The blockchain works great for currency systems, and to an extent, smart contracts (provided they are not needlessly complex, or else you risk losing all stored funds to a bug/hack/unintentional loophole).
This can be summed up by saying that blockchains are only useful for abstract information that benefits from being uncensorable, of which money and contracts are examples. Other examples I would include would be ID systems and data timestamping (Satoshi proved this in an inverted way by adding the famous "The Times 3.1.2009, Chancellor approves 2nd bailout" text to the genesis block, any data referenced in a blockchain can be verifiably proven to have existed at least after a verifiable time/date).
But when the 21m BTC supply was mined, miners will surely shift to a more mining-profitable coin that can lead to a lower Hashing power.
One of the side-effects of the rise of ASICs has been that Bitcoin is sharing its PoW scheme with only a handful of coins, signifcantly reducing the number of possible targets to alternatively point hashing power at. Nonetheless it is of course impossible to predict how the world and cryptocurrencies will look like a 100 years from now, given the latter still exist.
And
this is one facet of a very significant reason to not switch POW algorithms. Thank you.
Don't you mean this is the reason not to be holding any coin using the PoW algo to which Bitcoin's PoW algo is switched?