Seeing how many people agree with the post I am left wondering who the commenters on HN are. With a large code base and actual customers, comprehensive unit testing is by far the most cost effective way to create and maintain software. This is not being dogmatique, it's the shared experience of the majority of established software engineering firms. In the case of my company, we experienced so much growing pains that our productivity almost came down to zero before we switched to TDD. Many developers in our group were skeptics (including me) but today you won't find one of us arguing for less testing.

The kind of large pivot that the author refers to is only possible when you don't have established customers and you have a minimal product. You may as well call it prototyping. And prototyping with or without tests is indeed more a matter of taste than effectiveness.