Uploaded July 2026 | Updated November 2026, 3 hours ago
An agent that writes its own SQL will invent tables that don't exist. The fix could be as simple as including the database's schema in the system prompt. This video shows how to use PXI to run an experiment to find out!
Watch this video to learn:
• When to use a code evaluator instead of an LLM judge (hint: if you can verify it programmatically, do)
• Why both prompts must run on the same model — you're testing the prompt, not the model
• The first rule of failing evals: when an evaluator fails almost everything, suspect the evaluator before the model
#ArizeAI #LLMEvaluation #AgentObservability #Phoenix #PromptEngineering
🎵 Music by Phoenix engineer Tony Powell — check out more of his music at soundcloud.com/cephalization/sets/some-sounds
🔗 Try Phoenix: arize.com/phoenix/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=rnabors
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1
An agent that writes its own SQL will invent tables that don't exist. The fix could be as simple as including the database's schema in the system prompt. This video shows how to use PXI to run an experiment to find out!
Watch this video to learn:
• When to use a code evaluator instead of an LLM judge (hint: if you can verify it programmatically, do)
• Why both prompts must run on the same model — you're testing the prompt, not the model
• The first rule of failing evals: when an evaluator fails almost everything, suspect the evaluator before the model
#ArizeAI #LLMEvaluation #AgentObservability #Phoenix #PromptEngineering
🎵 Music by Phoenix engineer Tony Powell — check out more of his music at soundcloud.com/cephalization/sets/some-sounds
🔗 Try Phoenix: arize.com/phoenix/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=rnabors
🔔 Subscribe for weekly content on LLMs, agents, and evaluation: youtube.com/@arizeai?sub_confirmation=1










