7 INSANE loops you need to try right now
Channel: Matthew Berman
Uploaded:
Duration: PT971S
Read a transcript preview of “7 INSANE loops you need to try right now” from Matthew Berman, uploaded on 2026-06-19.
Transcript preview
Language: English (original, YouTube automatic captions)
[0:00] Loops are emerging as the single biggest
[0:02] unlock for people building software with
[0:04] artificial intelligence right now. But
[0:06] most people don't even know what loops
[0:09] are. And so today, I'm going to tell you
[0:11] what loops are. I'm going to show you
[0:13] why they're valuable. And then I'm
[0:14] actually going to give you many specific
[0:18] use cases that you can use loops for
[0:20] today. So, what is a loop? A loop is a
[0:24] way to allow your AI coding agent to
[0:27] work autonomously towards a specified
[0:30] goal. The most important thing about
[0:32] loops is that it removes humans. That
[0:36] allows the agent to work much more
[0:38] quickly towards this defined goal. And
[0:41] if it sounds very theoretical, I am
[0:43] going to break it down. So, what is a
[0:45] loop more specifically? Well, you need
[0:48] two things. You need a trigger and you
[0:51] need a goal. With those two things, you
[0:54] can complete the loop. A trigger is what
[0:57] kicks off the loop. And there are three
[0:59] ways to kick off a loop. One, you can do
[1:02] so manually. You literally tell the
[1:04] agent, "Go do this loop." Two is
[1:08] schedule. You can schedule a loop to
[1:09] happen at a certain time of day or on a
[1:12] repeating schedule. And then three, you
[1:14] have actions. You can have the loop kick
[1:17] off based on some kind of action like
[1:19] opening a PR. Now, to fully remove the
[1:22] human, we wouldn't want to kick
[1:23] everything off manually, but sometimes
[1:25] it is required. All right. And for the
[1:28] goal, the goal can be basically one of
[1:31] two things. It can be verifiable or we
[1:34] can use LLM as a judge. So, if it's
[1:37] verifiable, it is something concrete,
[1:39] some