Even as LLMs get better over time at understanding ill-formed prompts, I expect that API prices will still continue to depend on the number of tokens used. That’s an incentive to minimize tokens, so “prompt engineering” might stick around, even if just for cost optimization.