Qwen Office 2.0: Set Up Team Agents Without Confusing Access and Automation

Qwen Office 2.0, also called QwenWork or 千问办公, adds shared team-agent spaces, event-triggered tasks and remote computer control. The November 9 release connects more of the workflow, but linking a group, inviting a colleague and authorizing a computer remain separate actions.

That is the first thing to understand before connecting a working team. A shared interface can make the product feel like one continuous environment, while the files, people and execution permissions underneath it still belong to different systems.

What changed in Qwen Office 2.0

The official desktop release notes identify version 2.0.0 and add collaboration spaces with a dedicated digital employee, in-task document editing, a materials space and mobile remote control. The former scheduled-task area becomes automatic tasks, adding event triggers for messages, approvals and to-dos alongside time-based schedules.

The practical change is that a project can keep its discussion, linked information and agent work together. A launch team could use one space for its brief and project discussions, rather than asking each person to rebuild the same context in a private chat. That still requires an explicit decision about what belongs in the space and who should see it.

Set up the collaboration space and its membership separately

The collaboration guide says spaces operate within an enterprise organization. Switch to the intended organization, open Collaboration, create a space, name its digital employee and choose colleagues from that organization. Availability depends on version, organization and rollout. For web access, use a desktop browser; the guide excludes mobile web and DingTalk's embedded pages.

Then add the relevant data sources. A group administrator can confirm reading access for a DingTalk group. An ordinary group member must submit a reason and wait for administrator approval. Check the data-source status rather than assuming a selected group was successfully connected.

Connecting a DingTalk group does not invite all its members into the space. Membership and source association are different controls. Group-content authorization covers chat history and group files for use by space members and the digital employee; online documents and AI tables need their own source entries.

For a launch project, that suggests a useful review before the first agent request: who is in the space, which discussions are linked, and whether those discussions contain information intended for a smaller audience. Adding more context should be a deliberate choice, rather than the default response to a weak answer.

Choose a trigger with a narrow job

In the automatic-task documentation, time-based tasks follow a schedule, while triggered tasks react to selected DingTalk events and filters. A trigger can run in the cloud or on this computer. Local execution requires the computer to stay awake with the client running.

The free personal edition allows ten enabled automatic tasks in total. Triggered tasks have a 200-run hourly limit, with additional runs in that hour not executed. Runs consume Credits. A scheduled task can be run immediately for checking, but triggered tasks currently do not support that manual-run control.

Start with a narrow read-and-draft job. For example, the following is a suggested instruction for an internal launch-review workflow, not a tested configuration:

“When I am mentioned in the launch-review group about a missing asset, inspect the linked launch checklist. Prepare a note containing the missing item, the source message and any owner or date explicitly stated. Leave unknown owners and dates blank. Do not change the checklist, create commitments or send a reply to the group. Save the result for my review.”

This separates discovery from action. Test the underlying task in a normal conversation first, then check a controlled qualifying event after configuration. Include an irrelevant message in the pilot as well: a workflow that responds correctly to everything it sees may still be responding far too often.

Phone control does not move a job to the cloud

The remote-control guide is explicit: a remotely controlled task continues running on the chosen computer. Remote control neither converts it into a cloud task nor synchronizes every local file to the phone.

To use it, enable remote control on the target computer and sign in from the phone or web with the same account and organization. Keep the desktop client online and the machine awake. For a new task, check the execution-location selector and choose the intended computer. An unavailable switch can reflect an organization policy that needs administrator action.

Task-level questions and confirmations can be answered remotely, but operating-system permission windows must be handled on the original computer. That is a useful troubleshooting distinction when a task seems stuck despite an active phone connection.

A task's execution location matters more than the device showing its progress. If the required spreadsheet exists only on a laptop, opening the conversation on a phone does not make the spreadsheet available to a cloud worker. Confirm the selected machine and authorized working directory before changing prompts or retrying the task.

Keep document permissions visible

The materials-space guide describes a place to find and reuse generated files, uploaded files and supported online materials. It does not automatically import every file on a computer. Access to a source conversation is also checked separately.

Treat a material's presence in that list as a navigation aid, not proof that every collaborator can open or edit it. The 2.0 release notes say online documents, tables and applications retain their own permissions. If a colleague cannot use a linked document, inspect the document's access rules before adding broader workspace permissions.

The strongest use of this release is a repeatable team process with a clear owner, a bounded set of sources and a review point before consequential changes. Connecting an agent to more conversations is easy. Making its work dependable requires deciding which conversations it should use, where it should execute and when a person still needs to make the call.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help teams explain complex enterprise AI workflows clearly, so customers understand both the useful capabilities and the decisions required to adopt them.

Book a 30-minute call.