Of course, but why are we bad at it after so many iterations? I think a comment above mentioned that there is actually incentive to get estimate to skew lower and worry about the extra costs later. What does a world look like where estimation is correct within a reasonable degree and people don't race to the bottom to get the lowest bid so that they can inevitably put out a crap product?
The blessing and curse of digital delivery of software is that retooling is comparatively easy to physical products and distribution cost is pretty negligible.
Getting it right up front matters a lot less for most software than it does something like a product where you invested a bunch of money in tool and die, stood up an assembly line, have warehouses full of product to deal with of you release a new version that obsoletes them, and so on.
If software is incomplete or wrong or bad, the real costs are mostly just labor to fix it. If physical products are bad, the very act of doing the fix has costs beyond the labor of the design work.
(I’m certainly not endorsing sloppy planning in software, fwiw, this is just how the incentives play out much of the time)
There's also the issue of requirements changing many times over, based on new requirements are added or limitations that are discovered only after making some progress in the project.
