Alibaba’s QwenWork 2.0 Gives DingTalk Teams a Shared Agent That Acts on Events

Alibaba released QwenWork 2.0 on 9 November 2026, giving teams a shared AI agent and a way to start tasks automatically when certain workplace events occur. The update, reported by IT Home, changes how people can use QwenWork: an agent can work from a team’s shared discussion and files, and a configured task can start without someone opening a chat and prompting it again. This is a QwenWork product release, not a new Qwen model.

One agent for a shared workspace

According to Alibaba’s official release notes, each collaboration space gets its own agent. Teammates can discuss a project, assign work, and have the agent use the space’s materials and agreed decisions to move tasks forward. The work and its outputs stay in the workspace, so the next teammate can pick up where the last one left off.

Alibaba also says users can open DingTalk documents, spreadsheets, and Yida apps from a task conversation, make edits themselves, and ask the agent to continue from the revised version. The release notes say online documents and spreadsheets retain their existing permissions. That is Alibaba’s documented behavior, not an independently tested guarantee. Our coverage of agent permission boundaries explains why authorizing one action does not settle what an agent may do next.

DingTalk events can start a task

QwenWork previously offered scheduled tasks. Version 2.0 adds event triggers: instead of waiting for a set time, a task can start when a configured condition matches a message, an approval event, or a to-do. Alibaba’s example is a message from a manager that prompts QwenWork to create a DingTalk to-do. The company documents support for running triggered tasks on a local or cloud computer.

That matters because a team can connect an existing workflow to an agent without asking someone to relay the work manually. But an approval event that starts a task is not necessarily authorization for every action the agent takes afterward. If access changes between the trigger and the write, the team needs to know which permissions apply at execution time. The same distinction between starting and controlling delegated work came up in our coverage of OpenAI’s expansion of dots to enterprise workspaces.

The test: replay an event, then change the approval

Before connecting the agent to a production spreadsheet, set up a test workspace with dummy records. Configure a test approval to add a row with a unique request ID. Replay the same event, if the environment permits, and check whether the agent writes a second row.

Then use a separate request to test timing: withdraw its approval, if supported, before the write is attempted. Check whether the workflow stops, asks for fresh authorization, or still writes. Record the trigger, approval state, and resulting rows so the outcome can be audited. AgentsCalling has not run these tests.

Availability

The shared-agent and event-trigger features are documented in the Chinese desktop and web release notes dated 9 November. AgentsCalling has not verified the same rollout for the international version.

Reporting note: Prepared with AI-assisted research and drafting. Product capabilities are attributed to Alibaba; AgentsCalling has not independently tested QwenWork 2.0.