---
format: "aidr-story-markdown/v1"
id: "569d529b072331872f7c436a5ed2485c59be64bcfa69a6735209afe47a56ce3e"
canonical_url: "https://aidr.today/569d529b?lang=en"
title: "AI coding agents still need human cleanup on nearly three out of four tasks in a real-world test"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-08T09:50:59.000Z"
category: "Agents"
topics: ["openai","agent"]
source_urls: ["https://the-decoder.com/ai-coding-agents-still-need-human-cleanup-on-nearly-three-out-of-four-tasks-in-a-real-world-test/"]
summary: "In a real-world test using Alibaba's Co-Design Bench, AI coding agents required human cleanup on nearly three out of four tasks. The top performer, GPT-6 Astra, fully solved only 26.4 percent of tasks, despite scoring above 99 percent on older benchmarks."
---

# AI coding agents still need human cleanup on nearly three out of four tasks in a real\-world test

> [Open the canonical story](<https://aidr.today/569d529b?lang=en>)

**Published:** 2026-11-08T09:50:59.000Z
**Category:** Agents
**Topics:** openai, agent

## Summary

In a real\-world test using Alibaba's Co\-Design Bench, AI coding agents required human cleanup on nearly three out of four tasks\. The top performer, GPT\-6 Astra, fully solved only 26\.4 percent of tasks, despite scoring above 99 percent on older benchmarks\.

## Sources

- [Story source](<https://the-decoder.com/ai-coding-agents-still-need-human-cleanup-on-nearly-three-out-of-four-tasks-in-a-real-world-test/>)