Yes. I agree with that as well. If you can intuit that it's (for example) an off-by-one error somewhere in your logic, asking an LLM to pinpoint that issue can save you tons of time that is better spent thinking about your abstractions. Also, I tend to ask high-level questions to the frontier models about best practices with common abstractions. But "the screen flickers here, please fix" is a bad pattern, IMO. Understanding why the screen is flickering is likely to educate you about where your abstractions are leaking.