AI 能写代码，但你能看懂吗？
AI can write code. But can you understand it?
一位开发者提出，AI 已能生成函数、建议架构、解释报错、编写测试甚至构建完整功能，但产出代码与理解软件是两回事。若不懂 API 原理、请求与响应之间的过程、数据库行为以及复杂度、架构、安全和权衡，即便生成 500 行代码也无法判断其是否正确、可维护、安全或必要。他认为 AI 并未取代学习的必要，反而提升了学习的价值，不应把理解外包出去。
One thing I’ve been thinking about a lot lately as a developer is the importance of real knowledge.
We have incredible tools today. AI can generate functions, suggest architectures, explain errors, write tests, and even build entire features.
But there’s a difference between producing code and understanding software.
Knowing why an API works the way it does.
Understanding what happens between a request and a response.
Knowing how databases actually behave.
Understanding complexity, architecture, security, testing, and trade-offs.
These things become even more important when you have AI helping you.
Because if you don't understand the fundamentals, you may be able to generate 500 lines of code—but you won't necessarily know whether those 500 lines are correct, maintainable, secure, or even necessary.
For me, AI isn't replacing the need to learn.
It's increasing the value of learning.
The better I understand software engineering, the better I can use AI as a tool instead of depending on it as a crutch.
Learn the fundamentals.
Build things.
Break things.
Debug them.
Read other people's code.
Ask why, not only how.
And use AI. A lot.
But never outsource your understanding. 🧠💻
Tools can accelerate your work. Knowledge gives you control over the result.
来源：Google AI：DEV 作者专属（RSS） · dev.to