YAML doesn't know why: building the business context your agents are missing
How a data scientist built the context layer his AI agents were missing, capturing the organizational knowledge a dbt project alone can't hold.
Overview
Agents need organizational knowledge beyond dbt definitions—business rules, campaign decisions, product changes living in specs, meeting notes, and Slack threads. Covers building a living knowledge base distilled from organizational artifacts, what failed, and what's needed for a sustainable context layer agents can read and contribute to.
What you'll take away
- 01
Technical capability was never the real gap for AI agents; the gap was organizational knowledge, the kind that lives in Slack threads and PR explanations and was never captured as a durable record.
- 02
After two failed approaches, one giant memory file, then plain RAG, the working version extracts atomic facts and decisions with temporal validity, so a superseded decision stays retrievable without polluting current search.
- 03
In a live demo, an agent traced a revenue spike from the dbt project to a known backend bug, named the engineer who fixed it, and logged the full investigation back in for future reuse.
Session speaker
Pedro Heyerdahl
Kilo Code