In this lesson you’ll cover:
- The shift from a human→document pipeline to the agentic web (goal → decompose → query → verify → act)
- The three operations an agent performs (query, verify, decide) and how each fails differently
- The three agent types — research, recommendation, action — with their goals, criteria, and infrastructure needs
- Why a recommendation agent parses a structured field (e.g. hasMerchantReturnPolicy) rather than reading your prose
- The compute-economics argument: extracting from prose costs ~1,000× more than parsing a structured field
- Building a “non-human ICP” with the four-dimension framework (goal, disqualifiers, ranking signals, verification path)
- Why NLP comprehends but schema defines — and why they’re not interchangeable
- Entity URIs and disambiguation — the 95% NER ceiling vs 100% with a URI, and the cost of ambiguity at scale
- The action layer being standardized now (schema.org, WebMCP, ACP, UCP)
- The three-layer agent-readiness stack (entity → attribute → action), built in order