I often think it is hard to work test-first.
Then, I remember how difficult it is to work without tests.
Once I created a GUI for a physical product. Something similar to an admin panel for a router.
After a while, I realised I would not be able to keep up with the rquired changes, and at the same time keep my confidence in the code I wrote.
I added tests. The problem was: I kept the tested (testable) version aside from live code.
Reason: I did not trust it well enough. It was way different from the live, time-proven code.
With every other "refactoring" I did to the original application, my confidence in it diminished - because even though tests were passing, I had no quick way of checking that the application as a whole still behaved how its users expected it to.
This way, I never got the chance to put the refactored version into production. In the end, I still trusted the "not refactored" version of the code much more.
I would have been much better off working in tiny steps, all the while making sure that with each step, the application still works as the users expect it to. This way in the end there would be only one version of the app to maintain.
It turns out I was not the only one who fell into a similar trap of a "grand refactor". It must be pretty common, because Uncle Bob has a special term for this kind of endeavor – "the grand redesign" [1].