YAML doesn't know why: building the business context your agents are missing

In agentic data workflows, most failures come down to lack of context. dbt's MCP and skills made agents genuinely capable, YAML files and the semantic layer gave them fluency in your data model, but sessions still depend on constant prompting, and agents keep getting things wrong in ways that look right. The context they are missing is organizational: business rules that evolve, campaign decisions that shift acquisition patterns, product changes that alter user behavior. Knowledge that lives in the engineer's spec, the meeting notes, the Slack thread, but never reaches the agent session that needs it. 

The talk covers an attempt to capture that context through a living knowledge base distilled from organizational artifacts, what went wrong, and what it takes to build a context layer that multiple agents can use and contribute to, that stays structured, complete, and current as the business evolves, making context compound in your favor instead of against you.