Code Mending in the AI age
summary
Shows that legacy C from an IOC C obfuscated entry compiles and becomes maintainable by combining GPT-4 suggestions with syntactic refactors from clang‑format/clang‑tidy and JetBrains CLion, guarded by approval tests that capture the first 5,000 lines and a test‑commit‑revert cycle. Claims LLMs never directly edit production code; they propose renames and extracts while a human vets and the CI runs binary‑identical checks. Also reverse engineers a 15‑year‑old banner encoding (RLE + Huffman + 6‑bit ASCII) and successfully swaps in a new string.
AI-written summary · about
chapters (17)
- 0:00Legacy code in AI age
- 1:03Obfuscated code IOCCC 2006
- 2:48AI improves tangled code
- 5:19Uncompilable code compiles
- 7:05Tests and golden file validation
- 8:14AI refactoring suggestions only
- 12:45Custom AI refactoring prompt
- 14:51Renaming variables systematically
- 17:21Extracting functions like render
- 18:30Decoding ASCII with AI help
- 21:19Victory on maintainable code
- 23:28AI-assisted refactoring cycle
- 24:53Why AI cannot replace legacy work
- 25:55Need for skilled developers
- 27:01Reverse engineering ASCII banner
- 30:12Proving understanding by modifying
- 31:19Answering audience questions on refactoring
related talks
50mIdentifying and fixing Issues in Code using AI based tools
28mFixit linter+AI coding
55mSo, you inherited a large code base...
24mAI Hates Legacy Code
57mKeynote: Using LLMs and generative AI to fix software vulnerabilities
33mAccelerate Code Quality in the AI Age
29mOne-Click Code Fix: Securing Code Using AI
40mCode Rot at Machine Speed: Preventing AI-Induced Technical Debt