In a recent Software Architecture Superstream discussion, Neal Ford and Sam Newman argued that the surge of agentic AI coding tools demands a return to modular programming principles first developed in the 1960s. As AI agents generate increasing volumes of code, software architects face challenges in managing and verifying that output.
Newman highlighted that larger context files are actively degrading AI output quality. By breaking problem spaces into smaller, bounded units, architects can solve both the verification bottleneck and the context window problem simultaneously. This approach, which he calls "engineering through specification," allows teams to constrain AI agents within well-defined modules, making their outputs easier to review and trust.
Ford and Newman concluded that the old modular playbook—long a staple of software engineering—is exactly what's needed to impose discipline on the new wave of agentic code generation.