TL;DR
Developer and educator Ross Mikic joins Greg Isenberg to demystify agentic loops — the AI trend where agents run autonomously without human input. He argues that for most developers, fully autonomous loops burn tokens, make bad assumptions, and produce poor results, and that human-in-the-loop remains the superior approach except in narrow, binary-output use cases like automated code review.
Key points
1
Human-in-the-loop (you prompt, review, and iterate) is still the most effective and cost-efficient approach for building real products
2
Agentic loops (tools like /goal or /sloop) let an AI agent run autonomously using its own output as feedback, but they make unchecked assumptions that rarely align with your product vision
3
These loops burn significant tokens — Peter Levels reportedly spent $1.3 million in tokens in a single month running such loops
4
Loops do make sense in closed, binary-output scenarios: Ross uses a code review loop with Graphite that pushes code to GitHub, reads the AI review score, and iterates until hitting 4-5 out of 5 before allowing production deployment
5
Even his own constrained code review loop breaks when pull requests exceed 1,000 lines of code, illustrating that current agentic loops are fragile even in ideal conditions
Actionable insights
→
Do not use autonomous agentic loops (/goal, /sloop, etc.) if you are on a $20 or $100/month plan — they will exhaust your token budget without delivering production-quality results
→
Use agentic loops only for binary, rule-based tasks with a fixed feedback signal — code review scores, SEO page generation, or benchmark simulations where details do not matter
→
For serious app or startup development, stay human-in-the-loop: prompt, review results, test, and iterate yourself — AI cannot replace your product judgment or vision
Notable quotes
“AI can replicate sauce. It cannot create sauce.”
“Unless you want to donate to trillion-dollar companies, this is not it right now.”
“Human in the loop is the best loop.”
Worth watching?
⏭️
Worth watching the full video?
The key arguments and the concrete code review loop example are fully captured here — skip the video unless you want to see the hand-drawn diagrams Ross walks through live.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free