Alibaba launches Qianwen Office 2.0 with a multi-user Team Agent

On November 9, 2026, Alibaba released version 2.0 of Qianwen Office (千问办公), its office assistant built on AI agents. According to a text supplied by Qianwen and republished by the Chinese outlet QbitAI, it is the biggest update since the product launched three months ago, and every feature is already available to all users of the Qianwen Office client. The stated bet is to move the agent from serving one individual to serving a team, even a whole organisation, in what Alibaba calls a "Team Agent".

An AI agent is software that does more than answer a question: it chains actions, such as reading documents, filling in a spreadsheet or creating a task, to reach a goal. Until now most office agents have worked one-to-one with a single user. In its official blog post of the same day, titled "Qianwen Office 2.0: proactivity and coordination, towards the Team Agent era", Qianwen observes that the way teams collaborate has not changed and that the next generation of agents has to enter the place where collective work actually happens.

What does Qianwen Office 2.0 include?

• "Collaboration": a shared space where people and "digital employees" work together, presented by Alibaba as the first feature in the industry to handle several humans and agents at once.
• "Automatic tasks": the agent starts on its own when a condition is met, instead of waiting for an instruction.
• "Human-AI co-creation": editing documents, tables and slides directly from the conversation.
• "Material space": a library of files generated by AI or uploaded by the user.
• "Three-device link": phone, computer and web synchronised in real time.

This information comes from QbitAI, whose article states that it was written by Qianwen and republished with permission, and from the outlet CNMO, which reported the same launch on November 9. The "industry first" framing is therefore Alibaba's own, and we could not confirm it with an independent source.

How does the collaboration space work?

In the collaboration space, users attach several DingTalk group chats, documents, "AI Table" spreadsheets and "AI Listening" meeting transcripts to a single shared context. According to QbitAI, the digital employee can follow several conversations in parallel, manage and assign tasks, and then use the context to summarise progress, draw out points of agreement and create follow-up items. According to CNMO, this employee is called "Little Q": a company can customise its name, avatar, skills and reply rules, and set a usage quota in the admin console.

Alibaba cites the coupon operations team at Yintai department stores. Previously, five operators had to collect coupon requests from more than 60 store groups one by one, check them and copy them into a spreadsheet. Now those groups and one AI table are linked to the same space: communication habits do not change, and the agent aggregates information across groups and updates the table. The workload is said to have fallen "significantly"; no figure is given, and this is an example supplied by the vendor.

Automatic tasks, co-creation and three devices

"Automatic tasks" move the agent from reactive to proactive. The user subscribes to events and sets rules; when the condition is met, the agent goes to work. The first supported triggers are messages from a named person, messages that mention the user (@) in a given group, and approval tasks.

QbitAI gives an example from store management: a regional manager sets the rule "if someone @-mentions me about a stock shortage in the group, read the inventory table and start a transfer request". As soon as a clerk writes, the agent reads the stock, fills the transfer request into the table, and the manager no longer has to watch the group. Tasks run in the cloud and keep going even when the computer is switched off.

"Human-AI co-creation" lets users select a paragraph, a cell, a slide or an application area, leave several comments and submit them at once; the AI then edits the original document instead of regenerating it. For online documents, AI tables and Yida applications, several people and several agents can edit at the same time. The new "material space" gathers images, documents, tables, presentations, web pages, applications and local files, with search and sharing, and lets users start a new conversation from a selection. Finally, with the user's authorisation, a phone can remotely control the computer.

Opening to enterprises and known limits

On the enterprise side, CNMO describes three routes: third-party business applications connected to the platform, "Qianwen Office Inside" to embed the agent's abilities in an existing system, and "Managed Agent", which lets a company configure rules, tools and skills into a dedicated agent hosted in the cloud and linked by API to customer-relationship, recruiting or customer-service systems. Data from personal and corporate accounts is kept separate. QbitAI recalls that the product was already connected to DingTalk, Feishu and WeCom, and presents it as the first enterprise-grade agent.

Several points remain unclear. The reports we read give no price, no user count, no underlying AI model and no availability outside China. CNMO stresses that whether the product really improves business efficiency still has to be validated in more scenarios. And the "industry first" label is a commercial claim by Alibaba.

What this changes for businesses

For a company outside China, the lesson lies less in the tool, designed around DingTalk and the Chinese market, than in the direction it points to: the agent is no longer a personal assistant but a participant in team channels, with permissions, quotas and automatic triggers. That raises concrete questions of AI agent governance: who can hand it a task, what it can read in the groups, who approves its actions. Before copying the model, test it on a repetitive, well-bounded process, such as tracking requests across several teams, and measure the gain with your own numbers rather than a vendor's.