Simon Willison Discusses AI Coding Agents, Lines of Code, and Conceptual Integrity on Talking Postgres Podcast
On August 19, 2026, Simon Willison shared a blog entry titled 'Conceptual integrity and counting lines of code', which encapsulated his conversation with Claire Giordano on the Talking Postgres podcast regarding 'How AI is changing software development'. He maintained that using lines of code as a productivity metric remains relevant, particularly with the advent of AI agents, pointing out that developers have traditionally generated 50–60 lines each day. If AI enables the creation of a thousand lines of quality code daily, it indicates progress but demands expertise. Willison highlighted the importance of managing cognitive load within teams, despite the accelerated coding pace. The post also featured a sponsorship appeal for a $10/month LLM developments digest.
Key facts
- Simon Willison's blog post 'Conceptual integrity and counting lines of code' was published on 19 August 2026.
- The post summarizes his discussion with Claire Giordano on the Talking Postgres podcast about AI's impact on software development.
- Willison argues lines of code can be a meaningful productivity metric for coding agents.
- Pre-AI engineers typically produced 50-60 lines of production-ready code daily, with 200 lines being exceptional.
- He believes agents can produce a thousand lines of debugged code daily, but only with senior-level skill.
- Cognitive capacity, not code generation speed, is now the limiting factor for individual engineers.
- He compared poor software coherence to the Winchester Mystery House, which has 140 rooms built over 40 years.
- Willison cited The Mythical Man-Month's concept of 'conceptual integrity' to explain risks of AI-generated features.
Entities
Artists
- Simon Willison
- Claire Giordano
Institutions
- Talking Postgres
- The Mythical Man-Month
- Wikipedia
- Claude
- Qwen 3.8 27B
Locations
- Winchester Mystery House