How do you undo AI generated code that nobody can explain?
A Reddit discussion highlights the difficulty of removing AI-generated code, especially when its implications aren't immediately clear. While some AI output is easily modifiable, other parts can become deeply integrated into system schemas, tests, and team understanding before their true cost is realized. The conversation seeks examples of the most challenging AI-generated code to undo.
This discussion uniquely focuses on the practical challenges of undoing AI-generated code, unlike most conversations that center on its creation or immediate utility.
时间与来源
时间显示为 UTC
显示时区：UTC
本地时区尚不可用，暂时显示 UTC。
发布当时偏移：UTC+02026年9月12日 15:22 UTC
收录当时偏移：UTC+02026年9月12日 21:01 UTC
- 发布
- 2026年9月12日 15:22
- 收录
- 2026年9月12日 21:01
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
One thing I think gets missed in all the talk about agent speed is how hard some of the output is to take back out later.
Plenty of it is trivial to change once you know more. Other bits end up wired into the schema, the tests and everyone's mental model of how the thing works, before anybody notices what they are going to cost.
I have run into this with changes that looked completely reasonable at review time. The original call usually wasn't wrong. It was that unpicking it later touched a lot more of the system than the diff ever suggested.
The worst one for us is a caching layer an agent added to our pricing service back in February. It works, it has been in prod for seven months, and nobody on the team can tell you why it invalidates on write instead of on read. Two features built since then depend on the write behaviour.
What would actually have helped is not better code, it is a record of why. We run coderabbit on prs and its summaries are the closest thing we have to one, though they describe what changed rather than what was considered and dropped, which is the half I keep wanting.
The ones that worry me most are the decisions where nobody chose anything. The model picked, the review said fine, and the reason never existed in the first place.
What's the hardest thing you've had to undo that an agent put in?