00:00:00So, the Claude team have released Claude managed agents, which is what they call the next evolution
00:00:09after the agent SDK.
00:00:11This is something that lets you create custom agents without writing a single line of code.
00:00:16And these agents run on Anthropics own managed infrastructure, which has a very cool architecture
00:00:22that is perfect for shipping secure, scalable, production ready agents that can do pretty
00:00:27much anything you want.
00:00:28Also answer customer queries based on your knowledge base, or even do regular research
00:00:32for you and send it to you at a specific time using Slack.
00:00:36But why would anyone use this instead of installing Claude code on their own VPS?
00:00:40Hit subscribe and let's get into it.
00:00:46But before we get into it, let's talk a bit about Anthropic because they've been busy
00:00:49these last few weeks.
00:00:50I mean, they've recently prevented people from using the Claude subscription on third
00:00:55party tools like OpenClaw, which people think is because of managed agents, but they're
00:01:00not exactly the same thing.
00:01:01I mean, I would say OpenClaw is a bit like Linux.
00:01:04It's a tinkerer's agent.
00:01:06You pick your own hardware, you pick your own model, you deal with security and everything
00:01:11in between.
00:01:12Whereas Claude managed agents is, dare I say it like Apple, because you don't need to
00:01:17do any of those things.
00:01:19Claude takes care of the infrastructure, the security, you just tell it what you want in
00:01:24simple English, and it will go ahead and build your perfect agent.
00:01:28In fact, let me show you how easy it is by going through a simple yet very personal example.
00:01:34I have a private GitHub repo that contains all my medical information scraped from the
00:01:39NHS app.
00:01:40And I want to communicate with that data or get information from that data using Slack.
00:01:45So I can use it from my desktop, my mobile, basically anywhere I am.
00:01:49In between those two things, I would like a Claude managed agent to do the jobs of scraping
00:01:54the data, so using tools to get the right information and translating it into a way that I can understand
00:02:01it.
00:02:02So to get started, I could go to the Claude console, go to the new managed agents option
00:02:07and type here in natural language to create my agent from scratch.
00:02:11Now this will communicate with the Claude API using curl commands and will host any necessary
00:02:16code on Anthropix infrastructure.
00:02:19But I could also use the managed agent skill in whatever language I prefer.
00:02:23In this case, I will pick TypeScript and this skill will use the TypeScript to Claude SDK
00:02:29to create an agent for me.
00:02:31To do that, you'll need to have this version of Claude's code or higher, which has the built
00:02:36in managed agent skill.
00:02:38So in my case, I have a version higher than that and can trigger the skill using the slash
00:02:42Claude API command followed by managed agents onboarding.
00:02:46So after I hit enter, it asks me if I know what kind of agent I want to build, which
00:02:50I don't, but we'll see what it does.
00:02:52And it tells me it will walk me through these three steps.
00:02:55So tools, skills, files, and repos, and then environment and identity.
00:02:59Now surprisingly, it does use a lot of context, so there might be some compaction, but we'll
00:03:04see what happens.
00:03:05Anyway, I'll give it a command to create a medical agent that reads markdown files from
00:03:10a private GitHub repo, understands the information like a doctor and lets me communicate with
00:03:15it using Slack.
00:03:17And then it recommends me to use the pre-built tool set, so read glob grep and to not use
00:03:22write edit bash, since the doctor has no reason to mutate the repo, which makes sense.
00:03:27It also asks for the repo URL, so I'm going to tell it to go with its recommended tools
00:03:31and permissions, and I'll give it a link to the repo.
00:03:35Then after that, it suggests what round B and round C should be, which since this is a very
00:03:39basic agent, they're pretty self-explanatory.
00:03:41And then it creates a system prompt for my agent, as well as suggest the model it should
00:03:45use.
00:03:46Now I'm going to tell it to use sonnet because I don't want to spend too much money on Opus.
00:03:49I'll explain why later, but aside from that, this looks good and I'll provide it with these
00:03:53credentials as well as the language I want.
00:03:56And now it's created the two files for me in TypeScript.
00:03:59The first one is the setup, which will set up things like the environment, the agent,
00:04:04and any necessary skills inside Anthropix infrastructure.
00:04:07The second is the runtime, which is actually going to be the thing that communicates with
00:04:12the Anthropix servers and gives that information to Slack.
00:04:15So I'll go ahead and set this all up and show you what it's like when it's finished.
00:04:19So after running this setup file, it gives me an environment ID, which is over here.
00:04:24And it also gives me the agent ID, which is also over here.
00:04:27Now as I mentioned earlier, these things are created on Anthropix infrastructure.
00:04:32So inside the cloud console, I can see my agent over here, as well as the environment I've
00:04:36just created.
00:04:37I've also created my Slack app and have put all the information inside my .mv file for
00:04:42this apps.ts file to use, which means if I run that file, it should run my Slack bot.
00:04:48So I could ask it, what model are you using?
00:04:50And after a while it responds saying, I'm Claude made by Anthropix.
00:04:54Is there something medical I can help you with?
00:04:56This is very cool.
00:04:57But what's even cooler is that I can see the session here in the cloud console.
00:05:01Yes, I've been testing this a few times and here we get more details of what happened.
00:05:04So if I close this to make it bigger, you can see the question the user asked.
00:05:09Then it used the Slack message tool and then the agent responded.
00:05:12Now I forgot to mention earlier how the pricing works for this.
00:05:15So if we go to the documentation, we can see that all tokens used by the managed agents
00:05:20are priced with the pricing model that the Claude API uses, which is over here.
00:05:25So unfortunately your pro max or team subscription isn't useful for managed agents, but as well
00:05:30as paying for tokens, you'll also have to pay for sessions, which is 8 cents per session
00:05:35hour.
