In the latest edition of the excellent IEEE Software magazine, the non less excellent Philippe Kruchten has conjectured about the existence of a five years half-life for software engineering ideas. That would mean that 50% of our current fads would be gone within five years.
Though not (yet) backed by any hard numbers, his experience (as mine and yours) suggest that this is most probably true.
All in all, the sheer idea of such an half-life decay pattern helps letting obsolete knowledge go without regret. It is fine to forget everything about EJBs and code with POJOs. JavaBeans are evil: you can come back to pure OO development. You can bury your loved-hated-O/R-mapper skillset in favor of writing efficient SQL. It is OK to drop the WS-Death* for REST (this one is purposefully provocative, if a WS-nazi reads this blog I am dead).
Unlearning things that were so incredibly painful to learn may feel like a waste but it is not. This half-life decay is in fact great news because it translates into less knowledge overload and more focus on what matters... or, unfortunately, the adoption of new short-lived fads! And the latter is the scariest part. How much about what we are currently learning will survive more than five years?
Nothing platform or even language specific, I would say. But sound software engineering practices will remain: building elevated personal and professional standards, treasuring clean code, challenging our love for hype with the need for pragmatism...
This is where our today's investment must be: in the things that last.