Session
Stop Hand-Tuning Your Agent. Let Your Evals Tune It
Your agent has a failing case. You tweak the prompt, rerun the evals, and the number wobbles. Sometimes it goes up. Sometimes a different case breaks. You’ve been here. You've been in this hand-tuning agent loop.
Good news: you can let your evals do the tuning. An optimizer can search that whole space and score every candidate. Bad news: it can only fix what’s inside the space. A broken tool or a weak model stays broken, no matter how well you rewrite the prompt.
Attend this session and we’ll practice diagnosis. We plant three faults in one agent, read the traces together, and call the culprit before the optimizer runs. Then we find out who was right. Failures stay in. Keep it real.
You’ll leave able to:
- Tell a prompt problem from a tool or model problem
- Read a trace for the dial that failed
- Build a small eval set that catches real failures
- Compare candidates on quality, cost, and prompt length
Stop hand-tuning, start optimizing.
Ron Dagdag
Microsoft MVP / Research Engineering Manager @ Thomson Reuters
Fort Worth, Texas, United States
Links
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top