Legacy Rails and the AI That Couldn't
summary
Legacy Rails repo with messy conventions meets agent-driven experiments: build a pod-only directory of guideline markdown, teach agents a TDD workflow that writes a failing test first and pauses for human approval, expose that workflow via an explicit '/pi dev' skill, and implement PiPR that fetches historical PR edits to synthesize better descriptions and test plans. The tools are chat agents plus GitHub APIs; outcomes are fewer repeated corrections, clearer PRs with mermaid flow charts and executable test plans, and a steady drop in review overhead that restores confidence.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:15Legacy Rails struggles
- 8:16AI experiments
- 11:49Guidelines to workflow
- 15:16Human expertise for front-end
- 18:07Better PR descriptions
- 22:04Automation vs manual testing
- 26:16Self-improving AI workflow
- 28:45Key takeaways
related talks
32mFrom Toil To Triumph: Harnessing Agentic AI To Streamline Infrastructure as Code
40mHow To Tame Your Agents?
55mKeynote: Exoskeletons, not Autopilots
49mAgile for Agents Managing Robots The Way We Manage Humans
30mBuilding Tools with Tools: Modern AI-Driven Development
27mI know Kung Fu! RubyGems for AI
16mDay 1 Recap: What Practitioners Learned
37mAI-Driven Developer Productivity at Scale