Things in the "Context Plane" – By Shagility (agiledata.substack.com)

Shagility proposes a “Context Plane” as a single, organized layer of metadata and models to power AI agents within a broader "AI Data Stack." The post is a thoughtful, iterative blueprint rather than a finished spec: it argues we need a shared repository of business definitions, data models, quality contracts and operational logic so agents can reason, act, and be audited consistently. This matters to the AI/ML community because it bridges long-standing data engineering patterns (glossaries, logical/physical models, contracts, lineage) with agent-centric needs like reliable retrieval, reproducible transformations, and governance — reducing ambiguity in prompts and enabling safer, verifiable automation. Technically the Context Plane covers components such as Business Glossary and Core Business Concepts (canonical entities/aliases), Core Business Processes (relationships and life cycles), Conceptual/Logical/Physical Data Models, Data Contracts, Data Dictionary and Data Profiles, Transformation Logic, raw Facts, derived Measures/Metrics, Business Questions, Actions, and Information Apps. Shagility highlights overlaps between components and the need for platform-independent representations plus system-specific implementations and SLAs. The implication: a unified context layer would let retrieval-augmented models reference authoritative schema, contracts and metrics, support traceable model outputs, automate decision workflows, and enforce data quality — effectively turning organizational domain knowledge into machine-usable context for agents.