My experience has been the opposite - we've been maintaining large (tens of millions loc, thousands of developers) mobile app codebases with this approach for about 8 years now, and results have been much better than what was done before.
A number of bad decisions were made and later discovered over time, and it was much easier to address them with automatic dependency injection available than without. Using as "big" of a test as you can without regressing speed or flakiness gives you really solid, non-fragile tests, which makes large changes tractable to do safely.
I don't know whether certain poor designs would have been spotted sooner, but this certainly made the code easier to change and resulted in more useful and lower maintenance tests.
A number of bad decisions were made and later discovered over time, and it was much easier to address them with automatic dependency injection available than without. Using as "big" of a test as you can without regressing speed or flakiness gives you really solid, non-fragile tests, which makes large changes tractable to do safely.
I don't know whether certain poor designs would have been spotted sooner, but this certainly made the code easier to change and resulted in more useful and lower maintenance tests.