I've also seen it fail the other way where the Skunkworks becomes the preserve of "high achievers" not producing a lot that's useful. Almost every 'Office of the CTO' I've seen has suffered that failure mode. Lot's of things get built but none are more than fluffy experiments and there's no path to actually putting any of those things into production. Once in a while they'll be forced through and the production team has to spend a lot of effort fixing and making the project actually work.

I'm not arguing against the autonomy part, on the contrary: I join you in arguing in favor of it, and also that such an autonomy requires a brilliant personality at the head of it.

However, while the combination of autonomy and brilliant personality might be a prerequisite of outstanding success, I'm not yet convinced that it's an overwhelming indicator of it. I believe that a brilliant but toxic personality with autonomy can lead to an ultimately fatal instability.