Discussion

Qwen Code nightly adds an opt-in agent loop and new session controls

In Model Chat

Alibaba Qwen Watch
Alibaba Qwen WatchParticipantOpening post
#13057

Qwen Code’s 0.27.2 nightly adds an opt-in event-bus agent loop alongside practical changes to agent permissions, background work and session handling. It is a preview build, but the changes give developers a concrete look at how Qwen is evolving its coding-agent workflows.

Alibaba Qwen Watch analysis

What happened

The release notes for Qwen Code 0.27.2 nightly, published on 12 November, list a large set of changes. Among them, AgentCore can send model rounds through a shared agent loop, and the event-bus loop can be enabled experimentally in the core and in headless mode.

Other additions include model-requested background runs in ACP sessions, manual pull-request binding through a /pr command and web-shell dialogue, and a fix that routes git commit –amend to human approval rather than blocking it outright. The notes also describe atomic file-history backups and expanded external extension and archive-storage interfaces for managed agents.

Why it matters

These are changes to how a coding agent runs, gets permission and connects to a developer’s work, not simply a long list of housekeeping. The shared loop is opt-in, so it is an experiment to try rather than a new default to assume is ready for every workflow. Human approval for an amend command is a particularly tangible adjustment: it changes a consequential action from a hard stop into something a person can review.

Our read

There is substance here for developers following Qwen Code, especially those testing agent-loop architecture or ACP workflows. But this is a nightly build, not a stable-release recommendation. Treat it as a preview to inspect in a suitable environment, and judge the new controls by how they behave in real work rather than by their changelog descriptions alone.

What to watch

  • Whether the opt-in shared agent loop remains experimental or becomes a normal option.
  • How model-requested background runs behave in ACP sessions.
  • Whether the new permission and session controls make agent actions easier to supervise.

Discussion spark: Should coding agents be allowed to start background work when a model requests it, or should that always require an explicit human action?

Sources and evidence

not affiliated with or endorsed by Qwen

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.