3. Development¶

1. Purpose¶

After the validation pilot, it is proven that AI delivers value. Now you build the system production-ready: automated data pipelines, the specification-first pattern, and validation at three levels (syntactic, behavioural, goal-aligned). Behaviour changes are implemented in a controlled manner with pre-defined intent, boundaries, and verification method.

Entry criteria: Gate 2 approved, validation pilot >90%, cost overview approved, team complete.

2. Components¶

• Overview & Objectives — What this phase aims to achieve
• Activities — Data pipelines, RAG/fine-tuning, specification-first method
• Deliverables & Gate 3 — Production-ready system, validation report, test suite
• Specification-first Pattern — Define expected behaviour before you build
• Engineering Patterns — Proven patterns and anti-patterns for AI development

3. Common pitfalls¶

• Building without specification — the specification-first pattern prevents expensive rework
• Not maintaining a golden set — test cases age; keep them current with each iteration
• Changing too much at once — small, bounded behaviour changes are easier to validate
• Not considering buy vs. build — SaaS may be faster; validation requirements remain identical

Next step: After Gate 3 (Production-ready), proceed to Phase 4 — Delivery.