Alibaba Qianwen Work 2.0 Puts One Agent on a Whole Team of DingTalk Groups

Alibaba released version 2.0 of Qianwen Work (Qianwen Ban Gong), its enterprise office agent product, on November 9 — the largest upgrade since the product launched three months ago. The headline feature is called Collaboration, and it answers a question most AI office tools have avoided: when every employee has their own AI, how does a team actually coordinate?

Unlike products that drop an agent into a group chat to summarize messages, Qianwen Work 2.0 builds a shared collaboration space. Users link authorized DingTalk group chats, documents, AI spreadsheets and AI meeting notes as common context, and can upload additional documents alongside them. A digital employee — default name Xiao Q — can be configured with its own name, avatar, skills and reply rules. Team members assign tasks in the space; the agent combines the linked context to track progress, distill consensus, and create and follow up on to-dos, with multiple sessions running in parallel. When staff change mid-project, the context, division of labor and conclusions stay in the space.

The second new capability, Automatic Tasks, moves the agent from reactive to proactive. Users subscribe to events and configure rules; the first batch of supported triggers includes messages from a specified person, @-mentions in a group, and approval tasks. In the retail scenario Alibaba demonstrated, a regional manager sets a rule: when a store employee @-mentions him about a stockout, the agent reads the inventory sheet, pulls the right context and files a transfer request into the table automatically.

The reference customer is Yintai (In97) department stores. Its coupon operations team — 5 headquarters staff — manages group-buy vouchers for more than 62 stores with over 200 collaborating store, merchant and business staff, and an annual GMV above 200 million yuan. Each voucher requires verifying roughly 50 data fields before it can go on sale, and the work was previously scattered across more than 60 DingTalk groups. In the pilot, the team linked those groups and the coupon AI spreadsheet into one collaboration space; Xiao Q now aggregates cross-group progress, updates the table and reminds responsible owners of pending items.

Version 2.0 also ships human-AI co-creation with unified editing, a resource space, and three-terminal sync across phone, desktop and web — tasks execute in the cloud and continue running after the client is closed. All features are fully rolled out to the Qianwen Work client.

The strategic read: Alibaba is positioning Qianwen Work as an entry layer above IM platforms rather than a replacement for them. Daily communication stays in DingTalk; cross-group projects and long-running routine work move into the collaboration space. Anthropic has been testing the same team-agent concept with Claude in Slack, but Alibaba's version rides DingTalk's enterprise installed base — and industry observers now describe DingTalk and Feishu as the substrate layer beneath AI office products rather than the competitive endpoint.

What it means: the first phase of work AI was one assistant per person; the second is one agent per team, with the context that implies. The hard part is not the model — it is permissions, stability and which product owns the workflow context. Alibaba's answer is to connect to everything rather than replace anything, and the Yintai pilot is the kind of concrete, measurable deployment that will decide whether Team Agent is a category or a slogan.