"Monkey King" Sweeps Away "Shrimp Soldiers" - Alibaba Restructures DingTalk: Wu Yongming Launches the First Move in the B2B AI Battlefield

OpenClaw (an open-source AI agent framework, commonly known as Lobster) is sweeping the market, and “raising lobsters” has become a new industry trend. However, Alibaba has voiced a different perspective.

“I see articles about lobsters flooding the internet every day. Knowledge bloggers are constantly promoting lobsters, and I feel it’s irresponsible,” said DingTalk CEO Chen Hang (alias Wu Zhao) at the AI DingTalk 2.0 annual product launch on March 17. He warned that uncontrolled super intelligent agents pose risks of backlash.

Following this, Chen Hang unveiled Alibaba’s breakthrough plan, releasing the world’s first enterprise-grade AI native work platform, “Wukong.”

Interestingly, on the big screen at the launch event, a cartoon “Wukong” holding a staff stands in the center of a group of shrimp soldiers, filled with metaphorical meaning. Alibaba claims Wukong is a 24-hour “lobster army” directly embedded into DingTalk, which serves over 20 million enterprise organizations.

“Today, we are breaking DingTalk apart and rebuilding it with AI to forge ‘Wukong’,” Chen Hang said. In the past, people used DingTalk to work; in the future, AI will use DingTalk to work. “Unlike all other lobster Agents on the market, Wukong is naturally integrated into enterprise organizations and can be safely used in real business environments.”

On March 16, Alibaba announced the establishment of the Alibaba Token Hub (ATH) business group, directly led by Alibaba CEO Wu Yongming. The Wukong division made its first public appearance, signaling Alibaba’s focus on the B2B AI application market.

This is not only a restructuring of DingTalk but also a key turning point in Alibaba’s AI strategy. Behind this well-prepared launch, it reflects Alibaba’s latest layout for AI commercialization.

(Company provided images)

Safety and Control Are the Final Battlefield: Alibaba Wields the “Golden Cudgel”

At the start of the event, Chen Hang detailed security issues related to OpenClaw. He stated that releasing all “lobsters” would threaten the entire ecosystem, as installing them on personal or corporate computers could create backdoors, leading to Trojan viruses and other malware.

Industry concerns about OpenClaw’s security risks have existed for some time.

On March 8, the Ministry of Industry and Information Technology’s cybersecurity threat and vulnerability sharing platform reported that some instances of OpenClaw’s open-source AI agents, when misconfigured or by default, pose high security risks, easily leading to cyberattacks and data leaks.

Officially, the key difference between Wukong and other AI Agents is that while others focus on enabling AI to perform tasks, Wukong emphasizes safe, controllable, and accountable AI work within enterprises.

According to reports, Wukong’s dual-layer rule system sets absolute boundaries for AI behavior, with fundamental security rules taking top priority—no instructions can override them. Custom enterprise rules allow administrators to flexibly configure according to industry specifics.

Built on DingTalk’s 11-year enterprise permission system, administrators can precisely control who can use Wukong, which AI skills they can access, and what data they can see. All operations are tied to real enterprise identities, and insufficient permissions mean no data access.

In group chat scenarios, Wukong’s permissions are the intersection of “user permissions” and “questioner permissions.” Even if Wukong can theoretically access certain data, it won’t return data if the questioner lacks permission.

Additionally, its full-chain audit logs record every input, skill invocation, and output—who did what, when, with what identity, and on what data—making it clear and ensuring enterprise data security and AI behavior traceability.

iMedia Research CEO and chief analyst Zhang Yi told the “Daily Economic News” that OpenClaw indeed has security issues like overreach and data leaks, as it follows a development path prioritizing features over security. “Objectively, it lacks essential enterprise permission systems, data isolation, and operation auditing, making it more suitable for individual tech enthusiasts.”

While security and control are valuable, can this become Wukong’s core competitive advantage? Zhang Yi believes achieving secure and controllable technology is challenging. “It’s not just about stacking security features; it requires fundamental security design at the architecture level, involving enterprise organizational structure, deeply integrated permission systems, data isolation mechanisms, traceable operation controls, multi-AI collaboration security governance, and more—all crucial.”

(Company provided images)

DingTalk Cultivates “Wukong”: Can AI Truly Take Over Workflows?

Despite being a DingTalk launch, why did Wukong become the star?

Sources say Wukong was developed by the DingTalk team as Alibaba’s enterprise-grade AI native work platform.

On March 17, after updating the DingTalk app, a revamped, significantly changed interface appeared, with a new AI entry point on the bottom left.

Regarding the relationship between DingTalk and Wukong, DingTalk’s new “AI Search” feature explained that they are deeply integrated and co-evolving. Wukong is not a standalone competitor but an enterprise AI native work platform built on DingTalk, reconstructed and upgraded from the ground up.

