I think the gradient descent / machine learning metaphor is a good one, because it illustrate blind local optimization given a set of value functions. In other

I think the gradient descent / machine learning metaphor is a good one, because it illustrate blind local optimization given a set of value functions. In other words: exactly the behavior that organizations over a certain size tend to exhibit (or put another way, the middle-managers-revert-to-the-consensus-value-approximation decision metric). As you note, the problem with tech is that for truly profitable endeavors, there is always a technical barrier that first much be overcome. And from first-glance, that seems like an impossible / overly-expensive burden (especially given the unknown reward behind it).