AI’s Big Leap
summary
Shows a concrete system that turns parsed code into searchable, editable data using lossless semantic trees and modular open rewrite recipes; a migration example spans 3,371 steps of tiny changes and produces ready diffs. Uses embeddings to cluster 60+ build failures to prioritize fixes, an LLM to restore broken French accents in comments so JavaDoc compiles, and an agent called Modi that selects and runs recipes and summarizes results.
AI-written summary · about
chapters (18)
- 0:00Introduction and motivation
- 1:05What AI’s big leap isn’t
- 2:29Lossless semantic trees explained
- 4:12Recipes for code migration
- 6:38AI in the development rollercoaster
- 8:44AI for build failure clustering
- 13:19Embeddings for prioritization
- 16:29AI-powered recipes framework
- 17:55Visitor pattern in recipes
- 20:01Fixing encoding errors with AI
- 22:08Knowledge graphs and RAG
- 25:19Generating method descriptions
- 26:43Updating README files with AI
- 28:51Introducing Modi the agent
- 30:59Demo of Modi in action
- 33:56Why recipes are the backbone
- 35:22Resources and closing
- 36:47Q&A session
related talks
21mFrom Spaghetti to Structure: Decoding Complex Codebases with Graphs and GenAI
26mBeyond the code: Rediscovering the timeless skills of software engineering in the AI era
46mGenerative AI and Java
46mAI for Kernel Engineers
54mAI is Coming - Now What? Architecting the Future of Software
27mAI for software development: A reality check
34mJavaScript-Powered AI: Crafting Intelligent Systems with LLMs and Vector Databases
29mHow to roll your own LLM