The economics of enterprise software just inverted. AI-native, spec-driven delivery is producing calendar compression of 97 to 98 percent in early enterprise deployments, with three years of scoped work landing in six weeks and six weeks of work landing in two days. Software production has become nearly free, and most organizations have no framework for what that breaks.
Arthur W. Miller, the enterprise architect behind Zelle and Real-Time Payments infrastructure, has published one. His new book, AI-SDLC and the New Enterprise Architecture Approach: Spec-Driven Development for the Age of Intelligent Agents, delivers a complete, technology-agnostic framework for capturing AI’s speed without losing accountability.
Miller’s central warning comes from a career spent in systems where errors move money: “a 97% faster path to the wrong system is a 97% faster mistake,” he writes.
The book’s argument is that generation was never the hard part for long. Once code becomes cheap, the bottleneck moves to intent and verification, knowing what to build, proving it was built right, and establishing who answers for it. Approaches the industry has embraced under the banner of “vibe coding,” Miller argues, quietly reintroduce the risks disciplined teams spent decades eliminating.
His alternative works from artifacts organizations already produce. Requirements documents, architecture documents, project plans and release notes become living, AI-legible specifications that climb a three-level maturity path: Guided, Contracted, and Generated. Across eighteen chapters, the book walks all eight stages of the lifecycle, from Discovery through Analysis, each with worked examples, AI’s role, tollgate ownership and common failure modes. It closes with a migration playbook for moving a real organization from document-driven to spec-driven one artifact at a time, plus reference appendices covering enterprise integration patterns, agentic AI and MCP orchestration, a metrics starter kit and a troubleshooting guide.
“Every technology cycle, enterprises confuse cheaper production with lower risk. I built payment infrastructure, where that confusion costs real money in real time. This book exists so leaders can take the speed without inheriting the mistakes.” said Miller.
The book takes a deliberately vendor-neutral position. Named products are confined to a single dated appendix, on the reasoning that tools will change before readers finish the book while the framework is built to hold.
Miller’s career spans enterprise architecture for some of the most consequential financial infrastructure in the United States, alongside chief technology officer roles in regulated fintech. The book is written for three audiences: executives deciding whether spec-driven development deserves investment, enterprise and solutions architects responsible for how work flows through systems, and engineering leaders looking to change how their teams work immediately.
AI-SDLC and the New Enterprise Architecture Approach is available now in paperback and additional formats.
