Alibaba's answer to the enterprise AI agent race has a name: Wukong. Announced on March 17, the platform lets businesses orchestrate multiple AI agents from a single interface to handle tasks like document approvals, meeting transcription, and cross-system research.
The tool ships with pre-built solutions spanning ten verticals including e-commerce, manufacturing, legal, finance, and software development. It runs as a standalone desktop app or plugs directly into DingTalk, Alibaba's workplace messaging platform that already has 26 million corporate users.
What Sets It Apart
Wukong is built on Alibaba's Qwen 3.5 model, which uses a 397-billion-parameter MoE (Mixture-of-Experts) architecture - meaning it has a massive total brain but only activates 17 billion parameters per task, cutting inference costs by about 60% compared to the previous version. The model supports over 200 languages and a 256K token context window (roughly 600 pages of text), extendable to 1 million tokens.
The more notable technical claim is that Alibaba rebuilt DingTalk's entire backend so AI agents operate natively through command-line-style interfaces rather than simulating mouse clicks on screen. That is a fundamentally different approach from tools that overlay AI on top of existing software, and should be more reliable.
Security gets a four-layer architecture: unified identity authentication, dedicated sandbox isolation for each agent, network proxy management, and full audit logs. Alibaba is positioning this as a direct contrast to documented security issues with open-source agent frameworks.
Availability and Competition
Wukong is currently in invitation-only beta, with no public pricing announced. When it launches broadly, the DingTalk integration gives it an instant distribution channel across millions of businesses - though primarily in China and Southeast Asia. Alibaba says integrations with Slack, Microsoft Teams, and WeChat are on the roadmap.
The competitive landscape is getting crowded. Microsoft Copilot is embedded across Office 365 at around $30 per user per month. Google launched Agentspace for Workspace in late February. Tencent has its own WorkBuddy. Alibaba's advantage is the Qwen model being open-source under Apache 2.0, which could appeal to enterprises that want to self-host or customize their agent infrastructure.
China's AI agent market is projected to grow from under $1 billion in 2024 to over $30 billion by 2028, and Alibaba CEO Eddie Wu has committed $53 billion toward AI infrastructure. The bet is big. Whether Wukong gains traction outside Alibaba's existing ecosystem - particularly in Western markets where DingTalk has minimal presence - is the real test.