AI coding requires patience. Not because the models are slow, but because the workflow itself is iterative.
Many people say AI coding tools are overhyped because they donβt give perfect code in one prompt. But that expectation is wrong.
AI works better if you treat it like working with a very fast junior engineer. Give a small task. Look at the output. Fix the instruction. Repeat.
If you ask for a whole system in one prompt the result is usually messy. But if you break the work into smaller parts like functions or modules the results become much better.
The key skill is not prompt tricks. It is patience and breaking problems into small pieces and iterating.
It's fun vibe coding small, disposable bits of software with tightly bounded functionality. For anything real though, typing is faster.
i think that claude code tries to do the same, it tries to break the work into smaller parts, the only thing it relies on "intuition" when breaking problems, so of course leaving no ambiguity is the best thing you can do when AI coding (i have only started using claude code for a month now)
>If you ask for a whole system in one prompt the result is usually messy. But if you break the work into smaller parts like functions or modules the results become much better.
Full circle. If one has to do the full process of decomp, one might as well just noodle the code and save on the power/compute bill. The human brain is pretty goddamn efficient compared to a rack full of server hardware and GPU's. Of course humans also come with a problem from a capitalist's point of view. Three really. You can't own them, they aren't trivial to reproduce, and they can say no. AI has no such limitations. Hence the catnip like quality of the space.
Small task β review β adjust β repeat.
That's not an AI workflow. That's how any solid dev process should work regardless of tools.