Transcription
So, Anthropic just released agent loops today, and there's already a stupid amount of misinformation out there with people acting like agent loops allows you to use Claude code to do scheduled tasks persistently forever. That is not the case. That is not the case at all.
Anyone who's telling you, "like, here's how we do it, you can now use it with Telegram or Discord," has no idea what they are talking about.
The way loops works, okay, inside the CLI, and understand we're talking about the CLI, not Claude code desktop, because there is a difference, with how the scheduled tasks inside the CLI, loops allows you to do essentially like micro tasks that you would be repeating over and over and over again for a specific project. Think, "hey, you're doing a bunch of deployments. I now want to use the loop function to have it check my deployment status every five minutes."
What you would not use it for is like this webhook that's waiting for responses from Telegram for a number of reasons. One, it is session-based. You would never have Telegram do it like that because if you shut down your session, or you shut down your computer, or this goes over three days, it stops, right? So, it makes no sense in that context.
Secondly, if you want to do something that's more persistent, you would use Claude code desktop because there is no 3-day limit to it. It's not session-based. It runs in perpetuity. However, it still needs to have the computer on. It still needs to have a Claude code instance there.
So, like, loops are really cool, but their scope is actually pretty tight. And so, you're going to see a bunch of people just saying absolute nonsense. Understand they actually have no idea what they're talking about. They haven't used it, and they clearly haven't read the documentation because it's pretty clear.