one of the funniest things about using agents for coding is that when you're in the planning phase it'll give you these humble timelines of implementation that last weeks or months, not "realizing" it'll do the task in minutes or hours.

this is interesting. Am guessing it because that’s the kind of estimates it has learnt from and does that put a ceiling on llms can aspire for - so much similar to humans in that way

not entirely surprising because it's trained "in our image". but it's totally like us, overestimate how much we can do in 1 day, underestimate how much we can do in 1month.