Earlier quoted context omitted.
By specializing in a domain, you live and die by it. It usually happens anyways, but it isn't all upside.
Domain as in 'aerospace', 'construction', 'finance', 'computational geometry'... If any of those die software is kinda pointless anyway.
We are seeing domain specialization at work now, though not many really saw the Deep learning coming so most of those people aren't senior yet (and the ones that did are super successful in their domains; e.g. A friend of mine has done really really well in applying DNN to speech recognition very early). In research, we talk about areas, but even these are not very well defined and people aren't really interchangeable.