# 157 Claude Managed Agents Clearly Explained (and why it matters)

# 157 Claude Managed Agents Clearly Explained (and why it matters) Speaker 2: Anthropic just released Claude Managed Agents about 48 hours ago and everybody's freaking out about it, but they don't really know what it is or what it can do, or at least they don't fully understand it. I brought on Nick Spisak to go over what is Claude Managed Agents? What does it do? Who is it for and who is it not for? We're going to clearly explain it in detail in this video. Nick, let's go ahead and dive in. What is Claude Managed Agents? Speaker 1: Claude just released the capability to have no infrastructure and to package your existing agents directly on their platform to be able to get your first custom agent done today. As we cut through this, the biggest thing here with Claude Managed Agents is they're going to be broken into this concept of this nice little picture we have here on the left-hand side, where you have your existing agent harness, which is commonly known as Claude Code. And you take your tools, your sessions, your orchestration, your sandbox, and they're totally decoupled. So now you have your hands and your intelligence are broken into two, and there's something that are completely throw-away-able and portable to be able to reuse. Speaker 2: Awesome. So let's get into who Managed Agents is for because it's my understanding that certain people that are more technical, it might not be the best option for them and people who are totally non-technical, it also might not be the best option for them. So let's talk about who it's for and who it's not for. Speaker 1: Yeah, so we broke this down into really four different types of personas. So the personas that we have here is if you're the type of user that likes to use either ChatGPT chat or Claude chat or you're in co-work Environment, this is really specifically just for you. What do you care about? You care about results. So the best way to do that is this platform is perfect for the individuals, the non-technical users, the business owners that are really just looking for the ability to have results right out of the box, something you can put out there and it's ready to go same day. The second persona is your typical Claude code. User, they're looking for flexibility, speed and reuse. Yes, every one of these personas cares about results, but they also care about those specific things. And they're going to be the type of users that are going to be much more familiar with the command line. And one of the benefits of Claude Managed Agent is they have a command line interface known as Anth, what is the abbreviation for Anthropic. And it allows you to do everything that you do in the platform portal. That a co-work persona would typically do, but now you have the ability to be able to take your existing files, your skills, your MCP servers, and all the rest of the plumbing that you would normally do, and you can deploy that through the command line. The third persona, this is someone that is already familiar with the agent SDK or software development kit. This is your power user. They've already seen the light in terms of wanting to use agents. And the main difference for them is they already have their own infrastructure. So this is the big divider line where a Claude managed agent runs on Anthropics Cloud on their stack and a agent SDK persona still uses the agent persona, but they're doing it on their infrastructure. So they care about all the same things that the other personas of the stack do, but it's an extra point where they want that ultimate flexibility. This type of user is going to be using Python or TypeScript. So we're talking programming languages for this. And finally is the AI tinkerer. The AI tinker is someone that is potentially a solopreneur. They're curious about AI. They're always up to date on the latest trends. They're going to care more about curiosity and cost. And for this user, ultimately at the end of the day, at least right now, this is not the best platform for them because they ultimately care about the cost and those aspects. This managed platform is expensive to run. It runs on a token as a service model. Whereas this type of persona generally is going to be better equipped for using the Claude Code subscription. Speaker 2: Got it. Okay. And then so for the folks watching this too, what Nick's put together is a pretty lengthy Google Doc that you can literally hand to Claude Code or to your AI agent in Codex or wherever you've got a coding agent. And it will build and deploy your first Claude Managed Agent in 60 seconds, literally just by handing it to Google Doc. So if you guys are watching on YouTube, you can get it for free at the link in the description. If you're listening on the podcast, you can grab it in the show notes. But just grab that Google Doc, hand it to Claude Code or Codex and let it build your first managed agent and just try it out for yourself. But Nick, so what you're saying here, if I understand correctly, that Managed Agents is for the Claude user or the AI user that is kind of one step above Claude Chat and Claude Cowork, but, you know, also kind of in that Claude code range where, you know, they're used to these more sophisticated tools. They're kind of past the point of just using regular chat like Claude or Claude Cowork, but they're not quite to the level of like using an agent SDK. They don't want to build their own infrastructure, so they'd rather do it on Anthropics. Am I understanding that correctly? Speaker 1: Yep. This is a classic platform as a service model that you'll see in the industry across. This is an opportunity for those type of users to focus on business value as opposed to focusing on maintaining the infrastructure stack. Speaker 2: Got it. Okay. So it's probably going to be more expensive than running something on like an agent SDK because you're paying for their infrastructure as a service. Agent SDK is going to be a little cheaper, same results, but you've got to maintain your own infrastructure. Speaker 1: Is that right? Yep, you still pay for the token cost through the API. But in addition to that, there are minutes that you're paying for, for being able to run the agent on Anthropic stack. Speaker 2: Got it. Okay, so that's that's much clearer. Now, let's get into what it looks like. So I know you've got some screenshots of Manage agents under the hood. And I know we're going to pop into the actual console and kind of take a look at what that looks like. But why don't you show us? Speaker 1: Yep. So what we're going to