Hacker News Reader

> First, in the boom prior to around 2015, most software problems were accessible to a smart generalist, but nowadays I don't think that's true. Teams are more specialized.

Do you mean that new areas appeared that require specialization that didn't exist previously, or that areas that require some sort of specialization have comparatively grown? (Or something completely different?)

View on HN

Well, it’s more that the problems in an area like ML or security were solvable if you generally knew how computers work and were smart and good at learning new things. Switching to a new domain took a few months, but ultimately there wasn’t /so much/ you had to learn.

Nowadays, those easy problems are solved. If you want to contribute to an area, you have to learn all the context, read a bunch of papers, it basically takes at least a year. So you can’t quite be a generalist SWE, drop into a random team for three weeks and meaningfully contribute.

Put another way, the relative value of spunk and generalist ability has decreased and the relative value of domain knowledge has increased.

I think this depends on setting ci/cd and good data pipelines are like game changers in research ML projects and generalists definitely can do those it is not that flashy stuff.

Reply on news.ycombinator.com