WTF Is an “AI Agent Loop”? Genius or Hype?

Key Points

Ross Mike explains agentic loops, whether they’re hype, and a real use case.

  • Human-in-the-loop (how most work): you prompt an AI agent (Cursor/Claude/Codex), view + test the result, iterate. The loop runs through you. Example: build landing page → review → auth → backend, one step at a time.
  • Agent loop: the agent prompts itself (or agents prompt agents) without you in the middle — the mode people with “unlimited model access” (Boris, Peter, Anthropic/OpenAI insiders) use.
  • The contrarian take: for the average person, running full agentic loops is “a terrible mistake unless you have money to burn” — they burn tokens fast. The general hype consensus is, in his view, wrong.
  • He still demos one loop he actually uses — the concrete, cost-justified use case (shown at the end of the episode).

0 items under this folder.