A production agent's operational cost suddenly spikes to 100× normal. Investigation reveals it entered an agentic loop due to a stuck error state. What monitoring should have caught this earlier?
Answer: Both B and C: token rate alerts surface the loop; error rate alerts surface the underlying cause
Both B and C are correct. Token rate monitoring catches the loop (runaway token consumption is the defining financial signature of an agentic loop). Error rate monitoring on tool calls catches the underlying stuck error state that caused the loop. Latency monitoring alone would not distinguish a legitimate long-running task from a loop.