The LLMs/agents are that strong these days, and become stronger almost from month to month now, that they behave like pretty experienced software developers, even architects, you just have to give precise requirements to implement, give hints when they don't follow, and ask to review their results.

The software has its automated tests and I let it analyze what non-functional requirements are maybe missing from time to time. The same with the user experience. So the software architecture becomes strong(er). But you need real-life tests because of the unexpected situations that may come up.

In the end you need the hardware to test how the software reacts. Atm I just focus on what I have available. And I would test it in several gigs myself before sharing something.

So it's not that AI slop of people with almost no experience, but the advanced tooling of the experienced pros in creation.

This is what I'm seeing with other vibe coding approaches. Closer to the needs of the individual. With the option to become a solution for many. And those solutions can be adapted or further developed pretty fast and with way less money investment.

Advantage to me: I'm able to use my old (unsupported) Native Instrument controllers, or a combination of different controller vendors (without MIDI lean), and have a combination of DJ tools, that the established software vendors don't offer that aggregated.

Only the app creation is done with vibe coding. Even as software developer myself and a musician, both with decades of experiences, I'm not able to create that fast with a focus on audio-specific development aspects.