Best tools for logging & monitoring AI agent usage?

For those working with AI agents, what tools are you using to log and monitor usage?

I’m looking for something that can help:

1. Track which LLM calls or system prompts are failing

2. Identify where the agent logic went wrong

3. Provide analytics for improving the agent’s performance

Do you rely on existing tools (would love names + pros/cons), or do you build in-house logging/monitoring solutions?