There's complex technical work, and there's complex social work. The programmers who use a speech recognition API to solve a human problem - say, hands-free in-car navigation - are doing work that's every bit as difficult as building the speech recognition API in the first place. And they're often more highly compensated for it, if they succeed - it's just the risk structure of trying to match known technology to unknown problems is a lot more vague than trying to match unknown technology to known problems.