- Author
- Jay Baleine (GitHub user Varietyz)
- Published
- 2025
- Kind
- GitHub README for a method, with an FAQ in the author's voice
- Original
- github.com/Varietyz/Disciplined-AI-Software-Development
- Subject
- Coding with agents
What it is
Jay Baleine publishes the method under a 2025 copyright. A companion site is called Bane's Lab. The README is the front page of a larger set of documents. It sets out the problem, the loop and a getting-started path.
What it says
- Baleine says models fail on long projects through repeated code, drifting architecture and diluted context.
- Baleine says one ten-step loop applies to all work. Each check comes before the code.
- Baleine states four rules. One says every manual step counts as a failure of automation.
- Baleine says rules kept in project files hold better than rules kept in a chat.
- Baleine says a correction given twice should become an automated check.
- Baleine claims no deviation from the method since the first version.
What evidence it gives
The README offers description and the author's experience. It claims over eleven thousand hours with models. It points to a website as its exhibit. It reports no measurements or outside trials.
Interests and limits
The author promotes the method and a companion format, and asks that PAG be cited. Only the README was read, not the six chapter files or the linked site pages. Claims about results come from the author alone.