Transcription
So, Claudris made it even easier to build and launch agents that work even if you're not by your computer. So, if you've never built an agent before, in this video, I'm going to show you exactly how to build an agent and host it using a new open-source project that Claude just launched. And so, here's essentially what this launch your agent skill does. It will interview you to figure out what you want the agent to do. It will plan out how that agent should look. It will launch it and publish it in the cloud without you having to do anything. And then whenever it creates an output, it will grade and score the output against a rubric. It'll iterate over it over time and then you can have it run on a schedule. It's essentially a live agent or a workflow that you can run entirely autonomously um without you having to do anything.
And so this is essentially the idea of having workflows and then using the latest uh coding techniques that are going around. So everyone's talking about loops and a loop is pretty simple. It basically means that the AI will run and keep running until it has reached a desired goal. A lot of people in the agentic coding sphere, the like programmers are using loops a lot. Even if you don't know how to code, it's still helpful to know what it is because a loop essentially means that rather than just prompting an AI to do one thing, you're saying, "Hey AI, go run and try anything that you can to complete this goal and don't stop until you complete the goal." Right? And so why that's really helpful for agents and why that enables autonomous agents more ones that you don't have to babysit is now we can give an agent that runs like once a day or whenever a certain condition is met. We can say hey go and run and then it will go and it will keep running until it has solved the goal. Um and that way you can just have these agents running in the background.
And so I want to show you exactly what this launch your agent skill is that Enthropic just released. how to set it up inside of Cloud Code, even if you've never used Cloud Code before. Um, and everything that you need to know about it. And so, it's called Launch Your Agent. I'm going to put it in the description down below. And essentially what we're going to do is it's essentially just a skill that runs that helps us build this agent and runs inside of Clad Code. So, how do we actually set this up? So, this is a GitHub link. And what you're going to do is you're going to copy this link from the description below. You're going to go over here into Clad Code. Um, and so if you open up claude, you're it's not going to be cla. So if you're in the cloud desktop, make sure that you click on this code button. It will not work in desktop whatsoever. We're going to click on the code button. We're going to paste it in. Uh, let me actually paste it in a different workspace. Uh, and so what you'll do is you'll go click local. You'll then find any folder on your computer. If you've never used quad code, just create a folder on your documents. Um, you can click add new folder, select it. That's where the COD is going to run called the directory. And then we're going to paste this in. And we're going to say, uh, can you install this skill for me in this workspace? And then I'm just going to click go. And it's going to go and install this for me. And so now you will see it's going to go search the internet for that skill. It's going to clone it and install it into your directory, which basically means that it's going to go over here. Here's all the files. Here's the skill file. Um, here's the user interface. Here's quad.mmd. And it's going to take this, launch your agent in the wrap-up skill. And then it's going to go and essentially just copy all these files onto your computer um so that it can run. And if it asks you, do you need permission? Just give it permission. It's it's a cloud skill, so it's safe.
So now both of these skills are live. And now what you need to do is you need to start a new session in order to access this. So, I'm going to click command N. You can go over here and just click new session. And I'm going to say, uh, let's t test out the, um, launch your agent skill and let's build an agent. And so, that's all you have to do. You just say, hey, let's test it out. It's fine. Now, this skill is in our workspace. So, it's going to run this skill called launch your agent. Um, and we're going it's going to ask us a couple questions about how do we want our agent to be built? What are the conditions? What does good look like? Um, and then once it's done, it will build it for us. It will host it inside of a server on Enthropic so it can run even when we are not managing it with quad code. Um, and so what is something that we could have it do? Let's say we want it to um let's say we want it to pull information from the internet and summarize it for us and give it to us.
So the cool thing about this is you can give it tools. So any of the tools that you have inside of Claude um any of the connectors. So if you go to Claude and add any connectors in there. So I'm going to go here and show you. Um so over here, let me show you this. So inside of quad, you can click customize and you have all the connectors. So any of these connectors that you have in here, you can use inside of this agent. And so you can tell it, hey, can we use this connector? Can we use this tool? Um, so I'm just going to have it use um let's go and check dev news to find any of the new latest uh model releases with AI. What is it about? What are people saying about it? Is it good or bad? Um, and you can just check various different newsletters and magazines and uh blogs related to AI releases. Um, and so I want you to start by scraping that. Um, and then giving me a finalized kind of report of what the output is. And so here it is. Um, and it's going to go ahead and build this for us. And we'll probably want to connect some tools if we wanted to actually send it to Slack or Gmail or wherever we want to access this. But I'm just going to show you kind of how to set it up for now. And this can be helpful. you have if you have clients, let's say you want to build an agent for a client, you can build it, host it on cloud managed agents and then the agent will be basically running for them 247. You're not going to have to, you know, only build it locally on your machine.
