Chapter 0.5 — Enterprise trade-offs

🎯 Chapter objective

Map the main trade-offs in enterprise AI systems and offer a heuristic for decision-making.

🧠 Core concept

Every architectural decision involves a compromise between:

• quality;
• cost;
• latency;
• security;
• governance;
• privacy;
• maintenance;
• scalability;
• auditability.

No decision maximizes every dimension. The architect's job is to make the trade-off explicit and revisitable.

🏗️ How this shows up in production

More quality -> bigger model -> more cost -> more latency -> less scale.
More security -> more validation -> more latency -> more cost.
More privacy -> less data -> less quality or fewer features.

🧪 Simple heuristic

For each decision, write down:

1. Main objective (1 sentence).
2. 3 non-negotiable constraints (e.g., p95 latency ≤ 2s, PII never leaves the tenant, cost per task ≤ US$ 0.10).
3. 2-3 accepted trade-offs.
4. 1 contraindication (when this decision should be reverted).

📌 Checklist

• [ ] Are trade-offs written in an ADR (Architecture Decision Record)?
• [ ] Are reversal criteria defined?
• [ ] Is the ADR reviewed periodically?

---