I tend to have the opposite experience from the author's: I know enough about the domain to know that choices I make at the beginning could cause massive problems later on (in performance-critical software, for instance), yet not enough to know exactly where the traps are. Once the project is underway and empirical data can be collected, it's much easier to direct attention to known issues and opportunities.

By nature I enjoy the 'boring' stuff, to use the author's term, and often find a clean code-to-package CI pipeline and an up-to-date wiki more satisfying than the software itself!

I wonder what the ratio of 'non-finishers' to 'non-starters' is. I suspect I'm in the minority, but maybe someone here knows of a reliable study of this.