There's got to be a solution between those two extremes :)
I've started writing only integration tests, this is cool because most of the time it only breaks during refactoring when there actually are regressions. I've also made myself a chrome devtool to record my interactions with browser and generate capybara tests out of it, so it's quite fast to write and rewrite, it never happens anymore that I think "oh, maybe that change is not worth it".
But the thing is, this only works for apps/features having a web interface. I do more and more system programming, and I have no idea how to do integration testing on this (I think about using docker, but this would be even slower than piloting a browser).
So yeah, anyone who can solve the tests burden problem while still making code as robust, I'm interested.
(btw, you'll still find a bug in 6 months in production with 100% test coverage ;) )
