An agent that runs for forty turns, calls twelve tools and talks to three models is a distributed system with one process. We spent two years learning to debug those with traces. The GenAI semantic conventions in OpenTelemetry let you do the same thing here, and the first trace I looked at explained a bug that logs had hidden for a month.
The telemetry bill for one service passed its compute bill in March. The instinct is to sample harder and log less. We did the opposite: fewer, wider events, tail-based sampling that keeps every error, and a one-week retention tier for the boring 95 percent. The bill went down by two thirds and the incidents got easier to debug, not harder.