IQ is a correlation variable that pops out when you measure any group of people's aptitude at a battery of tasks that involve thinking. Basically, if someone's good at one task that involves thinking (say, chess), they're more likely to be good at another task that involves thinking (say, reading comprehension). Apply some Bayesian statistics, and boom, you've concluded that there's some confounding variable (that we call IQ). Then you can start measuring how strongly certain tasks correlate with variable. Turns out abstract pattern recognition very strongly correlated with this unknown variable, so we can use that to predict what someone's IQ likely is.

The point is, due to the very definition of IQ, it's not a narrow metric, and selecting for it does tend to find you individuals who are going to be better than average at most anything. That said, it would seem alarming to me for a job to give me an IQ test instead of cutting out the correlation coefficient and just judging me on the task they're hiring me for.