Plan an agent's share of observability work

Your Python microservice is a Flask API with a Postgres database, and it calls 2 other services. It has almost no observability: a few print statements, no metrics and no traces. You want structured logging, metrics and distributed tracing, and you want an AI agent to do most of the instrumentation.

How do you break the work up, what do you keep and what do you delegate, and where do the human checkpoints go?