| purpose | Ground lessons about why agentic software factories fail. |
|---|---|
| applies_to | Teaching sessions about safe AI-assisted software development. |
| entrypoint | Start with lessons/0001-why-software-factories-fail.html. |
| verification | Learner can diagnose reward, blind spot, delay, and steering point. |
| update_when | The learner's desired real-world outcome changes. |
Develop sound judgment about agentic software workflows so increased coding throughput does not quietly destroy maintainability. Use that judgment to recognize unsafe “lights-off” automation and choose where human steering has the highest leverage.
- Explain the causal mechanism behind software-factory failure without relying on slogans
- Distinguish fast verification from evidence of long-term maintainability
- Diagnose whether an agent workflow is optimizing the wrong feedback signal
- Choose an appropriate human review point for a real change
- Lessons should be short, concrete, and applicable to real repositories
- Prefer primary sources and observable evidence over AI-industry hype
- Build durable recall through retrieval practice
- Building a fully autonomous software factory
- Comparing specific coding-agent products