Demand-Driven Context: A Methodology for Coherent Knowledge Bases Through Agent Failure

An agent's failed attempt can help identify which organizational knowledge needs documenting. This workshop proposes giving agents real problems, examining where missing information blocks them, and adding that knowledge to a shared context base.

The published description calls for growing the collection problem by problem and storing it in Markdown, with agents of different roles using the same shared material. The intended benefit is a closer connection between documentation effort and questions that arise in actual work: each addition addresses a demonstrated information gap.