Alibaba Launches Enterprise Agent Collaboration Platform Wanyou Wujie

Alibaba launched an agent product called "Wanyou Wujie" on September 2. According to Alibaba Cloud's definition, it is an enterprise-level "human-agent collaboration platform" that decomposes complex tasks across multiple agents, with humans and agents jointly participating in projects. The process and results are ultimately stored in the enterprise account. Previously, one agent worked on behalf of a person; now, a group of agents works together.

Earlier, Qoder Work, MuleRun, and Wukong from the DingTalk ecosystem were reincorporated into Qianwen Office, handling computer operations, cloud execution, and enterprise collaboration respectively. These three different routes were ultimately placed under the Qianwen brand. This appeared to be a consolidation, but a month later, Wanyou Wujie emerged. From a product definition perspective, each has its own focus: Qianwen Office leans toward office scenarios, Wanyou Wujie emphasizes human-multi-agent collaboration, Bailian targets developers, and DingTalk holds enterprise organizational relationships. Each has its place and reason to continue existing.

Alibaba is no stranger to the horse-racing mechanism. When the market is unclear, letting different teams run forward simultaneously and ultimately leaving the answer to users and business results is a more realistic approach for a large company than betting on the future from the start. It can hide in a chat box, operate a computer, enter office software, or become a new entry point above software. Alibaba has enough cards: Qianwen has models, Alibaba Cloud has computing power and enterprise customers, DingTalk has organizational relationships, and Bailian connects developers. In the agent era, every business has the opportunity to move forward.

So Qoder Work could be done, Wukong could be done, and MuleRun could also be done. Later, they merged into Qianwen Office, and now there is Wanyou Wujie. This does not necessarily mean the previous integration failed. It is just that when every team in a company is capable of building agents, "who does it" is never the hardest question; the hard part is who does not. The real trouble with horse racing in large companies often occurs here: it is not that there are no good products, but that behind every horse, there is a sufficiently valid rationale. Alibaba now has many such rationales, and it still lacks an answer that makes the other rationales less important.

Wanyou Wujie paints an attractive future: in a company, there will not only be employees but also a large number of agents—sales agents, customer service agents, procurement agents, and finance agents. For enterprises, this is productivity; for cloud vendors, it is computing volume. Agents will not sit quietly like traditional software; they need to call models, search information, read databases, use APIs, and then hand tasks to another agent. The busier the agents, the busier the cloud. This is a business that aligns well with Alibaba Cloud's interests. Of course, as long as enterprises earn more than they spend, there is no problem.

The issue is that enterprises will not ultimately pay for the word "agent." Bosses will look at other numbers: how many fewer people were hired, how much faster processes became, how much costs decreased, and how much revenue increased. If an agent costs RMB 1,000 and saves RMB 5,000, the decision is easy; if it merely enables an employee to produce five PPTs a day instead of three, the matter becomes less attractive. Over the past two years, the AI industry has been best at proving "it can do it," but enterprise procurement will eventually ask another question: why spend money to have it do it? When agents truly enter enterprise budgets, this question will matter more than model parameters, especially for a company that simultaneously sells models, cloud, databases, search, and agent platforms.

In the past few years, if Alibaba needed an enterprise AI entry point, DingTalk was almost the ready-made answer: contacts, approvals, workflows, permissions, and organizational relationships were all there. AI entering DingTalk, and then DingTalk becoming the enterprise agent entry point, was a smooth path. But now, things have changed slightly: Wukong was merged into Qianwen Office, and Qianwen Office then entered DingTalk; Alibaba Cloud has its own agent platform, Bailian handles developers, and Qianwen is increasingly becoming Alibaba AI's unified brand. Previously, AI entered DingTalk; now DingTalk is also entering AI.

From Qianwen to Alibaba Cloud, from Bailian to DingTalk, and then to Qianwen Office and Wanyou Wujie, Alibaba is betting more and more resources on agents. Each piece can find its place, and that is also the problem. In the early market, having more horses is not a bad thing; no one knows where the finish line is, and running multiple routes may even be safer. But the truly difficult part of horse racing is never getting the horses to run, but deciding, after the race has progressed to a certain stage, who no longer needs to run. Qoder Work, Wukong, and MuleRun have already experienced such a choice once, and now Wanyou Wujie has stepped onto the track. It is still too early, but Alibaba's stable of horses is already growing larger.