Literally the prompting and task definition is main variable how LLM's performs. There are literally millions of examples of vibe coders and new AI adopters who run out of tokens since they don't know how the LLM's work.