Transcription
Anthropic just launched Claude Managed Agents. And if you're thinking this is just another release that you don't have to pay attention to, I want to break down in this video exactly why this is one of the biggest releases from Anthropic for any business or knowledge worker building with AI, especially if you've been exploring things like OpenClaw and building custom agents for your own business.
And in this video, I want to break down what Claude Managed Agents actually is, how to start using it today, and why this is now the easiest way to build agents for your business, for your work than almost any other platform.
If you don't know me, my name is Chris, and for the last 15 years, I've been designing apps and advising startups on product and design. And I've taught hundreds of people in my community how to build apps, agents, and more using AI tools. So, if you're interested in doing that for your business or for your work, then you can click the link in the description down below to find out more.
Otherwise, let's get straight into Claude Managed Agents and break down exactly how it works and how you can start building your first agent.
So, this is Claude Managed Agents inside the Claude platform. So, if you go to platform.cloud.com, you can access Claude Managed Agents today. And Claude Managed Agents is not just another way of using Claude. It's a full agent building platform inside of the Claude platform here. And if you've already been using things like Claude Code, Claude Co-work, and you're struggling to figure out how do I actually build an agent? What is an agent? How do I get it running and doing a task? I think what Anthropic have done here is basically pull together a user interface and an experience that makes complete sense for businesses for any knowledge worker to get started with building agents straight away without having to do anything in the terminal. Build your own custom agent setup all of that sort of stuff. Basically, Claude Managed Agents guides you through the entire process. And I think this is going to be the easiest way for so many businesses and knowledge workers to build agents with AI that are actually useful and can help you in your day-to-day work.
And usually I just talk about building applications and software with AI, but I want to talk about Claude Managed Agents a little bit here because building agents alongside applications and AI systems is really the way that you're going to bring a lot of value to your business or to your work. It's not just about building software or building applications. It's about combining all of these things together to create something that is going to create a massive amount of value for yourself and also for your business and just help you create a ton of stuff using AI.
So, let me show you exactly how this works. Even if you're non-technical, this is really super easy to set up and get started with. I'm going to break down all the key concepts, how to create your first agent, and how to get it up and running.
So, when you go onto the Claude platform here, the first page you're going to go to is the quick start page on the managed agents tab here. And it's got this little new badge here so you can see where to get to it. And basically, there's a few different concepts here to understand how agents work. And I also want to break down the use cases for building these agents as well, so you get an idea of the sorts of things that you can use them for.
Now, one thing to know about Claude Managed Agents is that it does use your Claude API. So, it's not running off of your Claude subscription, but if you run a business and you're serious about building agents, then running it from the API shouldn't be an issue. Essentially, you want to set a budget for all of your agents and then manage that budget directly through the Claude platform here just by checking your usage here from your API. You can see you can see all of the usage from your Claude API and how each of your agents is actually contributing to your budget that you have when running these agents.
So now that I've covered that, let's go back to the agent section here. So Claude Managed Agents are broken down into these four separate things. First, you have the agent. And an agent is basically a set of instructions alongside MCPs and tools and skills that the agent has access to in order to do its job. The best way to think about an agent is about is like an AI employee, but an employee that only has one very specific task to do inside of your business. So almost like a mini employee basically. And the more tightly scoped your agents are and specific you are about the jobs that they have to do and the tools that they have access to, the better the outcomes that you're going to get with these agents are.
The next concept in Managed Agents is sessions. And this is basically like a chat with an agent or like a task that it would be completing. So you can think of this as very similar to the agent chat that you're familiar with when you use Claude Code or Claude Co-work or any other AI chat tool.
The third concept here is environments, which is basically the environment that you create for your agent to run in. Now, what is incredibly powerful about Claude Managed Agents is that these agents are running in Anthropic's own cloud infrastructure. So, you don't have to deploy your own agents. You don't have to figure out the tech. It's all deployed on Claude's cloud infrastructure so that you can get up and running with agents very, very quickly. And basically the environment is like the sandbox, the environment that the agent operates within and the different networks and packages that the agent has access to. So you can see this agent here, my YouTube content strategist that I've just set up quickly has full internet access in its environment.
