Reading one paper a week is 520 papers a decade, and there's your "Oh wait, I've seen a solution to this problem before..." superpower as a senior dev. Not your only one, but one that's easy to acquire.
I would often do this in grad school:
* Go search for papers that broadly had to do with some structure or other mathematical gadget I was interested in at the time,
* Read the abstracts of those papers to find the ones that looked most interesting,
* Take the most interesting papers, and read the statements of the theorems,
* Finally, devote a little more attention to those papers that had interesting theorems that seemed to fall within the domain of what I was working on.
I did this with math papers, but there's no particular reason you can't generalize this to other fields. CS in particular can use almost the exact same methods. For less mathematical fields, you'd need to make some substitutions, such as "section headers and key topic sentences" for "statements of theorems," but you can make it work there, too.
Doing this, in a decade, you end up reading 1560 abstracts, which is probably more useful in terms of "Oh, wait, I've seen this before" type insights than reading 520 entire papers.
Knowing the shape and direction of research into big topics is just so critical to choosing the right topics yourself.
I was an on-and-off-again ACM member for 40 years, and one of the better publications was ACM Computing Surveys: https://dl.acm.org/journal/csur -- even older issues are pretty high value, and there are tons of references to follow.
[edit: update Usenix link to something much more current]
Sometimes I will just randomly see what has been posted to one of the areas I am interested in as well on arxiv.
Then often I end up just googling references in the paper to other papers.