distributed tracing, wheres the sweet spot?

hey everyone, ive been thinking a lot about distributed tracing lately. we're starting to hit that scale where tracking requests across multiple services is becoming a real pain. what are your thoughts on the best place to implement tracing logic? should it be tightly coupled within each service's codebase, or is there a cleaner way, maybe via service meshes or sidecars? im curious about the trade-offs youve encountered.