If your definition of “real coding” is “I personally typed every line,” that definition is going to die
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
这条记录涉及编程工具或代码能力更新，适合开发者评估工作流变化和可复用价值。
“亲自敲写每一行代码”的“真正编码”定义正在过时。ChatGPT/Codex等AI编码工具并不会淘汰工程，反而能增强系统思维、调试、架构、产品判断和验证等优势。关键在于理解契约、检查差异、测试边缘情况、验证集成并承担故障责任，而不是纠结于谁生成了最初的令牌。
I do not think AI coding makes engineering obsolete.
I do think it makes one old definition of engineering obsolete: equating professional legitimacy with manually producing the implementation.
Typing code is a means, not the outcome.
If ChatGPT/Codex generates a function and I understand the contract, inspect the diff, test the edge cases, verify the integration and own the failure when it breaks, the meaningful engineering question is not who emitted the first draft of the tokens.
The meaningful question is whether the resulting system is correct.
The “AI slop” criticism is valid when people skip that ownership. Prompt once, never read the code, deploy it, then act shocked when it fails — yes, that is garbage.
But a senior engineer using generation as leverage is a completely different workflow.
The irony is that AI should increase the value of experienced judgment. The faster code appears, the more important it becomes to know what should exist, what should never exist, and how to prove the difference.
If an engineer’s entire competitive advantage was memorizing syntax and moving boilerplate around faster than everyone else, that advantage is shrinking.
If the advantage was systems thinking, debugging, architecture, product judgment and verification, these tools can amplify it.