I'm not sure we need a whole new programming paradigm and languages (with the hiccup that they'll require bindings to other FFI-libs that do not have as much guarantees). If you have a rich domain and want to benefit from performance boost using "fenced" mutations. Haskell has linear-types, software-transactional memory, and the ST monad. Which are three ways to constrain side-effects over references. If you need tight control of "the CPU as a state-machine", for instance, for IO-heavy services, then you have Rust, ADA and other C-targeting DSLs