Why Domain-Specific Languages and good abstractions make LLM code generation reliable and reviewable.