It's really hard to put any single root cause on it. Firstly, mobile OSes are massive beasts with lots of different teams working on it. Co-ordinating this becomes very difficult. Additionally, you typically get about 1 build a day which means there's a long lead time to find a bug that's managed to propagate out and then a long lead time for it to get fixed (at least 1 day). I think this decrease in quality has been seen across all OS vendors, it's just that Apple has become one of the dominant OSes so there's more press about it than there was about previous Apple releases.
So now you have a larger volume of bugs since bug count is based on the amount of code written, assuming best case that developer quality is roughly consistent across all teams. Since you have a larger volume and only so many hours in a day, trying to ship all software at once at an annual cadence means your bug count will go up regardless of how much QA you throw at the problem (since in addition to finding issues you have to fix them).
Automated testing might help because you could catch bugs before they're even submitted. Well, automated testing is still very hard on these devices when you're building the OS. For example, how do you do something as simple as build & run unit tests? If the answer involves loading code onto a mobile device that means (at Apple scale) you need a farm of devices. Moreover, your test isn't valuable if it's running against yesterday's OS. Might not even build/run because you've now acquired a dependency on an API made available in the new OS. So now you have to co-ordinate building the software using an SDK that may be unstable (build breakages happen), run it on an OS you somehow have to pre-validate works otherwise & then have some confidence that the bug found was introduced by your commit & not some combination of build issue, compiler bug (oh yeah - the compiler changes drastically too while your building your next OS), or downstream dependency bug. Finally, on top of all this you need to make sure you can recover bricked devices in an automated fashion so that the maintenance overhead of your automated testing cluster is manageable. All of these aren't impossible problems to solve. The problem is that it requires a lot of effort & physical infrastructure. Additionally, Apple has had a very ad-hoc development environment where no 2 teams are the same (it was slowly changing to unify large pieces of infrastructure when I left) which complicates the ability to roll out central infrastructure. Additionally, there are different needs - the kernel has different testing requirements/strategies from system daemons, from drivers from 1p apps from backward compatibility testing.
When I left a couple of years ago Craig had actually directed his teams to improve the automated testing infrastructure but at the end of the day the software is growing more quickly than the investment that was made in QA. There's also only so quickly you can roll out central automation infrastructure to an organization as large as Apple so it's not as simple as "do more QA" or "add more QA resources" not to mention that "QA resources" in automated testing typically requires very skilled software engineers to design & build the systems. Allowing a longer bake time is exactly the right move here as it recognizes the need to decrease the volume of bugs while infrastructure catches up.
