Recursive Coding Agents
summary
Claims specific wins: RLMs let agents operate on external prompts as executable state and achieve real benchmark improvements — Qwen 3.59B as an RLM beats GPT-5.4 on Long CoT and DSPy yields state‑of‑the‑art long reasoning; Symbolica’s Agentica reaches ~30% on Arc AGI-3 versus 2–3% for frontier LLMs. Implementations include a CLI wrapper, a Y pi recursive coding‑agent extension for pie, Cloud Code with Claude Code Dynamic Workflows enables recursion, and OpenProse repos with runnable examples and verification hooks.
AI-written summary · about
chapters (18)
- 0:00Recursive coding agents intro
- 1:46Mismanaged genius in AI agents
- 2:48What are recursive language models
- 4:31Why RLMs matter
- 5:15RLM power examples
- 7:22RLMs vs frontier models
- 9:11Defining true RLMs
- 10:37RLMs to recursive agents
- 11:22Experiments with recursive agents
- 13:50Notable recursive projects
- 15:41Is Cloud Code an RLM
- 16:44Cloud Code dynamic workflows
- 17:27OpenProse language
- 18:54How OpenProse enables RLMs
- 20:20Recursive agents applications
- 21:00Golden sessions as workflows
- 22:05Key takeaways and future
- 23:10Q&A
related talks
48mAI Coding Agents and how to code them
40mAI Coding Agents and how to code them
48mAI Coding Agents and How to Code Them
01h 59mEmerging Patterns for Coding with Generative AI
15mAgentic Reasonings with Reinforcement Learning
01h 03mAI Coding Agents and how to code them
30mBeyond the Hype: Using AI Coding Agents Today
50mLet's build an AI agent