Almost no skill required to cook a steak

煎牛排几乎不需要任何技巧

Cooking a steak requires almost no skill. Put it in a hot pan, wait a little, flip it, and eventually you’ll have something technically edible. But a genuinely good steak, medium-rare from edge to edge, browned properly, seasoned right, consistently delicious, is a different matter entirely.

煎牛排几乎不需要任何技巧。把它放进热锅里，等一会儿，翻个面，最终你总能得到某种“在技术上可食用”的东西。但要煎出一块真正的好牛排——从边缘到中心都保持三分熟、表面焦香适中、调味精准、口感始终如一——那完全是另一回事。

Software development with AI is starting to feel much the same. We build nonstop now. With AI, without AI, during the commute, on the toilet, probably in our sleep. We create agents, harnesses, tools, skills, prompts, feedback loops, elaborate workflows. Then we throw everything at a model and hope it gives us what we imagined, without ever having to understand how any of it actually works.

如今，利用 AI 进行软件开发的感觉也大抵如此。我们现在不停地构建：用 AI 构建，不用 AI 也构建；在通勤路上、在洗手间里，甚至可能在睡梦中都在构建。我们创建智能体、测试框架、工具、技能、提示词、反馈循环和复杂的流程。然后，我们将这一切一股脑儿丢给模型，祈祷它能给出我们想象中的结果，却从不去理解这些东西到底是如何运作的。

And what do we want? We want the perfect steak. We want software that works, looks good, feels polished, and arrives exactly as we imagined it. Most of all, we want the same result every time.

我们想要什么？我们想要那块完美的牛排。我们想要软件能运行、外观精美、体验流畅，并且完全符合我们的构想。最重要的是，我们希望每次都能得到同样的结果。

Do we get it? Not every time. Not even close to every time. Sometimes the model hands us something surprisingly good. Other times it serves up charcoal with a sprig of thyme on top and calls it medium-rare, completely confident in the lie.

我们得到了吗？并非每次都能。甚至可以说，远非如此。有时模型会给我们带来令人惊喜的作品；而另一些时候，它会端上一块撒着百里香的焦炭，却信誓旦旦地称其为“三分熟”，并对这个谎言深信不疑。

So what do we do? We go to a restaurant. We pay for a premium AI product, hire an agency, subscribe to another coding assistant, jump to a new framework promising professional results. We hope someone else already solved the problem for us. Sometimes they have. Quite often, they haven’t.

那我们该怎么办？我们去餐厅。我们购买昂贵的 AI 产品，聘请代理机构，订阅另一个编程助手，跳槽到声称能提供专业结果的新框架。我们希望别人已经帮我们解决了问题。有时他们确实做到了，但更多时候，他们并没有。

That leaves two choices: learn to cook properly ourselves, or keep asking friends for restaurant recommendations while preparing our wallets for the next expensive disappointment.

这留下了两个选择：要么学会自己好好做饭，要么继续向朋友打听餐厅推荐，同时准备好钱包，迎接下一次昂贵的失望。

Most of us want to build something we care about with AI without getting lost in the implementation details. We want to treat it like a professional chef working in our own kitchen: tell it what we want, step away, come back when dinner’s ready.

我们大多数人希望利用 AI 构建自己关心的事物，同时又不想迷失在实现的细节中。我们想把它当作在自家厨房工作的专业厨师：告诉它我们想要什么，然后走开，等晚餐准备好再回来。

But AI isn’t a chef. At best, it’s a steak machine. It can follow a recipe. Watch the temperature, flip at the right moment, drop in the butter. Give it enough tools and instructions and it’ll repeat that process fast, at enormous scale. What it doesn’t do is know what you actually want.

但 AI 不是厨师。充其量，它只是一台“牛排机”。它能遵循食谱，监控温度，在合适的时机翻面，加入黄油。给它足够的工具和指令，它就能快速、大规模地重复这个过程。但它无法知道你真正想要的是什么。

It can’t see the picture in your head unless you translate it into requirements, constraints, examples, tests, feedback. And even then, it’s boxed in by its own capabilities, its context window, the quality of the system wrapped around it. You can stand next to the machine and correct it every thirty seconds. That might help. It won’t turn the machine into a Michelin-starred chef.

