Going Postel
summary
Argues that Ruby’s flexibility makes LLM-generated code practical and shows working tools: blueprints Rails server and editor plugin captures code patterns, creates vector embeddings, and generates variations that often compile; an agent DSL for TDD runs tests as goal checks, sends failing tests + files to an LLM and saves generated implementations until tests pass; a synonym module via method_missing resolves hallucinated method names at runtime; demos pass tests and recover from hallucinations.
AI-written summary · about
chapters (11)
- 0:00Ruby and generative AI
- 1:26Postel's robustness principle
- 3:41Ruby features inspired by Postel
- 6:07LLMs and Postel's law in Ruby
- 7:53Code examples with Blueprints
- 11:07TDD with AI agents
- 16:07Patterns and tests for chess app
- 19:25Handling hallucinations with Synonym
- 21:35Non-mission-critical scripts
- 23:04Future of Ruby and AI
- 25:11Call to action and resources
related talks
28mGetting More Out of LLMs for Ruby Developers Justin Wienckowski, Jake Gribschaw We're all hearing about LLM tools that are changing on a daily, if not hourly, basis
31mSeeing Metaprogramming and Lambda Function Patterns in Ruby
21mRubyLLM
24mRubyLLM: Making AI Development Beautiful Again
53mCode Generation: ....
30mRuby Techniques by Example
01h 15m10x Development: LLMs For the working Programmer
01h 10mDomain Specific Languages: Molding Ruby