So, it's an AI release radar. Um, yeah, let's do like a ranked or prioritized list of what the actual um most important things to know are and what the sentiment is around it, what people are saying about it. Um, this is a recurring thing that's going to happen every single uh morning at 9:00 a.m. And so you'll tell it when you want it to run. This is going to happen every morning at 9:00 a.m. I want you to create a ranked or prioritized list of what you need to know. So here we go. It's creating this ranked list. And now it is going to build it. And so we're going to just let this run. I'm going to pause this video while it runs. It'll probably take a few minutes. Um, and then once it's done, I'm going to show you how you can see the agent running. It's essentially going to be uh on this tool um called platform.claw.com is essentially cloud backend. You'll be able to see this manage agent um and here uh running and what it is.
So, it's going to choose a model for us. It's going to ask us uh questions about it. So I say, okay, I want to know about everything notable in AI and I want this to land I want this to land in my email. Um I want you to pick which uh which newsletters and sources you're pulling from. And so when it asks these asks you these questions, you're essentially telling it what does good look like? because if it if it's not clear on what exactly you want it to do, it's going to just do something that you don't care about. So, it's really helpful to say what is the exact outcome that you want it to do. And the cool thing here is that Claude built this so that it asks you these things and so it's so that it's getting all the information that it needs. So, what must a morning briefing always get right for you? I want um I want it to be skimable is probably the most important thing and ranked. And then do you have an example of a briefing you consider great? Um, no. Just do your best. So the cool thing about this is once it runs, we can then give it feedback and improve it over time. So I can then give it feedback and say, "Hey, I actually don't like how you did this. Can we do this it this way?" And it will update the initial prompt for the agent so that you're getting better outputs over time. So I will come back when this is finished running and I'll show you how to see where the output is, see where it's running and show you how manage agents works inside of the cloud console here. Um, and how you can understand what's actually going on.
Um, actually, uh, it's already came back. That was faster than I thought. So we have the agent. So it's going to call it the AI release radar. It chose the opus model. And so we can actually change that to a different model if we don't need it to be as expensive. Um, so here we go. And then the environment is in the cloud. So the difference between this if you've never used quad code before. So the cloud is the agent is running on a server and local means it's running on our computer. Most of the agents that we build inside of quad code are running on our computer. So the cool thing about this is it will run even if your computer is off. Whereas all the other agents that I built, I'm actively engaging with them and like dialoguing with them and coordinating them. Whereas this completely runs autonomously without me. And so we can give it different tools. So like I showed you before in these connectors, I can give it the Gmail tool, the Google calendar. So any potential skill that you have that you're already using inside of cloud, you can turn that into an agent that runs autonomously, which is really cool. Um, and so here's the outcome. Here's the deliverable. Um, it's going to be deployed, and we haven't set up any other tools or connectors here, but we could if we wanted to. And so here we go. It's found all of the top AI news sites. it laid it out and then it basically says it sends it to my inbox. Um, and it makes sure that um, yeah, so this is the version one. You can see it's already planning out what it's going to do on version two. So if it on version two, here's the memory store of the things that we already covered. And so it's already like building a plan about how to improve itself over time. Um, and so now we go here and it basically wants an API key. So, I should have the API key already in my uh in my workspace. So, it's probably going to set it up automatically. But, if you don't have a Claude API key, then essentially what you're going to do is go over here into platform.cloud.com. You're going to sign into your Cloud account. You're going to go over here to API keys. You're going to click create a key. Maybe call it like my launched agents key or something like that. You're going to click add. It's going to give you a little key. You're going to paste it right in here. Um, and and then Quad will will have the key. And so here we go. It's running. It's going to be launching it on here. Um, I'm not going to create that because I don't want to expose it, but we can see um the AI release radar. So, it just uh basically built this agent overview to show me what it's going to do. So, it's going to deploy every day at 9:00 a.m. American Central time. Um, this is kind of like the dashboard of everything that's happening. It's going to um sweep the source roster of all these different newsletters plus live web for AI developments in the last 24 hours. It's going to rank it by importance. It's going to write an email ready briefing to here. It's going to use these different web tools. It's going to save the memory store somewhere. And if we want to connect any uh connectors and MCP servers, we can. Um, and then what we're going to do is we can see the runs over here. Um, so the the V1, the agent sends the briefing to my inbox. So we have to set up the Gmail MCP server, which is fine. Um, and then here's version two. And then again, this is this is something that is actually really important to understand. The fact that it's clearly defining the exact outcome of what it looks like, that's how the loop functionality works. So, it's basically going to loop over itself and keep running until it has completed an output of this, which is until you have a newsletter that is ranked, that has the sentiment, um, that doesn't include anything from the last 24 hours and that is skimable, then you're not finished. And so, that's the cool thing is like usually you'd give it a prompt and it would run and then it would get like one quarter of the way there and you'd have to keep prompting it. Whereas now it's a fully autonomous agent because I can give it this is the outcome and then it's going to run completely without me. I'm not going to have to click enter, copy and paste any prompt. It's literally just going to run for me um and go in.