00:05:36And this is only when the session is running.
00:05:38So if I go back to the cloud console and click on sessions for all of these idle sessions,
00:05:42I'm not being charged.
00:05:43Okay, let's do something a bit more interesting with this bot.
00:05:46I'm going to ask it based on what you know about me medically, is it okay for me to eat
00:05:50calamari?
00:05:51Now here it's gone ahead and ran the bash tool to get information from the repo.
00:05:56It's done two file reads, and then a few seconds later, it sent me a Slack message, which tells
00:06:01me I should be cautious with calamari because I'm allergic to shrimp, which is true.
00:06:06I would say it's given too much information talking about my itchy tongue to swelling throat
00:06:11and so on.
00:06:12But to be honest, it works really well.
00:06:13In fact, I did tweak the agent a bit.
00:06:15So if we click here in the cloud console, we can see there are three versions indicating
00:06:19that this agent has been changed three times.
00:06:21I changed the system prompt to make it sound a bit more human-like and change the model
00:06:26from Opus to Sonnet.
00:06:27But here in the UI, I can change the agent system prompt, I could change the model and
00:06:32the tools it has access to, which is useful for testing.
00:06:35So that's pretty much it.
00:06:36Apart from a few small code tweaks that I figured out by going back and forth with Claude, that
00:06:41is how easy it is to build an agent.
00:06:44No need to learn how the Claude agent SDK works.
00:06:47You can just communicate with Claude using a skill and create your very own agent.
00:06:51But how does all of this work under the hood?
00:06:54So Anthropic have written a very detailed article on how everything works, which I'll have a
00:06:58link for in the description.
00:07:00But I would say the whole architecture is made up of three key components.
00:07:05So the session harness and orchestration, this is not to say the sandbox and tools are less
00:07:09important, but I would say these three are very unique.
00:07:12So the harness also known as the hands of the system is where the Claude model is used.
00:07:17And it's also known as a stateless router because it routes tool calls, resources and MCPs to
00:07:23their relevant place or runs code and edits files in a sandbox environment.
00:07:28Now I'll talk about the benefits of having tool calls separate from the actual harness
00:07:32itself.
00:07:33But the session here is like the memory of the system and contains append only logs of
00:07:37the harness.
00:07:38Now you may be thinking the harness is Claude code, but it's actually a custom made harness
00:07:43for the managed agents.
00:07:44I'll also explain why a bit later on.
00:07:47And finally, the orchestrator is what decides what modes the harness should be in.
00:07:51So build plan and so on.
00:07:54And importantly, it creates a new harness if this one fails.
00:07:58So imagine you have a harness here that fails or goes down.
00:08:02The orchestrator can create a new harness.
00:08:04And because the session logs are separate from the harness itself, the new harness can read
00:08:09the logs to find out what's been going on and exactly where to continue from.
00:08:13In fact, the whole thing is built for scale.
00:08:15So you can have as many models and as many environments as you want, and the architecture
00:08:20will be able to handle it.
00:08:21Also another benefit of this architecture is the security aspect.
00:08:25So if I go back to the Claude console and click on the credentials vault, you can see that
00:08:30the credentials are stored in a secure location.
00:08:32Now, if I have an agent running locally, that could be the .m file or something custom I
00:08:36have in place.
00:08:37But if I'm using the Claude console or the UI, then all of the credentials are stored
00:08:41here.
00:08:42And the beauty of that is these credentials are called at runtime.
00:08:45So if you have an MCP specific API key or tool specific key, then the harness or the model
00:08:50doesn't know anything about it and it can't have access to it.
00:08:54So, for example, if I needed to call the weather MCP tool and I had that API key, then the harness
00:09:00will call the tool and the API key will exist within the tool call itself or the MCP.
00:09:05And it will be used at runtime.
00:09:07Similarly, if the sandbox needed to use a key, then that will be stored in the vault.
00:09:12And that will also be used at runtime and the model wouldn't know about it.
00:09:15In fact, it doesn't even know about its own anthropic API key, since that is also used
00:09:20at runtime.
00:09:21I highly recommend reading the rest of this article to get a detailed overview of how the
00:09:25whole thing was put together, but it is very unique.
00:09:28Honestly, I really enjoy creating Claude managed agents.
00:09:31I mean, there are people out there who think this is going to die, just like the open AI
00:09:36agents.
00:09:37If open AI agents aren't dead, please let me know in the comments because I don't hear
00:09:39much about them.
00:09:40But I think this is going to stick around for a long time purely because it's very easy to
00:09:45create an agent.
00:09:46You don't have to learn about SDKs, you don't have to use the terminal to create one if you
00:09:51don't want to, and you can pretty much do all the hooking up to MCP servers tools all
00:09:56inside the UI, which is very nice for people who don't want to write any code.
00:10:01But I do have to say one kind of disappointing thing is the fact that you have to pay using
00:10:06API tokens or pay using API billing, because as someone who is a Claude Pro subscriber,
00:10:12I would love to use my limits, so the limits that are within the pro range on managed agents,
00:10:17so I don't have to pay for two different things.
00:10:20But in all fairness, it's not insanely expensive if you use a cheaper model like Sonnet or Haiku.
00:10:26And even though it's kind of curated, so as you can see, it gives you access to Notion,
00:10:31Slack MCP servers and so on, but if you want to create something that doesn't exist within
00:10:36those bounds, then you'd have to go ahead and write your own code, which OpenClaw is
00:10:40very good for.
00:10:41I mean, OpenClaw is super open, it's in the name and has many channels from Telegram to
00:10:46Discord to WhatsApp, whereas if you want to do the same thing with managed agents, you
00:10:51may have to code up your own solution or get Claude to code one up for you.