Conceptual integrity and coding agents with Simon Willison

Simon Willison appeared on the Talking Postgres podcast with Claire Giordano to discuss how artificial intelligence is changing software development. During the conversation, he argued that lines of code can actually serve as a meaningful indicator of productivity when using coding agents, provided the output remains high quality, tested, and maintainable. Because senior engineers possess the necessary skill and experience, they can leverage these tools to accomplish significantly more work than they could previously.

However, Willison explained that human cognitive capacity creates a hard limit on individual output. While an engineer can churn out code at a much faster rate, one person lacks the mental capacity to monitor one hundred times the usual amount of code. This reality means companies still need teams of engineers to load balance that cognitive capacity and maintain operational resilience.

The discussion also highlighted the threat AI poses to conceptual integrity, a software design principle where well-designed systems fit together logically without surprises. Because coding agents allow developers to generate features in minutes, software tends to grow random additions. Giordano compared this phenomenon to the Winchester Mystery House, a mansion continuously expanded for decades without a master plan.

Ultimately, Willison emphasized that discipline is now the primary challenge in software creation. In the past, the time required to build a feature naturally forced developers to abandon impractical ideas. When coding agents reduce that timeframe to an hour, justifying and adding erratic features becomes dangerously easy, causing the overall conceptual integrity of the software to fall apart.