What class of problems are being discussed? The OP seemed pretty open-ended, and it seems to me that for any problem, a well-designed organization will outperform a well-chosen individual. I agree that organizations have failures- but so do individuals.

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.