In simple terms, DingTalk provides the foundation and framework, while Wukong is the intelligent neural network and execution engine embedded within.

As Chen Hang said, DingTalk rewrote its underlying code for Wukong, making all capabilities accessible and operable via CLI (Command-line Interface).

“AI Search” states this transformation enables true “communication as execution.” For example, in a DingTalk group, a command like “Generate last week’s sales report and share with management” can be automatically processed by Wukong, pulling approval workflows, attendance, CRM data, generating reports, and pushing them—all without manual interface interaction.

Wukong is also a standalone app but will be directly integrated into the latest DingTalk AI 2.0 version, available to over 20 million enterprise users and 800 million users overall, ready to use out of the box without additional installation.

Furthermore, Alibaba’s B2B capabilities—such as Taobao, Tmall, Alipay, and Alibaba Cloud Skills—will gradually connect to Wukong, making it Alibaba’s unified AI capability platform for enterprise scenarios.

By embedding Wukong into DingTalk’s large user base, Alibaba aims to rapidly deliver its enterprise AI capabilities to countless businesses.

On the launch day, Wukong also released the TOP10 industry solutions for OPT (One Person Team), the world’s first AI Skill product that transitions from a technical concept to an industry-ready, plug-and-play solution. It covers scenarios like e-commerce, cross-border e-commerce, knowledge bloggers, developers, and retail stores, with one-click activation.

According to reports, from late 2025 to early 2026, the global AI industry will shift from an “Agent arms race” to “Skills ecosystem building.” Industry consensus is increasingly clear: AI’s core competitiveness no longer depends solely on model size but on Skills—standardized, reusable, and evolvable capability modules that can be embedded into business processes and scaled effectively.

Cui Lili, deputy dean and professor at Shanghai University of Finance and Economics’ Digital Economy Research Institute, told the “Daily Economic News” that Wukong is more like a platform aggregating B-end Skills, still exploring micro-team work scenarios, and remains in the experimental stage.

Notably, many of the requests made after tagging @Wukong in the updated DingTalk app were rejected due to permission isolation and security sandbox restrictions. For example, Wukong does not automatically scan or archive user history, meaning users need time to learn how to use the app effectively.

(Company provided images)

After the Lobster Frenzy, the B-End AI Competition Begins

All signs indicate that Alibaba’s AI strategy has now set its sights on the next development phase.

On March 16, Alibaba officially established the Alibaba Token Hub (ATH) business group, a new organization focused on “creating tokens, delivering tokens, and applying tokens,” directly led by Alibaba CEO Wu Yongming.

The Alibaba Token Hub includes the Tongyi Laboratory, MaaS (Model as a Service) division, Qianwen division, Wukong division, and AI Innovation division, covering everything from foundational model R&D and model service platforms to AI applications for individuals and enterprises. The Wukong division makes its first public appearance. The formation of an independent division indicates that the B2B AI application market has become a core part of Alibaba’s AI strategy.

What does the future hold for this emerging market?

Zhang Yi believes that enterprise B2B AI applications will become a key focus for major tech companies.

On one hand, the commercial cycle in the B2B market is becoming clearer—companies are willing to pay for cost reduction, efficiency, and security compliance, with more stable monetization paths than consumer markets. On the other hand, B2B scenarios have clear boundaries, more standardized workflows, and higher controllability, facilitating safe AI deployment. Coupled with strict requirements on data sovereignty, permissions, and compliance, this creates higher industry barriers, making B2B AI a battleground for giants.

Zhang Yi also noted that recent widespread discussions and experiences with OpenClaw have made the market more aware of the real demands from enterprises and users.

He emphasizes that future industry competition will mainly focus on: 1) integration of model foundations with MaaS platforms; 2) service capabilities of C-end AI portals and personal assistants; 3) seamless integration of AI Agents and workflows in B-end scenarios.

He stresses that success depends not just on the strength of the models but also on user acceptance and usage rates, which require security and compliance guarantees, as well as practical deployment in specific scenarios. Business monetization and ecosystem integration are also crucial factors.

Currently, Alibaba’s AI approach is becoming clearer: leveraging the “Qianwen” app for user penetration and model iteration on the consumer side, exploring commercialization and building competitive barriers with “Wukong” on the enterprise side, and connecting both through the ATH business group.

From the “lobster everywhere” phase to the emergence of Wukong, the landscape of B2B AI applications is at a critical turning point. In consumer markets, AI can be an all-powerful “magician”; in enterprise scenarios, it must be a safe, reliable, and clearly responsible “professional manager.”

Perhaps only when AI truly evolves from “being able to work” to “being able to excel in enterprise tasks” can the commercialization cycle be fully realized, and the industry competition truly begin.

(Article source: Daily Economic News)