1.4. AWS GenAI Architecture Principles
💡 First Principle: Building GenAI applications on AWS isn't just software engineering — it's applying structured architectural judgment to a class of systems with unique failure modes (hallucination, semantic drift, non-determinism) and unique cost structures (token-based pricing, context window economics).
The AWS Well-Architected Framework provides the evaluation lens for these trade-offs. The Professional exam specifically tests whether you can apply architectural principles to GenAI trade-off scenarios — not just whether you know which service to call.
⚠️ Think of the Well-Architected Framework as a structured checklist that forces you to ask the right questions about your architecture before a production incident does. For GenAI workloads, each of the six pillars has AI-specific failure modes that traditional architectures don't face:
| Pillar | Traditional concern | GenAI-specific concern |
|---|---|---|
| Operational Excellence | Deployment automation | Prompt versioning, model lifecycle management |
| Security | IAM, encryption | Prompt injection, guardrails, PII in context |
| Reliability | Multi-AZ, retry logic | FM throttling, model deprecation, hallucination |
| Performance | Latency, throughput | Token limits, retrieval quality, cold starts |
| Cost Optimization | Right-sizing instances | Token economics, caching, model tier selection |
| Sustainability | Energy efficiency | Inference compute footprint, model size trade-offs |
Common Misconception: The Well-Architected GenAI Lens adds new pillars specific to AI (like "Accuracy" or "Fairness"). It does not — it applies the same six pillars to GenAI workloads and adds AI-specific best practices within each.