除非你将其转化为需求、约束、示例、测试和反馈，否则它无法看到你脑海中的画面。即便如此，它仍受限于自身的能力、上下文窗口以及围绕它的系统质量。你可以站在机器旁边每三十秒纠正它一次，这或许有帮助，但它永远无法让这台机器变成米其林星级厨师。

Eventually, frustrated, you decide to just pay for the dream steak. You pick the expensive restaurant. Sit down, study the menu, finally, you can order with real confidence. You wait for the first bite. The plate arrives. Same burnt steak you made at home.

最终，在挫败感中，你决定直接为那块“梦想牛排”买单。你选了一家昂贵的餐厅，坐下，研究菜单，终于可以自信地点餐了。你等待着第一口美味。盘子端上来了。还是你在家做的那块焦糊的牛排。

Why? Because every restaurant in the city hired the same AI cook. “Cost optimization,” management says. “Most people won’t notice.”

为什么？因为城里的每家餐厅都雇用了同一个 AI 厨师。“成本优化，”管理层说，“大多数人不会注意到的。”

And they’re probably right. Most people won’t. Most of the time, software only has to be acceptable. Customers tolerate weird interfaces, pointless features, strange bugs, systems held together by generated code nobody actually understands.

他们或许是对的。大多数人确实不会注意到。在大多数情况下，软件只要“过得去”就行。客户容忍奇怪的界面、无用的功能、诡异的 Bug，以及由没人真正理解的生成式代码拼凑而成的系统。

But you’ll notice. You’ll notice because this was something you actually wanted to make. So you go home disappointed, hungry, a little embarrassed, and pull the cookbook off the shelf. There’s only one option left: learn to cook.

但你会注意到。你会注意到，因为这是你真正想要创造的东西。于是你失望、饥饿且略带尴尬地回到家，从书架上取下食谱。只剩下一个选择：学会做饭。

You learn what heat actually does. Which pan matters and why. Why thickness matters, why resting matters, why a timer alone was never going to save you. You ruin a few more dinners. Then you try again. And again. Eventually you stop depending on luck, you learned it the hard way.

你开始学习热量到底是如何起作用的。哪种锅重要，为什么重要。为什么厚度很重要，为什么静置很重要，为什么单靠计时器永远无法拯救你。你又搞砸了几顿晚餐。然后你再试一次，又一次。最终，你不再依赖运气，你通过艰苦的实践学会了它。

Software works the same way. AI can make you faster. It automates the repetitive stuff, spits out a starting point, explains code, helps you poke at ideas. What it can’t do is replace your judgment. It can’t define quality for you, can’t decide which tradeoffs are acceptable, can’t always catch the moment when something is technically correct but wrong in every way that matters.

软件开发也是如此。AI 可以让你更快。它能自动化重复性工作，吐出一个起点，解释代码，帮你推敲想法。但它无法取代你的判断力。它无法为你定义质量，无法决定哪些权衡是可接受的，也无法总是捕捉到那些“技术上正确，但在所有关键方面都错了”的时刻。

To build good software with AI, you still have to understand software. You need to know what you’re actually asking for, how to judge what comes back, and when the machine is just confidently serving you charcoal.

要利用 AI 构建优秀的软件，你仍然必须理解软件本身。你需要知道你到底在要求什么，如何评判返回的结果，以及何时机器只是在自信地为你端上一盘焦炭。

Keep learning. Keep building. Keep failing. Do that until you can produce the result you want instead of hoping to stumble into it. Then get good enough to open your own small restaurant. Then hire a few AI cooks. Most people still won’t notice the difference. But you will.

继续学习，继续构建，继续失败。坚持下去，直到你能产出自己想要的结果，而不是寄希望于偶然的成功。然后，磨练到足够出色，去开一家属于你自己的小餐厅。再去雇佣几个 AI 厨师。大多数人依然不会察觉到区别。但你会。