AI output often 'looks like something' on first try, which makes it easy to assume no effort went in.

But there's a big difference between prompting and accepting the first output versus someone using search, multiple LLMs, actually READING the underlying papers, and iterating until it's done.

Sometimes that still means getting to 'done' faster than by more traditional means. Sometimes it means more depth than you'd manage otherwise. Sometimes somewhere in between.

Of course, by that point, either way, it doesn't really look like lazy AI output anymore.

Maybe it's not so much about the tools/agents as it is about the intent-to-engage behind them?