On last week's episode of the Talking Postgres podcast, Claire Giordano and a participant identified as "Simon" discussed "How AI is changing software development." Below are highlighted excerpts from a lightly edited transcript (edits aimed to remove disfluencies).
Lines of code as a metric
The conversation presents an argument that in some situations it still makes sense to consider lines of code (LOC) as a productivity signal when using coding agents — with emphasis on "in some situations."
Simon argues that in the pre-AI era a software engineer could produce a few hundred lines of production-ready, debugged code in a day — and 200 lines was an exceptionally good day. Most days yielded 50–60 lines. If agents allow you to produce a thousand lines of debugged code per day, that represents a meaningful improvement — provided the code quality is the same (maintainable, tested, etc.).
He stresses that reaching and maintaining that quality with agents still requires significant skill, knowledge, and experience: the kind of work senior engineers provide.
Cognitive capacity and the need for teams
Simon raises the point that if a single engineer can produce far more code with agents, why maintain multiple engineers? Beyond the obvious bus-factor argument (a one-person team is poorly designed), the limiting factor becomes cognitive capacity. A person might churn out code a hundred times faster, but they do not have the cognitive capacity to oversee a hundred times more code. Teams remain necessary to distribute that cognitive load.
Conceptual integrity and the Winchester Mystery House analogy
Another central point is conceptual integrity. Simon cites the notion from The Mythical Man-Month: well-designed software has an integrity to it — no surprises, the domain is covered appropriately, and everything fits together coherently.
That integrity becomes much harder to preserve when coding agents let you conceive a feature, run a prompt, and have the feature in five minutes. Software can then grow "little weird bumps in funny different directions." Claire Giordano compared this to the Winchester Mystery House.
The analogy: the Winchester Mystery House was continuously expanded for decades, driven by the belief that ongoing construction would avert hauntings. The comparison highlights that when adding new "rooms" (features) becomes cheap and fast, continuous additions become tempting. The result can be a loss of conceptual integrity, which makes later architectural decisions more difficult.
Discipline as a countermeasure
The discussion returns repeatedly to discipline. Previously, the time cost of implementing an idea acted as a natural restraint: if a feature would take a week, developers might forgo it. If the same feature now takes an hour, it's much easier to justify. AI-driven speed removes that natural brake, so maintaining discipline and architectural control deliberately becomes crucial.
Closing thoughts
The episode underscores that while agents can increase development throughput (measurably producing more code), that alone does not guarantee better software. Preserving quality, maintainability, and conceptual integrity still requires expertise and disciplined practices.
(Note: the Wikipedia article referenced in the conversation contains credible sources that dispute the medium/psychic story associated with the Winchester House.)
