AI & Agility: Spec-Driven Development or the return of the V-model and how to survive it
summary
Presents problems with Spec-Driven Development (SDD) based on large language models—multi-agent, RAG, MCP setups—that assume perfect context and produce demo-ready but brittle production systems and heavy comprehension debt. Argues SDD revives a V-model unsuited to real-system complexity and limited human/LLM cognition, then proposes Empirical Product Development (EPD): break work into atomic tasks, enforce a feedback wall (tests, performance, system coherence, business validation), and iterate progressively to integrate AI into agile cycles while avoiding black-box failures and preserving quality.
AI-written summary · about
