whens it time to seriously consider a service mesh?

hey everyone, ive been thinking a lot lately about microservices and how much work it is to keep them running. we have a fair amount of services now, and keeping track of retries, circuit breakers, and observability for all of them is becoming a real headache. i know service meshes like Istio and Linkerd say they fix these problems by handling it all in the background. but, im wondering if adding a service mesh just makes things too complicated, more than the distributed systems are worth in the first place. are there certain team sizes, numbers of services, or ways they talk to each other where it really helps? or is it more like, 'wait till it hurts' kind of deal? would love to hear what you guys think and what youve gone through.