals Alibaba's formal bet on "AI Native organizations" as its core competitive edge for the next decade.

437 Days of "High-Pressure Experiment": Wu Zhao's Return and Exit

Reviewing Chen Hang's decade-long connection with DingTalk reveals a paradox of "success and entrapment through Wu Zhao's approach."

As DingTalk's founder, Chen Hang built this collaborative office product from scratch in 2014.

During the industry's early stages, domestic enterprise office software leaned toward gentle collaboration, while DingTalk differentiated itself with strong control features like "read receipts," "Ding messages," and "smart check-ins," precisely addressing SMEs' management pain points.

Paired with relentless execution and ground teams, DingTalk rapidly penetrated the market, achieving exponential user growth within years and securing its industry leadership.

Chen Hang's management style became deeply ingrained in DingTalk's DNA: goal-oriented, execution-first, and driven by high pressure.

This logic perfectly suited the mobile internet era's "land-grabbing" phase and was key to DingTalk's rapid rise.

But as the competitive landscape shifted, this management and product approach revealed limitations.

Chen Hang's 2025 return attempted to solve current challenges with past experience. He bet heavily on AI while maintaining a "fast-paced, high-pressure" strategy, hoping to replicate early successes.

However, AI products demand deep R&D, scenario refinement, and long-term trial-and-error—leaving no room for short-term gains. The failure of Project "ONE" essentially reflected incompatibility between old growth logic and new track ( track : competitive landscape) rules.

In August 2025, Chen Hang's flagship AI project "ONE" debuted at DingTalk 1.0's launch event, positioned as an AI work information stream entrance ( entrance : gateway). It aimed to extract scattered work information into card streams via large models, enabling "tasks to find people."

The project peaked at ~3 million DAU but was effectively abandoned after November 2025 due to vague positioning, neglected user needs, and internal high-pressure management. It was demoted from the main entrance ( entrance : gateway) to a negative first screen and eventually split into the new product "Wukong."

Accompanying high-pressure management was significant talent attrition at DingTalk. Reports indicate employee numbers dropped from ~1,900 at peak to ~1,600 within six months. One ex-employee told media: "99% of departures were because of Wu Zhao."

In the AI era, talent represents a company's most valuable asset.

After 437 days at the helm, Chen Hang's exit marks not just a personal career transition but DingTalk's definitive departure from the "strong control, heavy execution, speed-only" era.

Why Chen Yusen as Successor?

Examining this 1992-born "young commander's" background reveals a trajectory far different from traditional managers.

Chen Yusen began at Zhejiang University's Zhuken College studying computer science. During his junior year, he discovered the CTF hacker competition, reached the finals solo, then founded Zhejiang University's AAA team (Azure Assassin Alliance), placing 50th globally in their first US iCTF appearance.

In 2014, the 22-year-old graduate founded Changting Technology, securing 6 million RMB in angel funding from ZhenFund within a year. By 2019, the 27-year-old sold Changting to Alibaba Cloud, achieving financial freedom.

Following the acquisition, Chen Yusen joined Alibaba Cloud Intelligence Group, leading the construction of Alibaba Cloud's South America region from scratch. This region later became one of Alibaba Cloud's fastest-growing globally.

In 2025, he relaunched an internal Alibaba Cloud startup to develop the AI Agent product MuleRun ("Mule Runs Fast").

MuleRun's core positioning is AI digital labor—enabling users without technical backgrounds to automate repetitive, time-consuming tasks via AI. It employs a four-layer architecture (Base Agent + Knowledge + Skills + Runtime), assigning each user an independent 7×24 cloud virtual machine sandbox where all tasks execute in isolation.

It features a four-layer memory architecture (short-term session memory, long-term user habit memory, task template memory, team shared memory) that continuously learns users' work styles, becoming more intelligent with use.

By May 2026, MuleRun served enterprises and users across 43 countries including China, Japan, Brazil, and Mexico. Users paying over $200 monthly accounted for 34%, with paying users active 2.6 weekdays weekly and completing 13 end-to-end delivery tasks per person per week.

At Alibaba Cloud's May 2026 summit, Chen Yusen delivered a speech titled "MuleRun: Making Enterprises AI Native Organizations." He asserted: "The performance gap between AI Native and non-AI Native organizations will be at least 10x."

He divided enterprise AI adoption into two phases: Copilot (AI as co-pilot) and AI Native (work reorganized around AI, with humans becoming standard-setters and result-checkers). He noted ~95% of enterprises remain in the Copilot phase, with an 18-month window for AI Native transformation.

He also shared MuleRun's iteration speed: "R&D now works completely differently. Instead of large teams and long cycles releasing versions every 1-2 weeks, we now deploy at least daily. Using our own product internally, we iterate three versions per day."

This speed stems from an organizational logic starkly different from Chen Hang's: not forcing innovation through militarized control but empowering individuals via Agents to unleash creativity.

Chen Yusen's Clear labels ( Clear labels : distinct labels)—young, purely tech-background, unburdened by DingTalk's existing factions—make him Alibaba's ideal manager for the AI era.