And then the last concept here to understand with Claude Managed Agents is the credential vaults. And basically credential vaults are a way of storing secure credentials, things like API keys, login information, all of that sort of stuff in a secure vault that an agent has access to. So I'm going to show you in a second how to create an agent as access to the Notion tool to access some information on my Notion account. And I'm going to create a new vault with a Notion API key inside so that it can access the information on my Notion account.
So those are the four key concepts with Claude Managed Agents. And you get a bunch of templates here with Claude Managed Agents to get you started. So support agent is an example here. So answers customer questions from your docs and knowledge base and escalates when needed. And you can think about Claude Managed Agents as almost the next step in these automated AI workflows that you that a lot of people were building with things like n8n where you chain together all these different steps except the agent is just going through doing those steps by itself and actually figuring out a lot of different paths to accomplish the task instead of just a linear flow which goes through a certain set of nodes. And that's what makes building agents like this incredibly powerful. And this is why so many people building custom agents before this were able to do such incredible things. And Claude Managed Agents just really wraps the whole thing up in an incredibly easy to use package that I think is going to see a huge amount of adoption from businesses building with this tool.
So now that I've broken down Claude Managed Agents, let's actually create a managed agent here. And you can do this from the quick start page here just to chat to describe the agent that you want to create. Let's create an agent that pulls video ideas from a Notion database that I have in my Notion account and then researches title formats that are working for competitors in my niche and then gives me a brief script for each video idea along with five title ideas for that video concept.
So, this is an example of an agent that is just designed to complete a single specific type of task that's going to save me a bunch of manual time doing this myself every single week and maintaining this document. And the one thing that I can kind of go from here is to then iterate on this agent over time so that it starts to understand my business, my context better, what titles I like and don't like, get it to create documents that it can reference in order to give me better titles, better scripts for this task. And you can do this sort of thing for any small task across any number of different tools that you can connect to using any skills or MCP servers that you might want to access as well. And if you're just getting started with this, I would definitely think about setting up a simple basic task like this first and then iterating on your agents over time. And you can even go to Claude and do an audit of your business, an audit of all the processes that you're doing to figure out what sort of things you can build agents for if you're not sure where to get started.
So let's send this to Claude Managed Agents here and get it to build our agent for us. So you can see on the right hand side here it's creating the definition for this agent. So the name is YouTube Content Planner. It's got a description. It's got a system prompt here as well. MCP servers. It's going to set up the Notion server tools. It's got access to the Notion um the Notion tool here as well. So let's just go now. If we're happy with that, we can just say create this agent. So let's create this agent here. And then Claude Managed Agents is going to create this agent for us.
The next thing we have to do is configure the environment. So we're going to click configure environment here. Says an environment is a cloud container template that defines the compute workspace your agents tools execute in things like running bash commands, reading files or calling MCP servers. So would you like to reuse an existing environment or create a new one? So let's just reuse an existing environment that we've got here. Actually I'm going to click create a new one so you can see what it looks like. So let's create a new environment and we're just going to say unrestricted for this and you can describe the type of environment that you want to create for your agent here as well. Let's click unrestricted. That's going to create an environment for our agent.
And next we're going to start a session to test our agent. So it says before we launch this agent needs Notion credentials so it can connect to your Notion workspace. A vault is a workspace level store for MCP server credentials. It holds your OAuth tokens so sessions can authenticate with Notion without you re-entering them every time. And one of the great things about using Claude Managed Agents is that instead of using something like OpenCore, which is way less secure and way more risky to use, you're just managing agents in the cloud to complete tasks for you without having to set all of this stuff up manually yourself.
So let's go to credential vaults. Let's create a new vault here. And I'm going to call this YouTube Set Credentials. Let's create that vault. We go into that vault and we're going to add a credential. We're going to say this is the Notion MCP server and I'm just going to click this drop down and then search for Notion. We can click on Notion here. It says this credential will be shared across this workspace. And what we need is basically an access token and our so we can put these in then we can click add credential and you have to click I acknowledge this. So I'm going to go over to Notion and get these credentials quickly.
So I actually didn't need the um access token and the credentials there. I've just added the MCP server. When you click add it takes you through logging into your Notion account here. So that is now connected the credentials there.
