The Startup Ideas Podcast What are Agentic Loops?
375 snips
Jun 9, 2026 
Ras Mic, AI researcher and educator, breaks down agentic loops and practical AI tooling. He contrasts human-in-the-loop workflows with autonomous loops. He describes technical loop mechanics, token cost pitfalls, and when loops actually shine. He shares a hands-on closed code-review loop using Cursor, GitHub, and Greptile.
AI Snips
Chapters
Transcript
Episode notes
What An Agentic Loop Actually Is
- Agentic loops fire once from a human and then let the agent generate, review, and feed its own results back to continue building without ongoing human direction.
- Ras Mic explains the loop: spec/PRD.md → agent generates → agent reviews its output → agent iterates, aiming for autonomous progress.
Avoid Open Loops For Building Products
- Avoid running wide-open agentic loops for product development unless you have an enormous token budget and don't care about fine details.
- Ras Mic warns these loops make many assumptions, misalign with product vision, and burn tokens quickly in practice.
Don't Use Slash Goal On Small Token Plans
- Only consider slash-goal / loop features if you can afford high token consumption—reserve them for high-tier plans.
- Ras Mic recommends against using loops on $20–$100 plans because token burn is substantial.