AI Coding Agents: Redefining Design and Coding in Software Engineering

The rise of AI coding agents capable of generating syntactically valid and functional code, alongside design reasoning, has sparked debate among software engineers: should developers pivot solely to design, or does AI render both skills obsolete? Emily B., a software developer and creator, asserts that both coding and design remain indispensable. Citing Jack W. Reeves’ 1990s insight, she underscores that source code itself constitutes the software’s design, with compilation akin to manufacturing. This perspective implies that writing, testing, and debugging code are inherent parts of the design process. Furthermore, B. draws on computer scientist Peter Naur’s 1985 theory of “programming as theory building,” where the true design resides not merely in the repository but within the developer’s mental model—the “theory”—enabling confident explanation, query, and crucially, correct modification of the software.

This internal “theory” is critical given Kent Beck’s “Constantine’s equivalence,” which posits that software cost is primarily driven by the cost of change, directly linked to coupling. Effective design, therefore, aims to reduce coupling and simplify the mental model required for understanding, facilitating easier and safer modifications. While AI tools can generate code from high-level prompts, B. cautions that the underlying “theory” of that code doesn’t automatically transfer to the human. Developers must still engage with, critique, and understand AI-generated code to modify it safely, a process akin to comprehending human-written code. Although some engineers, like Llewellyn Falco, are exploring “programming in prompts” to elevate abstraction, treating AI as a compiler with the “theory” embedded in natural language instructions, B. expresses skepticism about the scalability of this approach for complex systems. Ultimately, B. advises aspiring engineers to initially focus on traditional coding to cultivate robust design skills, emphasizing their enduring value over rapidly evolving AI prompting techniques.