Execution Is Cheap. Thinking Isn’t
打开互动全文版（逐段中英对照 + 图/公式 + 论文问答）→这篇文章探讨了在工作和生活中，人们往往过于注重执行而忽视了思考的重要性。作者认为，虽然执行任务看似高效，但真正的价值在于深入的思考，它能带来创新和长期的成功。文章鼓励读者花更多时间进行战略性思考，而不是盲目地执行任务。
[](https://www.mynameismartin.co.uk/)- [x]
[](https://www.mynameismartin.co.uk/)- [x]
[](https://www.mynameismartin.co.uk/)- [x]
本周我发布了一个小型网络应用 Tone Ladder，这是我借助 AI 进行产品开发的第一次认真尝试。我在过程中做了笔记，所以这里混合了我的实时笔记和一些反思。
This week I launched a small web app, Tone Ladder, which was my first serious step into product development assisted by AI. I took notes during the process, so here’s a mix of my live notes and some reflections.
如果没有 Claude Code（一款 AI 智能体编码工具）的支持，我无法独自完成 Tone Ladder。我掌握的 JavaScript 足以让我作为一个偏技术的设计师应付过去，也发布过不少前端项目，但像 Tone Ladder 这样的应用所需的工程深度超出了我的技能范围。
I couldn’t have made Tone Ladder alone without the support of Claude Code, an AI agentic coding tool 11 Anthropic’s Claude Code was my weapon of choice, with thanks to Dan, who was willing to patiently walk me through his process to get me started.. I know enough JavaScript to get by as a technical-leaning designer, I’ve shipped plenty of front ends, but the engineering depth required for an app like Tone Ladder is outside my skillset.
我在工作时做了笔记并记录了一些想法，下面是我整理的内容，但最重要的收获是下面这一点，你读完这个就可以离开了：
I made notes and recorded thoughts as I was working, which I’ve written up below, but here’s biggest takeaway of all, and you can read this and leave if you want:
有了 AI，执行变得廉价，但没有好的计划，执行就毫无价值。现在比以往任何时候都更需要设计——我指的不是 UI，而是对问题空间的深刻理解——这是成功的关键。
With AI execution is cheap, but without a good plan, execution is worthless. Now more than ever design – and I don’t mean the UI, I mean the applied understanding of the problem space – is critical to success.
如果你在行业里待了几年以上，你应该已经吸收了敏捷开发的论点。尽快获得可运行的软件对我来说是第二天性，但这让 Claude 变得更糟——每个半成想法中的模糊性都被放大了。
If you’ve been in the industry more than a few years, you’ve absorbed the arguments for agile. Getting to working software as quickly as possible is second nature to me, but it made Claude worse – the ambiguity in every half-formed idea became amplified.
这感觉像是敏捷思维的一种哲学倒退。我对‘前期’理解得越深，结果就越好。我对正在构建的东西越确定，Claude 就能做得越多 22 我后来意识到这是一个有先见之明的想法。规范驱动开发是当下热门的‘新’事物。
This felt like a philosophical rollback of the agile mindset. The more I understood ‘up front’, the better the results. The more certain I was about what I was building, the more Claude could do22 I’ve since learnt this was a prescient thought. Spec driven development is the hot ‘new’ thing..
Claude 按字面意思理解委婉的语言，它填补了我理解中的空白而不要求澄清，并自信地加速朝着错误的方向前进。
Claude took soft language literally, it filled gaps in my understanding without asking for clarification, and confidently accelerated off in the wrong direction.
在没有清晰意图的情况下提示 Claude 感觉就像掷骰子：我让 Claude 做某事，猴子爪子卷曲，我咬紧牙关等待后果显现。
Prompting Claude without clarity and intent felt like rolling a die: I ask Claude to do something, the monkey paw curls, and I wait, with gritted teeth, for the consequences to reveal themselves.
这是我的第一个 AI 项目，随着时间的推移，我变得更擅长提示和检查 Claude 的工作，但我从未感觉我们像人类合作者那样融洽，彼此激发能量和热情，创造出超越各部分之和的东西。
This was my first AI project and I got better at prompting and checking Claude’s work with time, but I never felt like we gelled in the way I would with a human collaborator, bouncing off each others’ energy and enthusiasm, making something greater than the sum of its parts.
