Software Design in the Age of AI

Details

The article challenges the narrative that AI coding tools will make software engineers obsolete by separating coding from software engineering. It identifies five core software development activities — requirements gathering, design, coding, verification, and support — and argues that AI tools address primarily the coding step, leaving the other four intact and arguably more important. The central claim is that AI coding effectiveness depends heavily on design quality. For large systems, poorly structured codebases give AI tools insufficient context to generate reliable extensions or fixes. Clear module boundaries, well-defined interfaces, and separation of concerns make codebases more AI-legible — good design amplifies what AI can do rather than being superseded by it. The piece introduces the idea of treating AI coding tools as a stakeholder in software design decisions. Just as systems must be designed with human operators, maintainers, and users in mind, they should now be designed for AI to operate effectively within them — modular, well-documented, with clean APIs. The article is opinion and analysis rather than empirical research — there are no case studies, metrics, or controlled comparisons to support the claims. That said, the argument aligns with what senior engineers working at scale with AI tools are reporting in practice. The piece is most valuable as a framing tool for engineering leaders thinking about where to invest as automation spreads.