Good points. I would add to his last point though that hiring at least one person with a high y-intercept can increase the slope of your other hires significantly. (It's a lot easier to learn something quickly if you have someone around who already knows a lot about it.) And people who both know a lot and learn quickly are expensive and rare, so I expect the optimal hiring process would look for a large percentage of relatively inexperienced fast learners, along with a small (but non-zero) percentage of more experienced people, even if they aren't likely to develop as quickly. (Perhaps consultants would fit well in that role too.)
