Claude Code NEW Sub Agents in 7 Minutes

Anthropic has just released sub agents for cloud code. What the feature allows you to do is essentially create your own team of specialized AI agents for different aspects of your development

workflow. You can equip each agent that

workflow. You can equip each agent that operates independently with its own context and expertise as well as their respective MCP tools or system prompt.

This lets you delegate specific workflows like code reviews, debugging or testing to dedicated AI specialists that you defined while keeping the conversation focused on the big picture.

Now, in this video, I'm going to show you how you can set up sub agents as well as how you can think about potentially leveraging within building out your application. Let's dive in.

Once you're within Cloud Code, all that you need to do to create new agents is you can forward slash agents. Once

you're within there, you can create a new agent. You can specify whether the

new agent. You can specify whether the agent is going to live within the project or whether it's going to be globally on your machine where you can access it within any project. And then

from there you can either manually configure or generate with claude the agent that you want to create. In this

case I'm going to say I want to create a front-end engineer that is an expert in nex.js tailwind as well as shadui. What

it's going to do is it's going to generate a markdown file that we can go and we can iterate on as we build out that agent. And the really great thing

that agent. And the really great thing with this is we can specify which tools that we want to give the agent access to. We have all of the core cloud code

to. We have all of the core cloud code functions like being able to list out the different files within a directory, read or edit files, all of the different features that you're probably familiar with. Additionally, if you do have any

with. Additionally, if you do have any MCP servers, we are going to be able to leverage those within the agent. And

within here, you can go and turn on or off all of the different whether it's MCP tools or the core functionality that you want to give the agent access to.

Now, for this particular agent, I'm going to give it access to everything.

Where this is helpful is you can specify to have a backend agent that might have access to read your database and run SQL commands and potentially read the logs of your infrastructure or whatever it

might be. Whereas on the front end,

might be. Whereas on the front end, maybe it has access to a whole host of different tools. And additionally, what

different tools. And additionally, what you can do is let's say you're using a brand new front-end framework or something and the documentation isn't in the training data of the LLM. You can

specify it to leverage different MCP tools to actually retrieve the relevant information. Within here, I'm going to

information. Within here, I'm going to specify the color of the agent. So, we

have our Nex.js Tailwind Shad CN UI engineer. And we can also change the

engineer. And we can also change the title as well as anything within this markdown. The structure in terms of the

markdown. The structure in terms of the markdown, we have the name, we have the location where it's going to be relative to our directory. We have all of the different tools that will be listed out here. They're going to be comma

here. They're going to be comma separated or if we're leveraging all the tools, we'll just say all tools. Then we

have the description in terms of how cloud code will actually invoke this agent. We have use this agent when you

agent. We have use this agent when you need to build, modify, debug front-end components and applications using Nex.js, Tailwind, so on and so forth.

And then finally, the system prompts.

This is where it's really powerful because you'll be able to add to this over time the different biases of what things you like to leverage or not like to leverage. If there are certain quirks

to leverage. If there are certain quirks different models are doing that you don't particularly like, like maybe you don't like the linear gradients or the thick fonts or all of the different emojis, you can specify all of those

different aspects within the system prompt. And just to show you what it

prompt. And just to show you what it looks like, we're going to have a hidden directory called.cloud. And then within

directory called.cloud. And then within that directory, we're going to have our agents. This is going to be where all of

agents. This is going to be where all of that relevant context for each agent that you set up live. Okay. So now I'm going to go ahead and I'm going to create a new research agent. Again, I'm

just going to specify this within the root of our directory. I'm going to say you are an expert web researcher that's going to be able to search the web.

You're great at googling different things, but you're also really good at finding the relevant information on particular web pages. And then for this agent, I'm going to specify to give it access to scrape different web pages,

map the different URLs from web pages, and then finally also search different queries from the web. And then finally, just to show you the markdown file, we have the name within here. We have again the location. And this is going to be

the location. And this is going to be what it looks like when you break up all of those different tools. And the nice thing with Claude is you can also even specify to Claude to say, "Hey, update this even within just the chat

interface." You can do that directly

interface." You can do that directly from Claude and it will be able to find that agents directory and make the relevant updates that you're asking.

Now, I'm going to say I want to build out a SAS landing page. I want a pricing table as well as an FAQ. But

specifically, what I want to have in the main area of the website is I want to research the top five AI news stories from July 2025. I think first before we

actually start building out the website, let's just research the web and find the relevant details. I'm going to go ahead

relevant details. I'm going to go ahead and kick off that task. Now, the first thing that Claude is going to do is it's going to break down what we asked for into a to-do list. It's going to research all of the different stories.

Then here we can see the query that it's generated for us. It gathered all of the relevant research. We see great, I've

relevant research. We see great, I've got the research complete, and now we're moving on to the next.js process. And

what's really great with this is we're going to be able to define project specific agents. We can have global

specific agents. We can have global agents that are just helpful for things we use across different projects. We're

going to be able to spawn off a ton of different agents. And I think this is

different agents. And I think this is going to be interesting over the coming weeks just to see how different people leverage this and the creative use cases in terms of actually creating almost like a mini workforce of the different

things that we want to accomplish within Claude. Now, another interesting use

Claude. Now, another interesting use case for cloud code is you could set this up to access a ton of different MCP servers that aren't necessarily even coding related. Say for instance, if you

coding related. Say for instance, if you want to hook it up to Gmail or linear or a Canva MCP, you're going to be able to have a ton of different capabilities.

And instead of trying to squeeze in within just one system prompt all of the different facets of how your application works and what to do and not to do and all of that, instead of that, what we

can do is we can just give very poignant instructions to each of these respective agents. We can tell it what to do, what

agents. We can tell it what to do, what not to do, and we can just engineer that context to be very relevant for what we want each of those agents to do. And

what's really great with how the agents are set up is they're within markdown files. You're going to be able to take

files. You're going to be able to take that context. They're very portable. you

that context. They're very portable. you

can commit them to your repo. It's going

to be really interesting to see how different people leverage these over the coming weeks. So here's the starting

coming weeks. So here's the starting point of the landing page. So we have a simple Nex.js application. We have some pricing here. We have our FAQs that are

pricing here. We have our FAQs that are functional. But additionally on the

functional. But additionally on the outset of our application, we ask for it to search for up-to-date information.

Within here, we see five different relevant AI stories. Meta is 300 million talent war things like GBT5 is coming next month. All of those different

next month. All of those different pieces were leveraged from actually searching the internet and using that as context to feed in to what ultimately built out this application for us. I

just wanted to do a really quick one showing you how you can leverage cloud code. It is super exciting just to think

code. It is super exciting just to think about how we can leverage these different agents, especially now that we can equip it with different MCP tools and specify the relevant context for all of those different agents. Let me know

the overall structure of the different agents that you're going to be setting up within Claude Code. If you found this video useful, please comment, share, and subscribe.