99% of people prompt AI. The top 1% do this instead

@dominiccapaci

7.0K followers · usually 1.5K views a post

Views

387.8K

Likes

21.3K

Comments

290

Shares

1.8K

SoundTikTok sound · original sound - dominiccapaci

Promoted product

The AI Playbookhttps://skool.com/the-ai-playbook ↗

Why it worked

The video uses a strong contrarian hook that promises a 'top 1%' secret, immediately followed by a high-value, actionable technical explanation that establishes the creator's authority.

Summary

The creator explains the concept of using 'loops' in AI prompting to automate the iterative process of checking and refining AI outputs. He encourages viewers to follow him for more advanced AI workflow tips, linking to his course, The AI Playbook.

Structure

1Hook: 99% of people prompt AI, 1% use loops
2Explain the problem with standard prompting
3Define the loop concept and how it works
4Provide a practical example of building a landing page
5Emphasize the importance of specific, measurable goals
6CTA: Follow for more examples

Product placement

The creator mentions 'loops' as a workflow technique, which is the core subject of his course 'The AI Playbook' linked in his bio.

Original caption

The top 1% of ai users build with loops now. Prompting means you check the work and Loops mean it checks its own work and fixes it. Level up your claude output with loops #aitools #aitips #aiworkflow #claude #aiforbeginners

More from @dominiccapaci

Call to action

If you thought this was helpful, follow along because I'll be posting more examples of loops that you can steal from me.

On-screen text

99% of people prompt AI. The top 1% do this instead

Transcript

So the guy who built Claude code at Anthropic just admitted that he barely prompts Claude anymore. He uses loops instead. I've seen a lot of people talking about loops but not actually explaining them. So I'm going to break down exactly what they are and how to build them. So when you prompt AI like normal, it gives you a response once and stops regardless of whether that answer was good or not. You're the one who's responsible to check the quality of that answer and then if something's wrong, you have to go prompt it again to fix it. So you and Claude make this loop. And here's what he does instead. So all the loop does is adds a goal and Claude does not stop working until it hits that goal. Claude will check its own work and if it's not there yet, it will write a new prompt to get it there. And it will keep doing this until it genuinely passes the bar that you set. So you can just walk away and it keeps going. A good example of this would be to tell Claude to build a landing page for your business. Without the loop, you're the one who's got to open it up, test it out and then maybe realize it doesn't look good on mobile or something. But with the loop, you can set some goals like it's got to load in under two seconds and the landing page has to be mobile friendly. So Claude will then build that page and then the next step would be to test the loading time and then to test the mobile features. And if it's broken, Claude will just prompt itself again to fix it without you doing anything. And something that's very important that's overlooked is you have to set very specific actual goals and like use numbers when you can. This is how Claude becomes capable of actually grading itself. So the goal can't just be like make it look good, you got to be more specific than that. It's got to be something it can actually check. If you thought this was helpful, follow along because I'll be posting more examples of loops that you can steal from me.