In the past month, I've managed to merge 2 pull requests into other people's projects that prevented the application from starting up properly. They weren't even exotic configurations, but alas, it didn't run.
Of course, what I should have done was run the service, test it locally to make sure it works, etc. But both these projects required a significant amount of setup and run time dependencies, so I just trusted that the tests in CI (continuous integration) did its job.
The tests in CI... well they mostly unit tests, one of the projects had near 100k of them. One project did have a form of integration tests, but the main entrypoint was different and wired up through the testing framework. This is how we ended up in a situation where the deployed application can fail at startup in a live environment...
All of this is to say: test the real application in CI.