Alibaba's Qwen Codes Autonomously for 10 Days: Files Issues, Writes Code, Merges PRs

Due-Cup9574 · reddit · 2026-08-04

A developer shared results from an autonomous coding experiment using Alibaba's Qwen model. Given an empty repository and no human intervention, Qwen demonstrated strong long-horizon autonomy: it filed its own issues, wrote code, ran tests, fixed failures, and merged the code over a period of more than 10 days.

The author notes that while this proves the agent can grind for days and self-correct—making it a fundamentally different tool than a single-shot chat prompt—it still required a few rounds of human feedback to get things exactly right. The full trace of the run is public.

Original post →

More from coding & agent

coding & agent channel →