# Levels of AI use According to [[Steve Yegge]], these are the different levels to consider: - **Level 1**: No AI - **Level 2**: Using [[ChatGPT]], [[30 Areas/33 Permanent notes/33.02 Content/Claude|Claude]], and/or a coding agent in your IDE (permissions enabled, safety first) - **Level 3**: Still using [[ChatGPT]], [[30 Areas/33 Permanent notes/33.02 Content/Claude|Claude]], and/or a coding agent in your IDE, but in "YOLO mode" (increased trust) - **Level 4**: Not looking at the output in detail anymore, but at what the agent is doing. Not reviewing as much - **Level 5**: Using AI agents in the terminal with tools such as [[Claude Code]], [[OpenAI Codex|Codex CLI]], [[Gemini CLI]], ... And if you're a developer, you're not coding in the IDE anymore - **Level 6**: Multiplexing, managing several agents because a single one is not enough anymore - **Level 7**: Managing 10+ agents "by hand". Coordination becomes a real challenge - **Level 8**: Building your own orchestrator to coordinate more agents ## Video  ## References - https://newsletter.pragmaticengineer.com/p/steve-yegge-on-ai-agents-and-the - https://www.youtube.com/watch?v=aFsAOu2bgFk - https://www.bassimeledath.com/blog/levels-of-agentic-engineering ## Related - [[Gas Town]] - [[Levels of AI Context Management]] - [[Context Engineering]] - [[Harness Engineering]] - [[Agent System Engineering]] - [[Personal Context Management (PCM)]]