Part Overview
Part IV takes you from prototype to production. You will learn reference architectures, model routing, retrieval systems, state management, protocols, and security.
Interlock with Previous Part
What this part inherits from Part III:
- Vibe-coded prototypes (Ch 10-14) become the starting point for engineering work
- Multi-agent patterns and tool-based flows reveal production architecture requirements
- The "prototype to production" transition (Ch 14) defines the engineering quality bar
What this part changes retroactively:
- Vibe-coding choices get validated or invalidated: some prototype approaches scale, others must be rebuilt
- The Human-AI Product Stack (Part I) gets instantiated in concrete architectural decisions
Artifacts that now need updating:
- The build/buy/bake framework (Part I) now has architectural implementation implications
- Discovery insights (Part II) may require re-evaluation when engineering constraints emerge
Chapters in This Part
Common AI product architectures from simple to complex.
Router strategies, model selection, capability-based routing.
Vector DBs, Graph RAG, Self-Correcting RAG.
Memory patterns, state management, orchestration frameworks.
MCP, A2A, agent-to-agent communication.
Prompt injection, data privacy, adversarial robustness.
Bridge Notes
Earlier artifacts updated by this part:
- Part I, Ch 1: Build/buy/bake decisions now have architectural implementation paths
- Part I, Ch 3: Human-AI Stack becomes concrete in reference architectures (Ch 15)
- Part III, Ch 14: The prototype-to-production transition gets systematic treatment
Later chapters this part prepares for:
- Part V, Ch 21-25: All engineering decisions require reliability verification through evals
- Part V, Ch 22-23: Observability and guardrails are built on top of these architectures
- Part VI, Ch 28-29: Team topologies reflect the engineering patterns established here