Hook

Their other posts in the index, biggest breakout first.
The creator of Claude Code said he stopped prompting Claude. Instead he writes what he calls AI loops. And once you realize how easy it is to make this, you probably won't go back to regular prompting again. So here's how a typical person prompts today. You enter something like give me 10 ad ideas for my product. And the AI returns generic ideas. Then the user refines or prompts, hook stronger, give me better angles. The AI gives another output. Then the user refines and this process kinda just repeats. So you ask, AI answers, then you decide the next step. How an AI Loop Works. Generate, score, improve, then select winners. Objective, find the strongest ad angles. Scroll-stopping, specific, easy to demo. Metric, hook strength, pain point, specificity, demo potential, conversion intent. Boundary, three improvement rounds, remove anything below 7/10, show top 5 only. Generate 20 angles. Score each one. Remove weak ideas. Rewrite promising ideas. Repeat 3 rounds. Top 5 winners. Example: Weak idea, "Try our new skincare product." 5/10, remove. "Products that dry your skin." 7.5/10, improve. "Routine with this one product." 9/10, keep. AI becomes the feedback loop. AI judges and improves the ideas before you see the final list. The AI loop actually fixes that. You start with three things. What are you actually trying to achieve? So how can AI judge if those ideas are actually good? You set a boundary. So how many times will the AI improve an idea before it stops? So a full example of this would be, generate 20 ad ideas for my product. Then score each one on hook strength, pain point, demo potential, conversion intent. And remove any that score below 7 out of 10. And rewrite the promising ideas. Repeat this for three rounds and only shows you the top 5 winners. That's the tip. Try it out if you want to comment loop if you want me to share the guide.