Conceptual integrity and counting lines of code

Blog · Watch/Read

• AI coding agents dramatically increase code output, shifting the primary challenges in software development from production speed to managing cognitive load and preserving architectural coherence.
• Simon Willison argues that lines of code can be a meaningful productivity metric with AI, as skilled engineers can produce thousands of quality LOC daily (vs. 50-200 previously), but this speed makes human cognitive capacity the new limiting factor, requiring teams to manage the increased complexity.
• The ease of generating features with AI agents threatens “conceptual integrity” (from The Mythical Man-Month), potentially creating “Winchester Mystery House”-like software with disjointed additions, necessitating greater discipline from engineers to prevent architectural decay and ensure long-term maintainability.