Intel / video

Let Your Agent Cook: Using Skills to Evaluate and Improve Your App — Ankur Duggal, Arize AI

Why it matters

Shows how to turn vague quality complaints into trace-level evidence a coding agent can act on, with experiments to validate each proposed fix.

Key takeaways · AI-distilled

Arize's Ankur Duggal connects a coding to production traces and evaluation results through Arize's CLI and skills, so it can pull spans locally, diagnose failures and propose code changes to a flawed financial app.
Trace analysis surfaced numeric problems, fabricated answers, redundant searches and silently worked-around failures. In an earlier fix for non-US stock queries, changing the country alone failed until the underlying tool was also updated.
Duggal pairs dataset, evaluator and experiment skills so fixes are tested rather than accepted as unverified edits, and still recommends human review of proposed changes, since a coding agent can spend its effort on linting instead of the actual defect.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.