Simon Willison argues that measuring software productivity by lines of code makes sense when using AI agents, despite common objections. He states that a senior engineer could previously produce about 200 lines of debugged, production-ready code in a single day, with most days yielding only 50 or 60 lines. If agents allow a developer to output 1,000 lines of similar quality, that represents a very meaningful improvement. Achieving this output requires significant skill, knowledge, and experience. Willison notes he can now perform far more work as a single engineer than before. However, cognitive capacity remains the limiting factor. He can churn out code a hundred times faster, but he cannot stay on top of a hundred times the amount of code. Companies still need teams to load balance this cognitive capacity across multiple engineers.
Conceptual integrity
Willison explains the concept of conceptual integrity from The Mythical Man-Month. Well-designed software possesses integrity where there are no surprises and everything fits together. He argues this is difficult to maintain with coding agents. A developer might have an idea for a feature, run a prompt, and have the feature ready in five minutes. The software then grows little weird bumps in funny different directions. Claire Giordano compared this to the Winchester Mystery House. The house has 140 rooms because the builder was the widow of the inventor of the Winchester rifle. A psychic told her she would be haunted by the ghosts of everyone killed with that rifle unless she kept building the house forever. She added new rooms for 40 years. Willison says this is exactly the problem with coding agents and software. It is very easy to keep adding new rooms because the cost is so much cheaper. The result is a system where conceptual integrity falls apart, making future decisions harder.
Willison concludes that discipline is the solution. Previously, discipline was enforced by the time required. A developer would think a crazy feature would take a week and decide it was not worth justifying. If the feature takes an hour, it is much easier to justify. Willison recorded an episode of the Talking Postgres podcast with Giordano on this subject last week. The discussion covered how AI is changing software development. The transcript has been lightly edited to remove disfluencies.
What it means
Developers must be careful not to let speed erode system quality. The ability to generate code quickly removes the natural friction that previously stopped unnecessary features. Teams need new ways to enforce discipline that do not rely on time costs.
