Spec-Driven Development as the Governance Backbone of the AI-Driven Development Lifecycle (AI-DLC)

This thesis has argued that Spec-Driven Development is the necessary governance backbone of the AI-Driven Development Lifecycle — not a complementary practice but the structural prerequisite that makes AI-DLC viable at enterprise scale. The argument rests on four interlocking findings from the source literature, each of which addresses a different layer of the governance challenge. Piskala (2026) establishes that specification quality is the primary determinant of AI-assisted development quality, and provides a three-level rigor framework that maps SDD to context. Rehan (2026) provides the compilation mechanism — TDAD — that transforms informal specifications into verifiable behavioral contracts for AI agents, with mutation-tested robustness metrics that enable objective quality gates. Vilas Boas et al. (2026) provide empirical evidence that specification quality, not model capability, is the binding constraint on AI-DLC performance, and that when specification quality is high, the throughput and cost efficiency gains are substantial. And Augé et al. (2026) provide the Operations framework — MAS-Lab — that extends specification governance from individual agents to multi-agent systems operating in production, where behavioral compliance must be continuous, auditable, and resilient to specification evolution. Together, these papers describe an emerging discipline: specification-governed agentic engineering. This discipline treats the specification not as documentation but as infrastructure — the machine-readable, version-controlled, compliance-auditable artifact that governs AI agent behavior at every phase of the development lifecycle. AI-DLC provides the operating model; SDD provides the governance layer that makes that operating model trustworthy. The practical implication for organizations adopting AI-DLC is straightforward: the investment that matters most is not in the AI agents, the orchestration platform, or the Bolt compression. It is in the specification corpus, the engineers capable of producing it, and the institutional processes that ensure it remains accurate as systems evolve. Organizations that invest in specification infrastructure will find that AI-DLC delivers on its velocity and efficiency promises. Those that do not will find that agents are capable of producing code faster than they are capable of producing trustworthy behavior. The leader within the AI-DLC framework, this thesis argues, is not the AI agent. It is the specification — the precise, versioned, machine-enforceable expression of human intent that makes everything else governable.

[1] Augé, J., Carofiglio, G., Grassi, G., & Samain, J. (2026). MAS-Lab: A specification-driven validation framework for reliable multi-agent systems. arXiv preprint arXiv:2606.30546.   [2] Augment Code. (2026). What is AIDLC? The AI development lifecycle explained. https://www.augmentcode.com/guides/what-is-aidlc-ai   [3] DORA Research Team. (2025). State of DevOps Report 2025. Google Cloud.   [4] Piskala, D. B. (2026). Spec-driven development: From code to contract in the age of AI coding assistants. arXiv preprint arXiv:2602.00180.   [5] Rehan, T. (2026). Test-driven AI agent definition (TDAD): Compiling tool-using agents from behavioral specifications. arXiv preprint arXiv:2603.08806.   [6] Taylor, C. (2026, June). Everything you need to know about AI-DLC. The Pragmatic Engineer via Substack. https://carlytaylor.substack.com/p/everything-you-need-to-know-about   [7] Vilas Boas, M., Pinto, G., Monteiro, E. R., Carida, V. F., & Ribeiro, D. (2026). One developer is all you need: A case study of an AI-augmented one-person squad in a brownfield enterprise. arXiv preprint arXiv:2605.18461.   [8] AWS. (2026). Kiro: AI-Driven Development Lifecycle (AI-DLC). Amazon Web Services Developer Documentation.   [9] Kitchenham, B., & Charters, S. (2007). Guidelines for performing systematic literature reviews in software engineering. EBSE Technical Report EBSE-2007-01.