Highlights
- Software is a language artifact before it is a technical one — Hewitt's core claim is that systems fail at the level of concepts and naming, not code, so the architect's first job is designing the semantic field the software lives in. This is the philosophical grounding for MDDE: if meaning comes first, the model — not the platform — is the real system.
- The architect's primary deliverable is a shared conceptual vocabulary — he argues that clear, agreed-upon concepts do more for a system than any framework choice. That is exactly the concept-library move: a curated set of precisely named ideas that both humans and AI can build on without re-negotiating meaning every time.
- Design the concepts deliberately, then let implementation follow — the book pushes design upstream, treating code as the downstream expression of a well-formed idea. MDDE operationalizes this: capture the semantics as metadata-as-code and generate the repetitive implementation, instead of hand-crafting meaning into every pipeline.
- Ambiguity is the most expensive defect class — Hewitt shows how vague or overloaded terms silently propagate into architecture, teams, and data. In SBM terms this is why rules are the forgotten third pillar: explicit definitions and validation rules are what stop AI (and humans) from confidently guessing over fuzzy semantics.
- Technology choices decay; conceptual clarity compounds — the book treats frameworks and platforms as transient carriers of a design that should outlive them. Platforms churn, capabilities endure: a well-designed semantic model survives every re-platforming, which is precisely why "the model outlives the tool."
- Thinking and writing are architectural acts — Hewitt draws on philosophy and design thinking to argue that an architect who cannot articulate a concept in prose has not finished designing it. This maps to the second-brain practice of atomic, self-contained units: one concept, one artifact, precise enough to be composed and queried later.
Highlights on this page are generated with the help of AI.