HN 提问:谁才是 AI 编程的真正赢家?

2 分•作者: unclebucknasty•24 分钟前
我在三个不同的 Claude 编码项目中都观察到了相同的模式:初始势头强劲,但随后开始减弱,直到回报变为负数。 这似乎是模型行为变化和代码库规模不断增大的组合效应,其中后者尤为突出。代码库越大,以及记录的历史/进展越多(例如,通过更新已完成的工作和说明,如 .md 文件),就会出现越多的“漂移”。 在某个阶段,就会出现“崩溃”。此时,它无法在单个工作单元内保持一致性。例如,这可能表现为向视图添加一个相似的元素,但其样式与现有元素完全不同。这感觉完全是随机的。 它也可能表现为不连贯或实现不佳的逻辑和行为。我尝试了不同的建议,重构了项目,进入了计划模式,添加了确定性检查、样式、规则等。但感觉就像是“打地鼠”,而且不清楚它一开始是否真的遵守了这些规则。 此外,还有一个问题是它是否在“转移问题”。在我抱怨 UI 生成不一致时,它建议在 UI 更改后运行样式检查。当然,这引出了一个问题:为什么它不在初始生成时就应用这些规则呢? 但现在,我发现它会启动 AI 代理来进行完整的 UI/UX 审查——包括 Playwright。然而,它仍然在“漂移”,并且没有捕捉到问题。这与它的单元测试一致——在我最新的项目中,它已经生成了一套数百个耗时较长的测试,并且在提交前不断运行,但没有一个测试失败过。与此同时,却出现了一些 bug。当被质问时,它说它一直在测试“错误的层”,并且基本上生成了自我确认的测试。 我的问题是,这里**真正**的赢家是谁?在搜索 Google 时,它指向了许多成功的声明以及据称促成成功的各种方法。但是,当你深入分析这些声明的底层来源时,似乎主要来自 Reddit。 然而,进一步深入研究会发现,几乎所有的实际研究都表明,生产力几乎没有提高,甚至出现退步——尤其是在经验丰富的工程师方面。 那么,谁在这里使用 AI 编码开发了任何复杂度的优质生产应用程序?如果有,使用的是哪种框架/模型,以及使用了什么技术?另外,您能指向您的代码仓库吗?
查看原文
I&#x27;ve noticed the same pattern across three different Claude-coded projects now. Initial velocity is strong, but starts to fade until returns are negative.<p>It seems a combination of changes in model behavior and increasing code base size—emphasis on the latter. The larger the code base and documented history&#x2F;progress (e.g. through updates to completed work and instructions, as in .md files), the more drift starts to set in.<p>At a certain point, you get collapse. This is where it can&#x27;t maintain consistency within a single unit of work. It can manifest, for example, as adding a similar element to a view, but styling it in a completely different manner than existing elements. It literally feels random.<p>It can also manifest as incoherent or poorly implemented logic and behavior. I&#x27;ve followed different advice, restructured projects, entered plan mode, added deterministic checks, styles, rules, etc. But, it feels like whack-a-mole and it&#x27;s unclear that it actually adheres to these rules in the first place.<p>Then, there&#x27;s the question of it moving the problem. In response to my complaints of inconsistent UI generation, it recommended a style check to run after its UI changes. Of course, that begs the question of why it doesn&#x27;t just apply those rules in the initial generation.<p>But, I am at a place now where it spins up AI agents to do a full UI&#x2F;UX review—Playwright and all. Yet, it&#x27;s still drifting and not catching it. This is consistent with its unit tests—on my latest project it has now generated a suite of hundreds of somewhat time consuming tests that it constantly runs pre-commit, and not a single test has ever failed. Meanwhile, there have been a number of bugs. When questioned, it said it had been testing the &quot;wrong layer&quot; and had essentially generated self-confirming tests.<p>My question is who is <i>really</i> winning here? When querying Google, it points to so many claims of success and approaches that supposedly contribute to that success. But, when you break down the underlying source for those claims, it appears to be largely Reddit.<p>Yet, pressing it further reveals that virtually every actual study shows either little to no productivity gains or actual regression—particularly where experienced engineers are concerned.<p>So, who here has developed a quality production application of any complexity using AI coding? If anyone, which harness&#x2F;model, and what techniques did you use? Also, can you point us to your repo?