See how to go from unstructured documents in Box to structured, searchable metadata using Box AI and a lightweight vibecoding workflow. In this video, we build a small, CLI-first Python app that extracts fields from real documents and writes them back as Box metadata — all guided by a simple agents.md spec. We start by anchoring on a metadata schema, use agents.md as guardrails to generate the app in Cursor, and then run the workflow across a folder of files. The result is structured data written directly onto your Box content, ready for search, automation, and downstream workflows. Key takeaways: → Use Box AI Extract Structured to pull consistent fields from real documents → Anchor extraction on a shared Box metadata template (schema) → Use agents.md as guardrails to vibe-code a complete, runnable CLI → Process single files or entire folders with the same extraction logic → Write extracted fields back as Box metadata to power search and workflows This walkthrough focuses on the core pattern: schema → working app → populated metadata, using your existing Box content — without moving files out of Box or building a heavy system around it.

 133      1