Aoede needs a voice before it can do anything. The problem is that getting one is also the least obvious part of the experience. So the first launch now walks through it: a welcome, a short tour of how it works, a choice between Apple's system voices and Kokoro, the setup for whichever one you picked, and importing your first book. The entire flow can be revisited later from the Help menu or Settings.
The setup step is the reason onboarding exists at all. macOS hides its good system voices behind VoiceOver Utility, a place no one would think to look, so if you choose Apple voices, onboarding sends you straight there instead of leaving you to discover it. If you choose Kokoro, the model download happens on an explicit step with a real byte-progress bar, rather than kicking off silently the moment you select the engine. Only the step for the engine you picked shows up, so nobody sits through setup for a voice they are not going to use.