Context, Intent, and Action: The Semantic Foundation (Ch 11)

Chapter 11 is the Knowledge Camp’s showcase. Reis argues that AI is forcing a discipline humans should have had all along: explicit, machine-readable meaning for every data concept.

Semantics Stack

Controlled vocabularies map synonyms to preferred terms (“client” and “account holder” resolve to “customer”). Thesauri add hierarchical and associative relationships between terms. Taxonomies organize concepts into navigable trees (product catalogs, geography hierarchies). Ontologies formalize relationships and rules so machines can reason over them — concepts, properties, and constraints (every Order must have at least one Product).

Three Metadata Layers

Technical metadata (column types, schemas), business metadata (descriptions, owners, validation status), and semantic metadata (taxonomies and ontologies that connect concepts). An LLM needs all three layers to answer a question like “top-selling products in the Northeast last quarter” without hallucinating.

Context, Intent, and Action

Context engineering is the deliberate construction of the information environment for AI. Intent captures why a query is being asked — same semantic structure produces better results when the system understands the goal. Action-aware data modeling encodes what operations are permissible on data, critical for AI agents that don’t just read but act.

MCP and Semantic Grounding

Reis explicitly names Model Context Protocol (MCP) as a standard for exposing data semantics to LLMs. Semantic grounding ensures organizational definitions override an LLM’s training priors.

RDCO Relevance

Directly relevant to dbt semantic layer consulting and AI-agent data access patterns. The MCP mention validates our positioning around agent-ready data infrastructure. Cross-ref SDG pipeline articles on metrics layer and governance.