ZK STARKs are very powerful and will certainly be useful for off-chain bitcoin smart contracts, rollups, etc, but STARKs are very complicated and inefficient.

An average bitcoin dev could probably implement almost any hash-based signature algorithm in a day or two. Contrastingly, implementing a STARK prover/verifier seems to demand teams of people with years of expert knowledge in the domain. Even established STARK software like Winterfell suffer from awful usability/ergonomics. Read their README and examples, and you'll see what I mean.

I don't think we should build on-chain bitcoin security standards based on such things without a simple easy-to-use library to depend on, like libsecp256k1 is today. Perhaps there will be a more stable and usable STARK library in the future but so far I haven't found any. The closest is RISC0, but AFAICT it's bugged for secp256k1 usage, and they're not fixing it.