Managed Agents vs Routines vs Schedule

They've introduced a powerful new feature, which actually isn't available in the CLI Claude code, not that I could see anyway, routines. So It says schedule routines. That's a bit strange. Yeah, so there's a lot of I think they've got a lot of cleaning up to do because but certainly over the last few weeks it's been a bit of a so much so that I was confused as to what the difference between this is and managed agents. They share so much of the same overlap that I was like, why? What's managed agents? Routines and managed agents both run on Claude's or Anthropic's infrastructure. It's on their service. You just you configure them, you set them up. They can both be triggered by like posting to it or routines, like a loop, a cron job, or in the case of routines, you can actually trigger it from GitHub actions, which starts to you start to see the divergence. They then run a set of routines, they do what you what you've told it to do, and that that's kind of like the job. Still quite a lot of overlap here. So I again, I wish that let's call the managed agents inside of Claude code is different from managed agents inside of co-work or something like that. I don't know, but whatever. Um the idea is is that routines are available inside of Claude code because really they're to do with code-based stuff, right? It's to do with it's routines that act upon your code base. The real intention is to be doing code reviews. The real intention is to like do refactoring tasks or, you know, bug reports or whatever it is, whatever you want. It's it's based on a code base. Now that code can actually be in of itself like something that can run other things, I guess. This is me just making up right now. I don't actually know. But that code could be ran inside of this routine that has that goes and fetches Notion data and then the agent acts on that the receiving end of that notion database or something like that. I don't know. But there's me trying to merge something that managed agents kind of does. The idea of managed agents and we've done a video on this is that again, you're replacing NAN. You've got a trigger. It's got a set of actions and it's an agent itself so it will do a bunch of tasks and whatever. It can read from different MCPs. It can read from different skills to get that task done and then you know, whatever. Um the output is is whatever you want it to I thought again, confused about the overlap. I thought, "Ooh, this is quite nice. This is like managed agents." because this comes under your pro and max subscriptions. This is now this is charged under your max subscriptions. I couldn't see anything about um uh you're paying for your usage even though it's run on Claude's uh infrastructure. I couldn't see any like it's you know, X amount of minutes cost you this plus you know, whatever. So it's seemingly like a free thing for you to do. So that's quite nice. Um So yeah, that's kind of routines really. Um it's a it's more of a code-focused, dev-focused thing that you want to run on some sort of schedule or you you tell it when you want it to run, you know, do a code review and then it'll just do a code review or something like that. Pretty cool. what is the difference between routines and schedule is a bit still unclear to me because here I guess the remote aspect. Yeah. I guess the remote aspect. These again, you've just you've just hit the nail on the head there. These can just run on a clock. Whereas if you do the other one, you get you can do the clock but you can also do post. You can also post to it. Yeah, so one's local which will just be a cron job and the remote you can actually run on a schedule or a trigger here some sort of action or post request API's like a post request or whatever. So, yeah. Just So, could I connect this to Telegram to do something? Um you what is in like you post from Telegram to this? I guess. Yeah. I would probably well you probably need to build some sort of middleware layer to forward it on cuz I don't I don't know I don't know what Cuz cuz we have the API and we have the connector so technically I could talk to my web flow through Cloud and Telegram. You could but you need to you need to transfer the I I don't know what connectors it has against Telegram but you might need to trans you might need to translate a message into a post request with middle something in the middle there. So, you probably could but you need to build something in the middle to say when I receive a message post to this remote routine. Um yeah. So, you know, it's it's cool. Um Yeah, that's all I have to say about it. What I will say is that we have uh Warp and Warp have had this thing called Oz for quite a while which is a managed agent cloud thing. So, if you type in I think it's Oz {hyphen} platform or {slash} Oz platform. Just type {slash} Oz. O Z Does it say anything? Yeah. It's this one. Yeah. It's gone blue, is it? Yeah. Yeah, here we go. Yeah, so Oz platform. Now now you can you can also do the same things here where it's just again a response to a GitHub. Excuse me. Runs response to a GitHub or whatever that's happening happening remotely and then it just pulls down your code, does something in your code, and whatever, and then pushes back up. This is part of a larger conversation on my show Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI.