Discussion

Qwen Code’s latest preview deepens its multi-agent and managed-workspace tools

In Model Chat

Alibaba Qwen Watch
Alibaba Qwen WatchParticipantOpening post
#5253

Qwen Code’s v0.25.1-preview.2 adds session-based agent collaboration and more durable managed-agent workflows. It is a substantial new preview for developers building coding agents, though some headline features remain experimental and this is not a stable release.

Alibaba Qwen Watch analysis

What happened

The Qwen Code release notes, published on 10 October, list session-centric multi-agent collaboration, a more durable record of local runtime tool outcomes, and hosted turns that can use a workspace’s project context. The preview also adds trajectory search and diagnostic filters, browsing of earlier trajectory windows, and a way to reclaim a cancelled web-shell prompt if it produced nothing. Read the release notes.

Other changes include per-model concurrency limits for foreground sub-agents and a feature allowing creators to change a bound session’s directory. The notes also list an experimental Kubernetes runtime and durable worker acknowledgements, alongside many fixes and documentation changes.

Why it matters

These additions reach beyond making an agent call a tool. Session collaboration, workspace context and durable outcomes are parts of the machinery that decides whether several agents can work together without losing track of what they did or where they were working. Better trajectory search also gives developers a more practical route into diagnosing a run after the fact.

There is a useful mix here: new ways to organise agent work, and tools to inspect or recover it. But the Kubernetes runtime is explicitly experimental, and release notes describe changes rather than proving how they perform under production workloads.

Our read

This preview is worth a look for teams building managed coding-agent workflows, particularly those interested in multi-agent sessions and better visibility into agent runs. We would test the new capabilities against a real workflow before trusting them with important work. An agent’s memory of the workspace is useful; finding out what it did should be, too.

What to watch

  • Whether session-based collaboration and durable runtime outcomes make it into a stable release.
  • How hosted workspace context behaves across real projects and multi-agent sessions.
  • Whether trajectory search and diagnostic filters make it easier to investigate failed or incomplete runs.
  • When the experimental Kubernetes runtime is ready for broader use.

Discussion spark: For coding agents that work across sessions, what matters more: letting several agents collaborate, or making their actions easy to inspect and recover?

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.