Transcription
Less than 24 hours ago, Anthropic released agent loops for clawed code. And people are losing their minds. But they're losing their minds for the wrong reason. And let me explain.
So what agent loops allows you to do is run prompts on a schedule. So imagine you wanted clawed code to check what your deployment status is. And you wanted to check it every 10 minutes within the same session. We can now do that with loops.
But now you have people saying outrageous stuff like this is a way to essentially open claw clawed code. like this is you would use loops for Telegram or Discord notification connections. That is not the case at all. And that's because you need to understand how loops work within the CLI. And there's some key distinctions you need to pay attention to.
First of all, these recurring tasks only last 3 days max. And they are session-based, which means if you restart that terminal, if you close the terminal, if you shut down your computer, it ends. This is not a persistent version of clawed code.
And because it's session-based, that means the scope of these new agent loops is actually pretty tight in nature. It's really depends on what project you're using. This is not some silver bullet for like I just want claude code in the cloud to always be running and doing stuff. That's not what this is.
And in fact, Claude Code Desktop actually does stuff like that a little bit better. If you look at those same things, right, the Claude Code Desktop actually allows you to run scheduled tasks and it doesn't expire. You don't need to have that same terminal open. Although you do need the desktop in your computer open in some manner, although it runs in different sessions.
All that to say is that loops, while very cool, are not the crazy, you know, game-changing moment people are saying it is. So just know that going.