Indeed, it seems we're more likely to get a recursively self-improving AI (with or without the G) through organizational design than by approaching the problem directly.

Precisely the point I was trying to make with "well knit human computer team"