AI Briefing
KO

Cathedral, Bazaar, and the Winchester Mystery House — The Third Model of Software Development in the AI Era

·2026.06.04 10:55

Key point

With the rise of AI coding agents, the software development model is evolving into the 'Winchester Mystery House,' a personalized, ever-expanding approach.

Details

Beyond the 'Cathedral (closed)' and 'Bazaar (open)' models defined by Eric Raymond in 1998, the emergence of AI coding agents has given rise to a third development model: the 'Winchester Mystery House.' This refers to a personalized, ever-expanding approach in which developers use AI to endlessly add on their own tools.

The Plummeting Cost of Code Production AI agents like Claude Code can generate roughly 1,000 lines of code per commit. This figure overwhelms the daily output of a human developer (about 10-30 lines), effectively driving the cost of code implementation to near zero.

The Bottleneck Shifts: From Implementation to Review While implementation costs have plummeted, the pace of human-centered tasks such as code review, discussion, and testing remains unchanged. As a result, the bottleneck in the development process has shifted from 'writing code' to 'Attention' and 'Review.'

Characteristics of the 'Mystery House' Model

  • Idiosyncratic: Because developers run feedback loops with agents tailored to their own personal preferences, the resulting structure has a unique form that is difficult for outsiders to understand.
  • Sprawling: With virtually no additional cost, there is a strong tendency to keep adding new features rather than modifying existing code.
  • Fun: The process of refining one's own workflow becomes a hobby and a quest in itself.

Future Challenges The massive volume of AI-generated contributions (PRs) could paralyze the review capacity of open-source ecosystems. Going forward, 'tools that make attention cheap'—helping to sift out high-quality ideas from the flood of code—will play a key role.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.