What On Earth Are Claude Managed Agents?!

This article is the most horrendous article to describe what managed agents are. Command AI. Don't adopt a pet. Like, what is all this? It's mental. So, let me just break it down for you. Managed agent is an agent that sits in the cloud waiting or watching for something to trigger it. And it has its own system prompt, what it can do, what it knows about, what its task is. Like, it knows its task, and it's just sitting there waiting to be used. Very, very similar to N A N in that you can set a pipeline, and you just set a trigger on it. It could be a Slack message, right? So, you could build a managed agent that every time a Slack message comes in, fires off to your managed agent, it does the task of looking in the Notion database on your knowledge base. Uh it could be it generates an image I don't know, it could be anything. It could be anything, and then it fires it back. And that's managed agents in a nutshell. I don't know why they've over complicated it here. So, so is it like, you know, how you can have a cloud cool work and schedule tasks? Is it like that in the cloud? Exactly. It's like that, but it's in the cloud. So, I built a workflow, which maybe I can show you actually. And I think this confuses things because it's in N A N. Um but let's I hope you can bear with me here. So, I have this workflow that every time my one of my videos goes live on my YouTube channel, I transcribe the video, I get the video and then transcribe it, and then I send it into AI to like write the little description that happens below the video. And I also get it to generate like a little blog post and stuff. I do a bunch of things. I instead of using and I use Gemini for this. This is a This is a Gemini thing. I built a an agent that a managed agent that does that for me, but it uses Claude to do that. All I do is send it the transcript. My managed agent is waiting for it on the back end, which again, I'll probably show you that. You've got this new managed agent section here. Um Here it is. Here it is. It's basically the same thing that's over in NAN. It's waiting there in the cloud. And so you create a session with it, throw it the throw it the transcription. It's just waiting for the transcript. That's all it's doing is waiting for that transcript, and then it gives me a response. Beauty of it is because this is their own ecosystem here. You could just you you you have all of that like trigger the trigger built in. So this one, the support agent here, this is waiting for any Slack message. It goes off and fetches from the Notion database and responds as a support agent. You know, so you've got all of you've got skills built in as well. So can access skills. Um you you um it's got memories, which you have to request access to, but it will have memories soon. Um and the MCPs, of course. It has all of the MCPs that you have access to. So it's it's that ecosystem. So it's pretty nice. It's pretty cool. Definitely replace a few um NAN if that's if that's your bag. However, this is another one of those things that does not come under your max plan. This is I could not My notes here tell me it's 0.8 per agent run time hour plus model usage. Now that's that's on my notes, but I couldn't find that anywhere. I couldn't find that figure. I just I all I found that it's just um uh you just pay for the model. You pay for Sonic. $3.15 out or whatever the hell it is. So in my opinion it's pretty damn expensive to do something that you could easily do in an A N. Like, and also no not being funny, but in an A N you can still use Claude. You can still you it it's not that different. You you get access to skills, which is something you don't get in an A N. You can still access MCPs. Do you know what I mean? You can still go off and fetch your notion database or what or this or that. So, you don't actually getting an awful lot. It's it's not It's like But even with skills, can't you just have a bunch of text in a different database as a node in an A N and you tell your agent, "Hey, if there is a mention of X and Y, go look at this database and take the read the file that you need and it's just doing that?" You could. Um and and to be fair, let me get this back up. Like, I would I would have said the nicety would be that you could share your skills that you have inside of your Claude instance, but this is outside of all that. So, the only skills you have access to are this Excel one, the PDF one, the PowerPoint one. You can add your own. But, my point is there's not even a synchronization with your existing tools. So, yeah, you might as well have it as a base prompt or a file somewhere in your database that's effectively a skill uh to help it get jobs done. You just have to be really into the Claude and Anthropic ecosystem to justify this tool. So, yeah, I like it. It's quite cool and it's very easy to set up. You can just describe your agent and it will figure out everything for you. Which is nice or you can start with a template. But, again it's an expensive expensive tool that you could probably you're better off just doing it in an AI. 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