AI Won't Make a 60-Point Engineer a 90-Point One

Hello everyone, I'm Feiyu.

Today's article is another reiteration of a point I've made before, because recently people keep asking me: in the age of AI, what should I still learn?

First of all, AI is indeed very powerful, but it is absolutely not a universal key. It can help you improve efficiency, but the premise is that you already know what you want to do.

If you don't have your own judgment and lack your own solution approach to a problem, then in the end you definitely won't achieve the expected results.

Don't think you're amazing just because you used AI to write a personal homepage or made a small tool.

You have to think about what exactly you made this thing for? Where do you plan to use it?

If AI cannot truly enter your business, workflow, or the things you need to produce, and then concretely solve your actual problems, it is most likely just a tool that satisfies your emotions once.

So what is truly valuable now is not that you learned yet another tool today, like Codex, ClaudeCode, or Xiaolongxia:

It's that these tools genuinely helped you solve a specific problem.

When I was reading the ClaudeCode source code, one particularly strong feeling was its pragmatic realism.

Its overall design philosophy gives absolutely no sense of showing off skills.

ClaudeCode can be said to be the most top-tier Agent currently, but technologies like RAG are not used, because it simply doesn't need RAG. RAG is fundamentally not suitable for its actual use case.

It uses the most straightforward methods, such as commands to search files, commands to view files — just particularly simple basic commands.

Looking at each command and each small design individually, none of them are surprising, but when solving specific problems, straightforward methods are often the most effective.

Many people think that using AI is meant to save effort, so why does it instead demand more from me?

But that's just the reality. The quality of the questions you ask determines the height you can reach.

Prompts are not better just because they are longer. If you yourself haven't clarified what you want to ask, what knowledge to provide to the AI, and what constraints exist.

If you haven't clarified these yourself, then AI certainly cannot reply according to your expectations.

At this point, you can't complain that AI is no good; it's your ability to ask questions that is lacking. You need to know what you want, what you're missing, how to verify, and where to add constraints.

So don't pin your hopes on AI turning someone who can only score 60 points in 2 hours into someone who scores 90 points in half an hour.

At most, it can help you increase your answering speed, or compress a test paper that originally took two hours to finish into one hour.

But your upper limit can only reach 70 points at best. It's impossible for someone who scores 60 points to suddenly score 90 points just by having AI.

Just like giving an intern unlimited tokens for top-tier models won't turn them into an architect in the short term.

The determining factor is still the person, not the model.

Always remember: the experience accumulated after hitting a wall is the most valuable. This has more practical significance than using the tools themselves.

I hope everyone will do more trial and error and grow in the AI era.

Finally: The article will be included on my personal website: http://hardyfish.top/