And so it says, oh, it says it actually doesn't have my my key yet. So I'll just go ahead and grab it then. So I'm going to go over here. I'm going to grab my key. I'm going to create a new key. Um, I'm going to say delete because I'm going to expose it on the video. Um, launched agents. You're going to copy this key. Um, claude API key. I'm going to paste it in here. It's not the best practice because technically it's this chat is being hosted on a server somewhere inside of Enthropic. But the only way that this would get exposed is if someone hacked your computer or hacked into Enthropic themselves, which is highly unlikely. But a lot of times what you can do is um and then if it if you want to run it with an API key. Oops. Um, if you want to like add an extra layer of protection, what you can do is so essentially you add a couple bucks in here. um if it doesn't work to add funds. So you go ahead and add like a couple dollars and what you can do is basically limit it. And so maybe you add $2 and you say don't auto reload. Auto reload means it automatically spends money when you're out. Just don't turn that on. Um add a couple bucks and then basically that protects you. Whereas if someone were to get your key and use it, they'd be able to use $2 worth of credits. Um, and you would kind of see what the cost is, which is like how much uh credits have been used. If this was maxed out, I would know. And then I would just delete it and then you'd be good. Um, so we can see here, um, it's building it out now. It's la. Oh, it's launched. Okay. So, it's launched. It's live on my account. It gave me this link here, which should go directly into my agents. Yeah. So, you can see in the agents here launched one minute ago. The AI release radar. You can see it's essentially just like a prompt. So I have the prompt you are AI release radar a daily AI news briefing agent. So it's basically just writing the system prompt for me making sure the relevant tools are accessible and then it's running. So we it already started the initial session. And what will happen essentially is it's going to run this prompt um on its own and the agent is going to keep going until it reads the output. And so every morning at 9:00 a.m. this agent is just going to run that prompt. It's going to search the internet. It's going to find it. It's going to, you know, figure out what it needs. It's going to send an email to me with the output. Um, it's going to do it without me having to go in to claw.com and say, "Hey, can you find me the news?" Press enter, wait five minutes for it to come back. Um, and so we can see here, here's the outcome, and you can kind of see what it's doing. So, you see this little blue agent, and then the tool. So the tool is the obviously like the tools that it's using and then the agent is kind of like the the prompt or the message. So like if I went over here to Claude, this would be like the agent. This would be the tool, right? So this is the agent. Here's the tool is the things that it's doing. And then the agent is like the responses. The tool is what's what it's using. So you can actually go ahead and see the log of what's happening. If this looks technical, like I urge you to just like try to understand what's going on because you probably will understand more than you think, right? Because the agent I will start by checking the date and then begin searches. Then I'm using web search. Even if you don't know what bash means, that's okay. I'm using web search. Um, and then the result is this. Uh, it found um all of this information about what's happening in AI. Then it said, "Here's the date." And then it just keeps finding more information, saying, "Okay, I found out about this. Let me research this. I found out about this. Let me research this." Um, and it's still running. Uh, did it finish? Actually, I think it finished. Um, it doesn't show that the time is increasing, but it stopped at a web search, which is weird. Um, so it should Oh, it says it's still running. Okay, so this is a first session. Says it's still running. Um, and so whenever this finishes, we will have the output here, which will be the actual like email that it sends. And if I want to update it, I could go over here to manage agents. I could go back to my chat actually and update it from there. I can also go just directly in here and try to edit it. Um, I would not recommend that.
