On messy jobs
Luis Garicano has some good advice about choosing work as generative AI progresses. The core claim is that you should choose "messy work," which involves "a wide bundle of complex tasks." A few notes:
- One obvious caveat is that the work should be messy and need doing. Garicano's example of Base44 (a one-person software company that hit it big) is in fact several examples. For Maor Shlomo, the founder, it was a case of doing a messy job brilliantly well. But it also means that sales, marketing, and many other messy jobs simply did not need doing, except as part of Shlomo's messy job.
- Many people in software seem to be underrating the messiness of software engineering, not only because there are so many human dimensions of the job but because the engineering itself, insofar as it forces you to manage a complex system, is messy. So far, generative AI is great at helping me manage that complexity, but does not seem close to itself managing it. (This is the sort of observation that could make me look very foolish in a couple years, but that's how I see it.)
- More generally: predicting what will be messy, but still need doing, is a lot harder than observing what is messy.
- And even more generally: whenever you get frustrated with AI, and find yourself thinking it's just not getting better at X, that's a sign that we'll need humans to do X for a long time.