Simon Willison: LOC metrics matter with agents, but conceptual integrity is harder to keep

Simon Willison · rss · 2026-08-20

Simon Willison discusses AI's impact on software development, arguing that lines of code is a valid productivity metric if agents produce high-quality, debugged output. However, the new bottleneck is cognitive capacity; engineers need teams to manage the increased code volume. He also warns that cheap generation via agents threatens "conceptual integrity," comparing the result to the Winchester Mystery House—software grows in chaotic directions without the discipline enforced by previous time costs.

Original post →

More from coding & agent

coding & agent channel →