Self-Healing Software Using Autonomous LLM Agents
summary
Elixir system pairs GPT-4 with gen_server agents that monitor exceptions and CPU spikes, pull source from a context repo, generate and compile fixes via code_eval_string, run generated tests, and apply patches with hot code reloading using code_load_binary. Communication runs over Phoenix PubSub with a summarization agent capturing chatty LLM exchanges and stderr warnings. Demo at SpawnFest fixes recursive and CPU‑hog bugs while 10,000 concurrent jobs run, loop‑retrying LLMS until tests pass and returning a working solution.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 3:13LLMs and coding assistants
- 6:21OpenEI agent for Elixir
- 8:44Multi-agent self-healing design
- 11:15Runtime code fix challenges
- 14:07Demo setup with Sasha’s app
- 16:33LLM agents architecture
- 18:40Live self-healing demo
- 23:57Future possibilities and Beam benefits
- 24:43Q&A
related talks
16mBuilding Real-Time AI Agents with Elixir
28mLegion: Agentic Code Execution with Pure Elixir
19mLLM codegen fails and how to stop 'em
14mAutonomous Testing Agents with LLMs in React
35mBuilding LLM-based Agents
01h 03mAgents Everywhere
32mProgramming language evolution in the AI era
41mAutonomous Quality Engineering: AI Systems That Generate Execute Heal Learn and Govern Quality