Spec-Grounded Modernization of Brownfield Software Systems: An Architecture-Aware Framework for AI-Assisted, Traceable, and Safe Evolution

Brownfield modernization is difficult because required behavior, implemented architecture, operational constraints, and business rules are distributed across code, documentation, tickets, data schemas, and tacit knowledge. AI coding agents can accelerate implementation, but without reliable system context and explicit specifications, they can also amplify architectural drift, duplicate capabilities, weaken traceability, and introduce unsafe changes. This article develops a Spec-Grounded Modernization Framework (SGMF) that integrates architecture recovery, specification-driven development, AI-assisted planning, bounded implementation, and evidence-based release assurance. The framework extends a practitioner proposal that combines GitHub Spec Kit with architectural context generation and requirement parsing. It aligns the current official Spec Kit workflow—constitute, specify, clarify, plan, tasks, analyze, implement, and converge—with brownfield-specific services for context recovery, impact analysis, architecture conformance, security validation, and continuous context refresh. The article contributes an eight-stage lifecycle, an artifact and traceability model, a conceptual reference architecture, a Modernization Assurance Index, and an evaluation design for controlled industrial pilots. An experimental billing-system scenario shows how a requirement concerning pending transactions can be connected to affected services, data changes, tests, architecture decisions, release gates, and runtime evidence. The article argues that the unit of control for AI-assisted modernization should not be the prompt or code diff alone, but a versioned evidence chain linking business intent, specification, architecture, implementation, verification, and operational outcomes. The framework provides a foundation for safer, more explainable, and more governable modernization of long-lived enterprise systems.