Our AI tests passed but users found missing details: when evals don't match reality
Plastic_Secret1303 · reddit · 2026-10-26
A team built an AI tool that summarizes support chats. Internal tests looked good, but real users reported summaries missing important details. Root cause: their test examples didn't match real conversation patterns. They're now adding more real-world examples and having humans review selected outputs. The post asks how others ensure AI evals reflect actual user needs—a useful cautionary case of offline evals diverging from production.
More from coding & agent
- SF 'Code Engineering' event picks up where 1,000-person harness engineering meetup left off — LangChain · 2026-10-27
- Prompt turns Claude Code into an iPhone backup forensics tool exposing 30-day 'deleted' photo copies — alex_verem · 2026-10-27
- Prompt turns Claude Code into a forensics tool finding deleted iPhone photos in backups — alex_verem · 2026-10-27
- Phones won't be mobile IDEs but control planes for agent fleets, argues Yuchen Jin — Yuchenj_UW · 2026-10-27
- FrontierCS 2 launches on Kaggle: 48 tasks make agents beat published CS papers — MeganRisdal · 2026-10-27
- Stop remaking old games with AI: a call for more ambitious agentic software — rschu · 2026-10-27