efficient-ai-codegen-approach-comparison.md

Comparing the Two Approaches

	Pure Inference	Rules-Augmented Inference
Time-to-first-output	Fast	Slow at first
Output variance	High — every run produces a different design	Low — designs converge to the rules
Team-wide reproducibility & scalability	None — best prompter wins; results don't generalize	Yes — sharpen the rules, lift everyone's output
Review burden	Heavy — review at the design level	Light — just check rule conformance
Stability across model switches	Fragile — re-tune prompts	Stable — rules outlive models