Monitoring and Reliability › Using Logs for Monitoring

In Chapter 12, we added logging to make the agent's behavior observable. Those logs are also valuable for monitoring. By analyzing logs, you can detect patterns and problems that metrics alone might miss.

For example, you might notice in the logs that a specific tool is failing frequently:

2025-11-10 14:32:15 - assistant - ERROR - Weather API call failed: Connection timeout
2025-11-10 14:35:22 - assistant - ERROR - Weather API call failed: Connection timeout
2025-11-10 14:38:45 - assistant - ERROR - Weather API call failed: Connection timeout

This pattern tells you the weather API is having problems. Even if your overall error rate is low, this specific tool is unreliable right now.

Many deployment platforms provide log aggregation and search tools. You can set up alerts based on log patterns. For example, "notify me if ERROR appears in the logs more than 10 times in an hour." This turns your logs into an active monitoring system.

You can also use logs to track user experience issues that don't show up as errors. For example, if you log when the agent asks for clarification, you might notice users are frequently confused by a certain type of request:

2025-11-10 14:32:15 - assistant - INFO - Requesting clarification for ambiguous query
2025-11-10 14:35:22 - assistant - INFO - Requesting clarification for ambiguous query
2025-11-10 14:38:45 - assistant - INFO - Requesting clarification for ambiguous query

This pattern suggests your agent's understanding could be improved for these cases, even though nothing is technically broken.