Mandatory dogfooding ONLY works if the development/QA/product teams listen, FIX THINGS, and also dogfood themselves.
I say this having seen the jumps betwen Solaris (outsider-only)2.0, 2.3, 2.4, (joined-Solaris-team-here) 2.6.
But at least even as an outsider I could answer, "What problem is $NEW_THING trying to solve?"
I do NOT see *any* hard-core GenAI proponents answering that question honestly. There *are* honest answers, but none that justify the hype some want to generate.