Simon Willison Warns AI Coding Threatens Software Integrity

During a recent episode of the Talking Postgres podcast with Claire Giordano, programmer and blogger Simon Willison challenged the traditional industry consensus that measuring developer productivity by lines of code is meaningless. In the past, Willison noted, a software engineer might produce 50 or 60 lines of debugged, production-level code on an average day, with 200 lines representing an exceptionally good day. With AI coding agents, however, skilled engineers can now generate 1,000 lines of high-quality, tested code daily.

While this represents a massive acceleration, Willison explained that the ultimate bottleneck has shifted from typing speed to human cognitive capacity. Even if an individual developer can churn out code 100 times faster using AI assistants, they lack the mental bandwidth to oversee and understand that volume of software. Consequently, engineering teams remain essential to load-balance this cognitive demand, rather than companies downsizing to single-developer operations.

The ease of generating code also introduces a severe design risk. Willison cited the classic software engineering text "The Mythical Man-Month" to highlight the importance of conceptual integrity, where well-designed software remains cohesive and predictable. With AI agents, developers can implement new features in five minutes, causing software to grow "little weird bumps" in random directions. Giordano compared this chaotic expansion to the Winchester Mystery House, a 140-room mansion built continuously over 40 years without a cohesive master plan.

Historically, the sheer time required to write code acted as a natural filter against unnecessary features. A developer would abandon a marginal idea because it would take a week to build. Because AI agents reduce that barrier to an hour, developers must exercise deliberate discipline to maintain conceptual integrity and prevent their codebases from becoming unmanageable.