2 min read
If you've been using AI for a while, you probably have a routine by now. You type something in, you get something back, you tweak it, maybe go back and forth a few times. It works. It's kind of like texting a really smart friend.
But lately there's been a shift in how people are building with AI, and it's worth understanding even if you're not a developer. Because it changes what AI can actually do.
Think about the difference between following a recipe and using a smart oven. When you're following a recipe, you're involved at every step. Preheat to 375. Check it after 20 minutes. Is it browning too fast? Turn it down. Is the internal temperature where it needs to be? Switch to broil to finish it off. You're reading, checking, adjusting, the whole way through.
A smart oven does all of that for you. You tell it what you're cooking and how you want it done, and it monitors the temperature, adjusts the heat, switches modes when the meat is ready to brown. You come back when it's done.
That's basically what a loop is in AI terms. Instead of you going back and forth with the AI at every step, checking the output and deciding what comes next, a loop handles that process on its own. The AI does something, checks whether it worked, and either moves on or tries again. You set the goal upfront and let it run.
This is what people mean when they say "loops are replacing prompts." Not that prompts are going away, but that instead of you writing a prompt, reading the response, writing another prompt, and so on... someone has built a system that does all of that automatically.
Loops can go wrong in ways that are hard to predict. They can get stuck. They can keep trying the same thing over and over. They can rack up costs while nobody's watching. So the people building these systems spend a lot of energy on guardrails — ways to make the loop stop when it should and ask for help when it needs to.
For most of AI's history, the skill was knowing what to ask. Now it's starting to shift toward knowing how to design a system that asks for you.
The skill used to be knowing what to ask. Now it's knowing how to design the thing that asks.