Definitely this. The hard work, from my experience, in hooking up to a fuzzer is writing the wrappers to hook in to the fuzzer, and creating a set of examples

Definitely this. The hard work, from my experience, in hooking up to a fuzzer is writing the wrappers to hook in to the fuzzer, and creating a set of examples from which to begin fuzzing intelligently. Automated fuzzing is the icing on top of that hard work. It magnifies the utility of doing the work to fuzz by making it run continuously and by throwing more computational resources at it to find more bugs. But, unlike, say, coding to some closed API where you're up a creek if the service goes away, most of the hard work is portable, and you can still run it on your own cluster. AFL and libFuzzer are open source and quite awesome on their own.