Build a Prompt Learning Loop – SallyAnn DeLucia & Fuad Ali, Arize

SallyAnn DeLucia, Director of RISE at Arize, and Fuad Ali, Product Manager at Arize, lead a technical workshop on building prompt learning loops—automated systems that continuously optimize AI agent instructions based on real-world performance data rather than manual prompt iteration. Recorded at an AI engineering conference, the session combines conceptual framing with a live code walkthrough participants can follow along with.

The presenters open by diagnosing why agents fail in production: weak or static environment instructions, missing tools, poor context engineering, and the absence of adaptive planning. They then introduce prompt learning as a systematic solution—a continuous loop that collects agent failures, uses that data to generate improved system instructions, and pushes updates to production automatically. A key insight is the concept of two co-evolving loops: one that optimizes the agent’s prompts, and a parallel loop that refines the evaluation criteria used to judge what a good output actually looks like.

The workshop also covers how prompt learning compares to DSPy and genetic algorithm-based prompt optimization approaches, and addresses the organizational challenge of bridging AI engineers (responsible for pipelines and performance) with domain experts (who define what correct behavior means). For teams building production agents and struggling with prompt brittleness or inconsistent reliability, this workshop from the Arize team offers a principled, data-driven framework for moving beyond static system prompts.

📺 Source: AI Engineer · Published January 06, 2026
🏷️ Format: Course Lesson