Talking Shit About AI Agents
summary
Claim: what the industry calls agents is a new programming primitive akin to Smalltalk objects, reframed as prompt objects that combine an LLM, prompt, loop, tools and an interface. Demo uses a Ruby gem prompt_objects where markdown files are programs, objects self-modify, add a read_file tool at runtime, create new objects, and run ARC‑AGI puzzles on a small cheap model. Presenter shows semantic late binding and a Sonic Pi composition that actually runs.
AI-written summary · about
chapters (8)
- 0:00AI agents and Ruby
- 2:28Anthropomorphic trap in design
- 5:15Prompt objects as primitive
- 10:16Demo: prompt objects gem
- 13:08Self-healing pipelines
- 14:11AI ownership of Smalltalk
- 15:37Prompt OOP principles
- 17:26Conclusion
related talks
50mLet's build an AI agent
48mStop Fighting Your AI Engineering Prompts That Actually Work
21mDemystifying AI Agents: A New Primitive for Flow Control
19mDon't Build Slop (4 Levels of AI Agent Maturity)
59mFrom Pair to Peer Programmer
24mThe truth about AI development
40mAgents in Action: Advancing Open Source AI Missions
32mProgramming language evolution in the AI era