How we automated code maintenance
summary
Describes a production system that generates ready to merge PRs to automate refactors and cleanups across Kotlin, Java, Swift, JS, Go and more using a compiler-driven engine plus framework-aware heuristics and a thin layer of specialized AI. Claims merged over 1,800 PRs with deleted over 500,000 lines of code in six months, reports ~90% of PRs require no manual edits, and says the open source Polyglot Piranha approach scales in monorepos.
AI-written summary · about
chapters (9)
- 0:00Automated code maintenance intro
- 1:51Why automate code maintenance
- 3:17Kotlin 2 upgrade challenges
- 7:47Feature flag cleanup case study
- 13:47Refactoring tool limitations
- 18:02Polyglot Piranha solves issues
- 20:54Using Polyglot Piranha
- 23:22Future of GitHub automation
- 24:05Key takeaways
related talks
12mHow we automated code maintenance
17m50 shades of PR
19mA Lightweight Polyglot Code Transformation Language
01h 17mRefactoring the ROI of Software
39mMoving fast and staying aligned how automation powers progress
48mSay goodbye to bugs and anti-patterns with Error Prone
01h 57mAutomatically improving Java codebases with Error Prone
34mPiranha: Building a Multilingual Deadcode Deletion Bot
