April 22, 2026 Melissa Pan — UC Berkeley
LLM-based agents already operate in production across many industries, yet we lack an understanding of what technical methods make deployments successful. We present the first systematic study of Measuring Agents in Production, MAP, using first-hand data from agent developers. We conducted 20 case studies via in-depth interviews and surveyed 306 practitioners across 26 domains. We investigate why organizations build agents, how they build them, how they evaluate them, and their top development challenges. Our study finds that production agents are built using simple, controllable approaches: 68% execute at most 10 steps before human intervention, 70% rely on prompting off-the-shelf models instead of weight tuning, and 74% depend primarily on human evaluation. Reliability (consistent correct behavior over time) remains the top development challenge, which practitioners currently address through systems-level design. MAP documents the current state of production agents, providing the research community with visibility into deployment realities and under-explored research avenues.
Melissa Zhiyang Pan is a second year Ph.D. student in Computer Science at UC Berkeley, advised by Prof. Matei Zaharia. Her research interests lie in building efficient and sustainable computing systems for emerging machine learning and data-intensive tasks (eg: agent systems) at a large scale, and how to use AI to support faster systems research. She is currently investigating energy-efficient and reliable agentic/compound AI systems through resource scheduling and cross-stack optimization. Melissa is also Amazon AI Fellow, and Laude AI Resident.