Yes, creating my own boilerplate is attractive from a learning experience. Except when it's not learning but figuring out how to configure Webpack. Or how to get HMR working. Then it just becomes a chore and I explore Github for the best boilerplate for my needs.
To me, a boilerplate and a library aren't significantly different. The former is just exposed a lot more.
Sharpening your saw is great but if you just want some firewood you should use your trusty old blocksplitter.
However, if you code up a custom solution that ignores some important detail of some deeper area of knowledge like security or concurrency, then your "3-day" solution might literally wind up costing millions or hundreds of millions. (Exercise: Name some famous historical precedents from famous companies relevant today.) Of course, the downside for your garden variety consultant is probably only worse by a factor of 2 to 10, but the point remains.
This is why the mastery of knowledge and practical experience necessary to reliably do such a cost-benefit analysis is very valuable. The upside in making decisions like this is something like saving 4 days or a week of programmer time. The downside can be literally worth a 21st century fortune.
When picking the defaults to use without config, be sure they're settings that you want people to use.