如何处理经理在我们的代码库/设计等中添加人工智能“废话”。
3 分•作者: LandOfMightDev•大约 9 小时前
我大学毕业后就加入了这家公司，在此之前有大约 5-6 年的业余编程项目经验。我入职一年了。我基本上是我团队里表现最好的人，承担更多的任务和更复杂的项目，并且能以极快的速度完成工作，几乎没有缺陷等等。相比之下，团队里的大多数人需要花费整个冲刺周期（sprint）的 2/3 来部署他们的任务，然后在冲刺结束前修复缺陷（通常会产生新的缺陷，直到我介入他们的工作领域，或者测试人员在几个月后的回归测试中偶然发现）。
最近公司一直在试图推高股价，所以他们要求每个人都全力投入到人工智能（AI）开发工作中。好吧，没问题，AI 有好有坏，只要有人检查代码，就没什么大不了的。然而，我的团队却把这理解为完全不用动脑子了。基本上，我看到的所有代码都是过度设计的垃圾，几乎总是会导致缺陷，或者引入大量的技术债务。我认为他们可能只是在制造负面工作，因为在下一个冲刺周期，通常都是我被指派去修复这些问题（因为我总是能很快完成我的工作）。
到目前为止，我一直认为这种糟糕的做法仅限于我的同事，而管理层会认真对待，产出高质量的工作，并仔细审查我们的 PR（Pull Request），在利用 AI 的同时进行彻底检查。我错了。我注意到的第一件事是，技术设计文档有 5 页长，但由于我不太熟悉代码库的这部分，我有点想当然地认为它是准确的。这是一个大错误，大约一天后，我发现文档的一半只是 AI 生成的垃圾，里面随意散布着一些无意义的内容。我曾多次询问同事他们的代码是如何工作的（因为 AI 添加了过度设计的垃圾），而他们却完全无法解释。
总之，我弄清楚了那些问题，并将 PR 发布给我的老板进行审查。发布后，他让我让 AI 审查工作并添加更改。我照做了，其中有一些明显的 AI 幻觉，但也有一些有效的建议，比如更新文档、添加一些错误处理等，我进行了修改。我将修改后的版本发回给他，他要求进行一次通话，我们进行了讨论。原来他的 AI 还有其他建议（给他的），我们一起分析了哪些是垃圾，哪些是有效的（由糟糕的设计文档引起的问题），我在通话中进行了更新并推送了。
一个小时后，他回来告诉我：“我让 AI 给出了一些额外的建议，你能检查一下分支并对比一下你的差异吗？”我查看了那个分支，果然，里面包含了我们在通话中讨论过并确定为垃圾的所有内容。它还随意更改了结构，使其更加复杂，难以解耦和维护。据我所知，要么没有人理解代码库，要么无法理解 AI 生成的代码，要么就是太懒惰而不愿思考。AI 的很多行为都是基于代码库中以前的代码，而这些代码可能已经有 10 年了，写得很糟糕，但 AI 仍然会建议遵循这种思路等等。
我感到非常沮丧，我是团队里级别最低的开发者，但却做着最多的工作，质量也最高（团队中的业务分析师已经明确指出了这一点），但我仍然是薪水最低的，并且不得不处理最多的烂摊子。
查看原文
Joined the company out of college with like 5-6 years coding hobby projects etc. & started a year ago. I'm basically the highest performer in my team, get more stories & points/higher complexity tasks & have my work done way quicker with virtually no defects etc.etc. In comparison, most of the team spends like 2/3 the sprint getting their stories deployed & then up until the end of the sprint fixing the defects (usually producing defects that we don't catch until either I have to work in their area or the tester's find it randomly months later in regression)<p>So recently the company's been trying to pump the stock so they asked everyone to go all in on AI dev work. Ok fine, there are good & bad uses for AI & if ur atleast checking the code, it's np. Well, the team took that as they should stop using their brain at all. Basically all of the code I see is over-engineered BS that will almost always cause defects or introduces a ton of technical debt. In my opinion they are probably just producing negative work where in the next sprint Im usually the one who's assigned to fix that stuff (bec i always compl my work quickly).<p>So far I thought this sort of bs was just limited to my coworkers & managers would actually take care to produce good work / check our PR's carefully, using AI but checking it thoroughly. WRONG. First thing that I noticed is that the technical design document was like 5 pages long but since I wasn't really familiar with that part of the codebase I sort of just assumed it was accurate. Big mistake, after about a day turns out half the doc was just AI bs with nonsense randomly scattered throughout. I have had occasions where I ask my coworkers how their code works (bec AI added overengineered bs), and they literally can't explain it at all.<p>Anyway I was able to figure that stuff out and released the PR to my boss to review. Upon releasing it, he asked me to ask the AI to review the work and add the changes, I went ahead and did that and there were some obv AI hallucinations but some valid points like update the docs, add some error handling etc, & I made the changes. I release it back to him and he asks to have a call and we talk about it, turns out his AI had some other suggestions (to him), we went through it and figured out what was BS and which parts were valid (issues that were caused by the shitty design doc) & I made the updates on the call & pushed them.<p>an hour later the guy comes back and says something like "I asked the AI for some more recommendations, can you check the branch and compare the diffs to yours". So I look at the branch & sure enough it had all the BS we discussed on the call & determined was BS. It also changed the structure arbitarily making it more complicated & difficult to decouple & maintain. From what I can tell, either no one understands the codebase, can't understand the AI's code, or are just too lazy to think. A lot of what the AI does is based on previous code in the codebase, code that could be like 10 years old and be written like shit yet the AI will still recommend will follow that idea etc.etc<p>Im getting pretty dishearted, im literally the lowest ranking (in level) developer on the team yet doing the most work with the best quality (business analysts on the team have literally called this out), but im still paid the least & have to deal with the most BS.