You don't need to do this forever. Usually you can just focus on a stack and related technologies and do well. Once you have experience and a clearly defined need, ad-hoc research is good enough.
The thesis of the linked article is that you do. For the most part, it matches my experience. I'm 30 years in and already wondering how useful the Hadoop/Spark/Scala stuff I spent the last few years mastering is going to be in the next 5 years.
* Scala introduced us old Java hands to a whole different world of modern languages. If you know it you get Kotlin or the latest Java changes for free (probably TypeScript-like other ecosystems too).
* Spark introduced a generation of backend developers to distributed query engine technology (my generation is unlikely to delve into postgress codebase in comparison) and made ETL trivial in real life systems
Hadoop clearly died in the last few years or so (outside of EMR where it's mostly invisible anyway). But it's a perfect example of a complete technology lifecycle - it had a good run for a decade starting around 2010 which in our line of business is incredibly long time. Not to mention how many things about distributed systems people like me learned from that stack over time.