Testing isn’t a one-time event 👇 (Logging): You cannot test what you don’t see. You must log every response your agent generates in production. Tools like braintrustdev help you capture these traces persistently. Evaluate (Scoring): Run your logs against your chosen evaluation method (e.g., have an LLM grade the logs against a rubric) to find exactly where the model fails. Iterate (Improvement): Once you find the failure, you fix the prompt, the skill, or the harness configuration, and then re-run the evaluation to confirm the fix works without breaking something else. Follow for more AI tips and updates 🚀 #aiagents #evals