So let's go over to sessions and create a new session here. Call this weekly content outlines. I'm going to select the agent that I want to use which is the content planner. We're going to go to the environment which is the content planner environment. Select the credentials which is the YouTube content credentials. I'm going to I'm going to tick I own or I'm also authorized to use this vault. We're then going to click create session here.
What I'm going to ask this agent to do is can you look at the next video in the content calendar? This should be a video about the consensus MCP server. And can you research some title idea directions for this and give me a brief overview of the script for this video based on the information on that card.
So let's send this to the agent now. This agent should now be able to go into Notion, have a look at this content calendar database in my Notion account. And just to show you what the content calendar looks like here, I can give you an idea. This is my content calendar here. So this next one that I'm doing is about building apps with real data using the consensus MCP. I've got a rough outline here of what I want to do, but I don't have a script for it. I don't have any title ideas. So let's use the agent to actually add that to my Notion page so that I can basically whenever I have a video idea or every week on a Monday I can run this agent to research the best title approaches based on my competitors in this niche and then create a brief script for me so that I can go into filming it knowing exactly what the title is going to be for this video and what I need to cover in the video as well. And this isn't going to be perfect the first time, but the idea is that you refine these over time based on how they complete the task. It's about iterating the same way that you would train an employee, the same way that you would update SOPs for an employee, the same way that you would improve your internal processes in a business.
So, let's see what this agent comes back with here. You can see here in the tool use that it's gone through. It says, "Great. I've found the content calendar. Let me fetch it and look up more about the Consensus MTP server at the same time." So, it's found that video idea page in Notion and it's then going and researching information around that around the consensus MTP around different title ideas and different titles that other competitors are using on my for my YouTube channel and then giving me information to actually help me write a script here. And you can see this last task here has actually done it. It didn't actually just give me a message saying great, I've finished. And so, I wasn't sure whether it had finished there, but it's actually written this idea out here. So hook open with a provocative question. What if your app could cite 220 million peer-reviewed research papers as a source of as its source of truth? Most building builders are slapping generic AI onto their products but plugging into real cited scientific data is an underused mode. Show a quick demo of Claude pulling a research paper result in seconds via the consensus MTP. So key points here and then we've got some title ideas here. So I used 220 million research papers to find my next app idea. Your app should be citing science. Here's how. Stop guessing app ideas. Let 200 million research papers tell you what to build. These are all actually really great titles here. And it says, "Quick note on priority. Title number three is probably the strongest for cold audiences, nonsubscribers. So, stop getting app ideas. Let 200 million research papers tell you what to build."
And if I go over to Notion, it hasn't actually added this to the card in Notion, but let's just ask it to add that information to the page in Notion. Can you add this information to the page in Notion? And there it's just said done. I've updated the Notion page with the full script outline and all five title ideas. So, let's go back over to Notion. And I can see here that it's added the script outline, key points into Notion for me, title ideas, as well as the recommended pick. This is now all inside of my Notion database here for this video idea. And this is the exact same way that you can build agents with the Claude Managed Agents platform for any process, any task in your business that can be automated this way. and even to like get you way better results than if you were just researching this and looking at competitor title ideas yourself. This is way way better. And this gives you all of the tools that you need with an agent to perform these tasks from web access to running its own tools to using MCP servers and skills and connecting to all of the tools you already use in your business.
And that and hopefully that overview can help you get started with Claude Managed Agents. So you can see in that brief demo the power of the Claude Managed Agents platform and building agents this way just makes so much more sense for so many businesses, so many people working with AI. You don't have to worry about building sub agents in Claude folders using Claude Code. You don't have to worry about building agents on any of these other agent building platforms. That's confusing and complicated. And you don't have to worry about running your own infrastructure for managing agents either. You're just running agents in Anthropic's cloud environment, getting them to do real tasks, connect to your actual tools, use any agent skill available and can connect to a bunch of different things through MCP servers. I think this is just an absolute no-brainer for so many businesses and knowledge workers building agents with AI and it's super easy to get started with as well.
And if you are building with AI and you want to build software, agents, and systems for your business or for your work that actually work, I've got a community helping people do just that over at school.com/ai-apps where I've taught hundreds of people how to build real applications, real software, and agents using AI tools. And so, if you're interested in finding out more, just click the link in the description down below. Otherwise, thank you for watching. I hope you enjoyed the video. Don't forget to like and subscribe and I will see you next.