我无法信任 Claude 能保持上下文或尊重人类可以推断的隐形边界，除非我事先明确定义了它们。
I couldn’t trust Claude to hold context or respect the invisible boundaries a human can infer, unless I’d specifically defined them in advance.
当我在一个容器内工作时，Claude 表现最佳。我用通俗语言编写了一份行为契约，描述了应用应做什么、哪些结果可接受，并定义了一系列测试，以防止回归，或标记 Claude 决定采取一条短而具破坏性的路径来达成目标的情况。
Claude worked best when I’d given it a container to work inside. I wrote a behaviour contract in plain language describing what the app should do, what results were acceptable and defined series of tests to protect against regression, or flag where Claude had decided to take a short, destructive route to a goal.
这让我重获掌控感，同时又不限制 Claude 带来不同且创造性解决方案的潜力。而且它解锁了这种方法，Claude 能够在我并不熟悉的领域提出想法和解决方案，而我自己原本不知道该如何提示才能获得这些。
This gave me back the feeling of control without constricting the potential of Claude to bring different and creative solutions. And it unlocked the approach, Claude was able to bring ideas and solutions in a space that wasn’t familiar to me, and wouldn’t have known to prompt for.
下次与 Claude 合作时，我会在流程中更晚的阶段才使用它。我会先完成思考、界定边界、写下预期。我会把 Claude 当作原型设计工具，而非协作者。
Next time I work with Claude, I’ll work with it much later in the process. I’ll get the thinking done, the boundaries defined, and the expectations written. I’ll bring Claude in as a prototyping tool, not a collaborator.
用 Claude 制作东西既便宜又快速。但制作错误的东西也同样便宜且快速，有时 Claude 甚至会激烈地为自己辩护，声称它*确实*做了正确的事情——_完全_按照我的要求做了。
Making things with Claude was cheap and fast. But it was cheap and fast to make the wrong thing, and sometimes Claude would even defend itself passionately that it *had* done the right thing – _exactly_ what I’d asked.
在开始这个过程时，我有一些预期。
I had a few expectations going into this process.
我希望得到一个能用的应用程序。我希望自己能看穿炒作和虚张声势，并安心地知道我的职业暂时还是安全的。我实现了其中一件事。
I hoped to get a working app. I hoped I’d cut through the hype and bluster and relax in the knowledge my profession was safe for a while yet. I got one of those things.
我没想到会重新发现多年前刚开始尝试网络、为朋友的乐队建网站时的那种感觉。一种充满可能性和乐观的感觉。
I wasn’t expecting to rediscover the feeling I got when I first started experimenting with the web, building websites for friends’ bands all those years ago. A feeling of possibility and optimism.
我最终得到的是既恐惧又兴奋：创作突然变得毫不费力，而决定什么值得创作现在可能才是设计的真正工作。
What I came away with was equal parts fear and exhilaration: making things is suddenly effortless, and deciding what’s worth making may now be the real work of design.
1. 1 Anthropic 的 Claude Code 是我的首选工具，感谢 Dan，他愿意耐心地引导我完成他的流程，让我得以入门。↩
1. 1 Anthropic’s Claude Code was my weapon of choice, with thanks to Dan, who was willing to patiently walk me through his process to get me started. ↩
2. 2 后来我才知道这是一个有先见之明的想法。规范驱动开发是当下热门的‘新’事物。↩
2. 2 I’ve since learnt this was a prescient thought. Spec driven development is the hot ‘new’ thing. ↩
上一篇 构建音调阶梯下一篇 作为设计师，我如何应对采用 AI 的压力
Previous post Building Tone LadderNext post How I’m dealing with the pressure to adopt AI as a designer
Martin 是一名设计师，致力于为政府、医疗、慈善和私营部门设计复杂的数字服务和产品。
Martin is a designer working on complex digital services and products across government, healthcare, charity, and the private sector.