Yet opportunities come with daunting challenges. For the young Chen Yusen, this represents a major career test.

DingTalk in the AI Era: The Need for Simplification

Traditional internet-era innovation at enterprises typically flowed top-down, with product evolution driven by market insights. In the AI era, this approach fundamentally changes—more innovation now stems from grassroots employees deeply mining their practical needs.

Chen Hang's "daily package" model—executives proposing requirements morning, teams completing and integrating them into install packages same day—maintained short-term high output but eroded long-term thinking and creativity.

The ONE project described in "Inside DingTalk" exemplifies this: foundational work was constantly sidelined by emergencies, depriving teams of "holistic thinking" space.

In contrast, Chen Yusen's MuleRun team embodies the AI Native organization model: small scale, short decision chains, and heavy reliance on AI collaboration.

Their rapid product iteration doesn't come from overtime but from AI-reconstructed R&D processes—multi-Agent task division, automatic code generation, and automated testing.

If Chen Yusen brings this organizational approach to DingTalk, product iteration rhythms and logic will fundamentally change. Only by evolving into an AI-native organizational architecture from the ground up can the product flywheel truly start spinning.

How should DingTalk evolve in the AI era?

Currently, all AI-native applications pursue minimalist architecture design, enabling users to obtain desired outcomes, generate specific application products, or even write code with a single sentence.

However, DingTalk originates from the mobile internet era, with all underlying architectures designed for enterprises' needs during that period. Despite Chen Hang's AI-driven reshaping attempts and efforts to simplify DingTalk after his return, the changes he brought ultimately proved limited.

As an AI-native organization representative, Chen Yusen plans to deeply integrate MuleRun with DingTalk, potentially reconstructing DingTalk's foundational architecture. The crux of this reconstruction lies in simplification.

Meanwhile, introducing MuleRun's capabilities to DingTalk suggests its AI Agents may evolve from functional call capabilities to contextual understanding and self-optimization.

To illustrate: The current Wukong Agent resembles an intern who checks manuals, knowing how to operate DingTalk's functions. The upgraded Agent would act like a subordinate who's worked with you for three years—understanding your preferred weekly report tone, your company's unwritten approval process rules, and which data your boss cares about most.

Another key change is lowering barriers.

Chen Yusen's core assumption with MuleRun is that as Vibe Coding and Claude Code reduce development barriers, non-technical personnel can encapsulate their work knowledge and processes into Agents.

He emphasizes "simplicity, stability, and usability," arguing that "a sufficiently good prompt holds immense commercial value." Applied to DingTalk, this means enterprise employees and business experts could become Agent producers rather than just consumers. DingTalk would evolve from merely an ISV (Independent Software Vendor) platform to one where ordinary employees create and share their own Agents.

From a business perspective, DingTalk commands 700 million users and serves over 26 million enterprise organizations, dominating China's collaborative office sector. Yet beneath this massive user base lies stagnant growth.

Industry data shows DingTalk's overall paid conversion rate remains below 1%, with revenue heavily dependent on large clients contributing over 60% of revenue. The commercial value of SME users remains untapped, and the monetization loop remains unclosed.

Chen Yusen might introduce new commercialization models to DingTalk.

MuleRun serves 43 countries, with 34% of users paying over $200 monthly. These paying users remain active 2.6 weekdays weekly, completing 13 end-to-end delivery tasks per person per week—representing rare heavy usage data among Agent products.

More importantly, MuleRun's pricing logic is clear: Team version at 420 RMB/seat/month, Enterprise version at 2,500 RMB/seat/year. It charges not by features but by "how much human labor AI replaces"—essentially selling productivity.

However, the challenges Chen Yusen faces after taking over DingTalk are far more complex than with MuleRun.

MuleRun started from scratch without historical baggage, with a small team and rapid decision-making. DingTalk, however, is a super-platform with 800 million users and 26 million enterprise organizations—where any change ripples across the entire ecosystem.

Chen Hang's year-long CLI transformation essentially amounted to "changing engines mid-flight"—maintaining daily use for 800 million users while preventing disruptions to millions of enterprises' business processes.

Chen Yusen must balance two extremes: excessive caution risks DingTalk being overtaken by more agile competitors in the AI era, while aggressive iteration like MuleRun's could trigger enterprise clients' stability concerns.

His security background (Changting Technology) may prove valuable here—he understands better than most the technical philosophy of "maintaining controllability while remaining open." The real test, however, is whether he can combine MuleRun's entrepreneurial speed with DingTalk's enterprise-grade stability.

Looking ahead, DingTalk under Chen Yusen's leadership may transform from a "management software" into a "productivity operating system."

While Chen Hang positioned it as AI infrastructure, Chen Yusen's mission is to make this infrastructure truly "alive"—enabling AI not just to operate DingTalk but to understand enterprises, learn from users, self-evolve, and ultimately empower ordinary people with 24/7 intelligent work partners.

Whether this path can succeed depends on whether Chen Yusen can find that delicate balance between the stability requirements of super platforms and the innovation speed of AI Native.