Opening Keynote – Yukihiro 'Matz' Matsumoto
summary
A new AOT compiler, Spinel AOT compiler, compiles a useful subset of Ruby to C and produces single-file native binaries with no runtime VM (limitations: no eval, limited metaprogramming). Benchmarks show ~6× average speedup across 28 tests and specific gains like 86×, 32×, 22× on microbenchmarks; the repo reports 1,744 tests passed and 58 benchmarks passing. Development is AI-assisted: AI writes and rewrites code, runs tests and benchmarks, and triages issues while human judgment gates merge decisions.
AI-written summary · about
chapters (26)
- 0:00Welcome to Matz
- 1:07Ruby’s philosophy and vision
- 3:36Best language for AI coding
- 10:15Ruby for AI and intelligence
- 11:20From projects to AI coding
- 15:38Building tools with AI in five minutes
- 20:25RSS reader demo
- 21:11Big integer AI multiplication
- 23:16Dream of Ruby AOT compiler
- 26:11Spinel: Ruby AOT compiler
- 31:11Self-hosting Spinel in Ruby
- 33:01Optimizing Spinel daily
- 38:07Spinel performance gains
- 40:21Spinel compilation demo
- 45:04Spinel vs JIT comparison
- 46:30Spinel limitations and features
- 47:15Spinel progress and updates
- 48:19Spin package manager
- 50:10AI agents for Spinel
- 53:06AI handling PRs at scale
- 55:36Human role in AI coding
- 58:33Why AI coding works
- 1:00:28Responsibility in AI coding
- 1:02:43Extreme vibe coding
- 1:07:23Spinel status and open source
- 1:08:30Closing thoughts
related talks
49mRubyConf 2024 Keynote: Yukihiro "Matz" Matsumoto
49mKeynote: Simple goal. Hard to accomplish- 49mKeynote: Simple goal. Hard to accomplish
48mQ&A with Matz
31mThe AST Galaxy to the Virtual Machine Blues
44mKeynote
01h 03mRuby3.0 and Beyond - Yukihiro "Matz" Matsumoto
33mRuby the Hard Way: Writing Bytecode to Optimize Plain Ruby