Spec-Driven AI Coding
The AI-Coding Revolution: Spec-Driven Development is a high-fit source because it treats AI coding as a discipline problem instead of a hype story. Alex di Gioia and Michele Brissoni present AI as an amplifier: a team with strong requirements, tests, review habits, and product thinking may get leverage, while a team with weak practice may get faster confusion and technical debt. Their proposed route is not one giant prompt, but staged agent work: discussion, design, distillation, implementation, artifact contracts, adversarial review, acceptance tests, unit tests, and mutation testing.
The strongest Spiralist relevance is the attempt to keep delegated software craft legible. In this framing, the agent is not a mystical programmer and not a mere autocomplete tool. It is a worker-like process that needs role boundaries, context hygiene, measurable outputs, review gates, and a human who remains responsible for direction. That belongs beside AI Coding Agents, Vibe Coding, Context Windows and Context Engineering, AI in Employment, Agent Tool Permission Protocol, Agent Audit and Incident Review, and The Erosion of Apprenticeship.
External evidence supports the caution while limiting the talk's stronger claims. METR's 2025 randomized study found that experienced open-source developers working on familiar repositories completed tasks more slowly with early-2025 AI tools, even though they expected and perceived speedups. DORA's 2024 State of DevOps report found more positive survey associations between AI adoption and developer flow, productivity, satisfaction, code quality, and documentation quality. The difference matters: AI may help some workflows, teams, and task types while hurting others, especially when review burden, context drift, weak tests, or unfamiliar generated code enter the system. OWASP's LLM Top 10 and NIST's AI Risk Management Framework give the broader security and risk frame for systems that use models inside real software workflows.
Uncertainty should stay visible. This is a practitioner talk from a software-craftsmanship community, not a neutral evaluation of the presenters' framework. The demo and argument are useful, but they do not prove that the same method transfers across languages, organizations, legacy systems, regulated environments, novice users, or high-security codebases. Treat the video as a serious craft proposal for making AI coding more inspectable, not as proof that spec-driven agentic development has solved software reliability, apprenticeship, security, or accountability.