It makes sense to estimate R&D, it also makes sense to have limits. (Eg. some startup wants to solve global fidgeting, they raise a lot of money, and they have 3-4 ideas, so they allocate half of their money and one year on their best idea, then half of the remaining and half of a year on the second best, and ...)

If there's some set date (eg. demo/launch at big industry expo) it makes sense keep that constraint in mind, plan work in small chunks, iterate, and always try to deliver tangible things needed for the demo ... and it also makes sense to periodically check in (let's say at the end/start of every iteration) how the demo is coming along ... and it again makes total sense to adjust course if it seems that progress is better/worse than expected ... and it makes sense to try to do this tracking and estimation in some consistent/rigorous way, instead of just gut feeling.

"And that's how story points were born". (Which is just a fancy name for reference class based forecasting - https://en.wikipedia.org/wiki/Reference_class_forecasting )

And, again, if there's enough reliable data it's possible to give educated guesses about how the remaining tasks compare to the already finished ones.

...

Now, doing this top-down, from day 1, expecting 100% accuracy, and whatnot is complete business madness, and anyone who does that needs to be thrown out of a fucking (ground floor) window again and again until this gets through their (apparently) very thick skull.

...

I've seen this done well. (Not the defenestration.) And seen this done horribly. The "methodology" is not magic, and it definitely doesn't make sense if the people trying to "do" it are senseless imbeciles forcing things they don't even understand on people whom they also don't understand.