do here is I'm going to show these quick three screenshots and then we're going to pop directly into the console. So it's really to answer three questions. What does this actually look like in action? So what we're going to do is I'm going to briefly show you guys just how we utilize the agent SDK and more specifically here, the Claude Managed Agents with our AI tools assessment. We'll briefly give you a quick look at how much a total usage was for running an AI assessment, and then how much it costs to run those. When you come over into the platform, it's literally platform.claude.com, and you'll be dropped into a workspace here on the left-hand side. This is the experience for the persona of someone that is accustomed to chat or co-work, and they'll be spending the majority of their time in the Workbench section and in the Quickstart. What we have here is we already have an agent that is deployed. We've deployed ours with the Ant CLI, and we have a session here of a recent run. Essentially what this does is it gives you the visibility into the full transcript of what we actually had the agent run step-by-step by its tool calls and its agent calls. Then on the right-hand side is the actual finished report of an AI tools assessment, For Kate that we had here as the main item. And as a deliverable, it spits out the document that we're then able to use for the rest of the process that we work with with the client. So this is a quick rundown of how we're able to package it. We have a session which gives us the visibility into what the agent was able to do. And at the end of the day, we're then able to create or add any of the files that we need for this. So this is, even though it's still in beta form, is a very comprehensive solution for a first version of how you'll be able to use this service effectively. Speaker 2: Okay, so when you click into sessions, like what we're looking at here is an actual session, and essentially on the left-hand side is what's going on under the hood. All the tool calls, all the scripts that it's running, all the skills that it's calling, and then on the right-hand side is the output. Is that the right way to look at it? Speaker 1: That's exactly it. Yeah, so then what we do is you'll also have an analytics dashboard for these. So in the analytics, you can see here, this is the usage that we've done to date in the month of April. And when that comes up, it'll show a breakdown of the The amount of tokens that were used both in and out. So you can see here, this was just from running our assessments a couple of times. And then you'll see a cost breakdown, which is ultimately what people care about, right? So we run it on the managed stack for Anthropic. This cost us $2.58 to be able to use the token as a service. And now we can get a realized amount of expense that we're doing for this. So for anyone that is operating with a business, It doesn't want to manage their infrastructure for some of these products where in the case of our assessments where we will actually charge $1,000 for these working with clients. This is a very minimal fee to be able to factor in to have all of the technical details be completely figured out. Speaker 2: Right. That's kind of where I was going to go with this. It's like, one, how much does this cost? And you've kind of broken that down. But two, it really comes down to the use case that you're using a managed agent for, right? If you built a managed agent to just do things for yourself or within your business internally, It's probably going to be pretty expensive and you better make sure that what it's doing inside your business is producing real ROI, whether it be you're making more money, you're saving a ton of time, or you're increasing the quality of your product, because it's probably going to be pretty expensive to run at scale. On the other hand, if you've got a Claude managed agent that you're using to fulfill a service or to fulfill a product in your business, and like you said, with our AI tools assessment, we're charging $1,000 for that deliverable, and it's costing us, according to the screenshot, $2.58 to have a managed agent fulfill that process end-to-end. I mean, that's obviously worth it for us. So I think just for the audience, it's like managed agents is great, But it's expensive and depending on what you're using it for, it might be a good fit for that workflow and it might not be a good fit. You might be better off. Going and using the Claude agent SDK, running something on your own infrastructure to do something that's going to be, you know, very compute intensive because that's going to end up saving you a lot of money over time, right? Or even something like Nick, I would assume like an open Claude agent that you're running on a much cheaper model to do some just internal housekeeping stuff in your business is probably going to be a much more cost efficient option than, you know, using a Claude managed agent for stuff like that. Is that kind of looking at it the right way? Speaker 1: It is, and it really gets us right into the next steps for these, right? So we gave you the clearly explained version. There's more that we can do with this, right? So if you're interested in Claude Managed Agents and want to learn more, both Corey and I both have articles out on X that go into the introduction of it and how you can deploy your first agent. And both of those will have the actual Google Doc that shows you how to get started. Go ahead and play with it a little bit. I think that's the biggest thing is you can do a lot of this testing especially if you're using the CLI to be able to set everything up locally first and then when you deploy it is when you're going to start incurring the credits or the API costs that are associated with the service. Speaker 2: Right. Speaker 1: Yeah. Speaker 2: And you did a great job explaining it very clearly, which was the goal of this video. So for the folks in the audience, again, grab that Google Doc link in description or link in the show notes. If you're listening on podcast platforms, literally download the doc. Hand it to Claude Code or CodeX and it will build and deploy your first managed agent for you to, at the very least, see what it can do and take it for a test drive. And then we will also link both my article and Nick's article that we wrote and published on X going into a much deeper detail. We'll put those in the description and in the show notes as well so that people can dive deeper if they'd like. So Nick, thanks so much for your time and for everybody in the audience. We'll be back next week.