4 min read ai

Daily Productive Sharing 1277 - Writing Code Was Never The Bottleneck

Daily Productive Sharing 1277 - Writing Code Was Never The Bottleneck
Photo by Sebastian / Unsplash

One helpful tip per day:)

Pedro Tavares has realized that the marginal cost of adding new software is approaching zero, especially with the advent of large language models (LLMs). But what about the cost of understanding, testing, and trusting that code? It’s higher than ever.

  1. He’s not sure whether the author fully understands what they’re submitting.
  2. He’s noticed that generated code often introduces unfamiliar patterns or breaks established conventions.
  3. In his view, the real bottleneck in the past—and still today—is code review, knowledge transfer through mentoring and pair programming, testing, debugging, and the human cost of coordination and communication.
  4. Ultimately, he finds we’ve landed in a situation where producing code has become easier, but verifying it has become more complex—which doesn’t necessarily make teams move faster overall.
  5. He recalls how developers used to joke about “copy-paste engineers,” but the speed and scale enabled by LLMs have only amplified those copy-paste habits.
  6. In his view, software engineering has always been collaborative work, relying on shared understanding, team alignment, and the transmission of expertise.
  7. He firmly believes that LLMs haven’t removed the need for clear thinking, careful review, and thoughtful design. On the contrary, as more code is generated, these things have become even more crucial.
  8. He acknowledges that the cost of writing code has indeed fallen, but the cost of helping the team collectively understand it hasn’t decreased at all.
  9. In his view, while LLMs do reduce the time it takes to produce code, they don’t reduce the effort needed to understand how that code behaves, uncover hidden bugs, or ensure long-term maintainability.

If you enjoy today's sharing, why not subscribe

Need a superb CV, please try our CV Consultation

Pedro Tavares 发现，添加新软件的边际成本正在接近零，尤其是有了大语言模型（LLMs）。但理解、测试以及信任这段代码的代价呢？比以往任何时候都高。

  1. 他不确定，作者是否完全理解自己提交的内容。
  2. 他注意到，生成的代码引入了不熟悉的模式，或者打破了既有的惯例。
  3. 在他看来，真正的瓶颈过去是、现在依然是：代码审查、通过指导和结对编程进行知识传递、测试、调试，以及协调与沟通所带来的人力成本。
  4. 最终，他发现我们陷入了这样一种局面：代码生产变得更简单，但验证却更复杂，这并不一定让团队整体运转得更快。
  5. 他回忆，开发者们过去常常拿“复制粘贴工程师”开玩笑，但 LLM 所带来的速度与规模，却让这种复制粘贴的习惯被进一步放大。
  6. 在他看来，软件工程从来就是一项需要协作的工作，它依赖于共享理解、团队协同和经验传授。
  7. 他坚信，LLM 并没有消除清晰思考、仔细审查和深思熟虑设计的必要。恰恰相反，随着越来越多的代码被生成，这些事情变得更重要了。
  8. 他承认，写代码的成本确实下降了，但让团队一起搞明白代码的成本，却一点也没减少。
  9. 在他看来，LLM 确实缩短了产出代码的时间，但并没有减少理解代码行为、发现隐藏漏洞，或保障长期可维护性所需的努力。

如果你喜欢的话，不妨直接订阅这份电子报 ⬇️

Dr Selfie

Dr Selfie