Same observation here.

I remember a time, not so long ago, when serious people were advocating JITed languages (Java, C#) for performance workload because the JIT was theoretically able to produce better code for the platform given it had full knowledge of the architecture and execution context.

For (long running) data stream workloads, JIT does have an auto-profiler benefit. It doesn't preclude writing code well.

For long running workloads with consistent usage patterns it has a benefit.

It is not merely enough to be long running, that is a requirement, but not enough by itself.

Are there any studies measuring it on real-world workloads?