Fair — AI scaffolding can regenerate the CRUD layer fast, and I'd absolutely use it for that. But the part I'm worried about isn't generating code, it's the ownership call: a stale plugin in a client project still has migrations, config, and real production data behind its table names. AI can help write the replacement faster, but deciding fork-and-maintain vs rip-and-replace (and planning the data migration on a live project) is still a judgment call that eats hours. That's where the day you saved can quietly evaporate.
The migration-collision warning is the most important paragraph here — it's the real cost of plugin-based admin panels. "Rename yours first" is honest, but it means part of the day you saved goes into migration ordering and table-name conflicts, especially once two plugins both want to own a concept like categories. Question: with dozens of packages in vendor/, what's your policy when one goes stale mid-client-project — fork and maintain it, or replace it? That decision is what determines whether plugin-first actually stays cheaper than building the resource yourself.