AI is great at design but terrible at algorithms
by•
Claude and Gemini crush system design and high-level modelling, but the moment you drop into actual function logic, they fall apart. It’s like pairing with someone who writes clean diagrams but chaotic code.
And if you ask them to improve it, they vanish into SOLID-theory rabbit holes and over-engineered OOP madness.
These days, since I vibecode more, I have found myself refactoring what the robots are writing ever so often that it has got to a point where I straight up delete function logic and write it myself.
Curious if others are seeing the same?
51 views
Replies
I'm not sure really what you are talking about, LLms like gpt or claude can literally crush any algorithm case scenario as long as they are provided with the full context.
Plus I don't really buy this algorithm thing, algorithms start to be important when you start serving over 100k+. other than that, the difference is nothing.
@sharif_hasan4 I disagree, I find that basic functional logic can be written easily. When things get more complex, either mathematically or based on my recent experience - proper parallel/reactive programming, AI written code can only serve as boilerplate or "hints" as to what can be done. Sometimes, it even throws me off because I am biased to the solution presented in front of me, when instead, I can come up with something simpler on my own.
Well, these challenges are real. I have experienced them. You have to write your rules well and not to be dependent on one tool. I usually purchase 3,4,5 different $20 subscriptions rather than going for a single $100 subscription. So, keep refactoring your code time to time or as soon as a set of features is ready. Especially, clean up the .md files written by AI which creates problems in context. But, yeah over the past few months I have seen a drop in performance, not sure if it's intentional to launch something new.
I think if the functionality or design is simple and straight forward then these tools are great but when you try to achieve something which is complex then you need to do the at least logic building yourself and take boilerplate code and enhance it yourself. although AI is of great help as you now don't need to write all the basic code again and again.