And so what I'm going to do is can we update it to make sure that it is actually able to send an email to me when it's done with the output. Uh how can we add that MCPN connector into this workflow? So I'm going to ask it here and it's going to help me set up that connector. So I'm going to wait for this to run. So essentially what it did here is it found us a tool. So it said you could use these different tools. the Gmail. Uh, we could have you sign in. You could use these tools. It found one called uh resend, which I've never actually heard before, but it looks like a way for AI to send emails. So, what I'm going to do is just click get started. I'm going to uh go ahead and sign in here, and then we're just going to grab it an API key, and then it's going to know what to do with it. So, I literally just created an account. I grabbed an API key here. This is the first thing that it sent me. I'm going to say here's the key. And then um it's going to go ahead and set all this up. So you don't need to be a developer. You don't need to understand what any of this means. It's going to go ahead and send this for us. Um, and now once this is done, it's going to connect this email so that it actually lands in my email. Now if I go back to the managed agents as well, we should see um this run. Okay, the output here. So, it ran uh it got the grading. Um, it said it was not fully met. So, it's like continuing to improve itself until it's satisfied. You can see it ran. It said, "Okay, we finished it." Okay, but it doesn't meet the desired outcome. So, let's run it again until we do. And it took two iterations, which is which is not too many. And we can see uh the output. So, if I go here on the output, this is the markdown file. So markdown, you can kind of just think of it like text that has a couple symbols in it to make it look different. So if I go over here to um claude, actually let me go over here to this workspace and then boom. Uh, and I'm just going to paste this in here. And I'm just going to say uh can you turn this into a markdown file in this workspace? I just want to show you what the output is. Uh, but basically you can see OpenAI unveiled Jalapeno, their custom inference chip. That's a weird name. Um, Enthropic accuses Alibaba of a distillation attack. Qualcomm, Google bakes computer use into Gemini. Capital keeps flowing into the application layer. Um, and so these are all the things that we got from that. So this is going to show up in a markdown file while this is building itself out. Um, can you uh test it out now and actually send the email? Uh, also try just logging me into Gmail and see if we can just log directly into Gmail as well because um I don't want to pay for credits for this app that I've never used. Okay.
And so now we should see this. All right. So AI release radar. Here we go. Oops. Um, right there. So this is what it looks like. Here was the first output. So the AI went researched the web, found all this data. Here was the first output. Release radar June 25th. Uh Open AI, Anthropic, all the things I just mentioned. So this is what's going to show up in my email. Um, and now because I said let's use Gmail, what's going to happen is it's going to try to set up Gmail in here and then it's basically just going to ask me to log into my Gmail account um so that it can um run this. And so boom, using the recent account, it actually looked like it worked. So, I'm actually going to go and open my emails and see if we received it. So, it didn't really format it correctly, but we did in fact receive the email. So, we have uh resend.dev is the email. AI release radar. It needs to format it better. Um, so I basically would just go in here, copy this in. Um, uh, let's just do this. Here was the final output of the email. You can see that it's not formatted correctly. So, I want to make sure that when it gets sent, it actually gets formatted. So, can you update this? And basically, we're building the agent. We're going through a couple iterations of refining what it is so that it can then run fully autonomously without us for as long as it wants and then we never have to actually like engage with it from there on out. And so the idea is that it takes a little bit of work upfront to build out this agent, but then it just runs in the background and it runs forever without us needing to do anything about it. And so the idea is we could have like more and more agents over time that we're building. Whether you're building this for a client, you can just have them running for them uh so that they can use them whenever if we're building them for ourselves and then we're building a team of agents where we might have agents inside of cloud code that we're working with every day. And then we have these in the background which are essentially you can think of them like workflows more than agents. Um, and so any like type of workflow that you might have created um on like make.com or NADN, these can now just be managed agents so that it runs in the background. Whenever this happens, it does this and it connects to this tool. And so here you're just connecting to tools um and it's automating things for you. Um whereas now you're just telling AI, hey, I want you to build me this thing. Can you do it? And then it's figuring out how to connect all the tools itself and to build everything itself. and then it's just running in the background without you having to think about it. So, it's a pretty huge level up. Uh, and so hope this video gave you a deeper understanding of how Launch Your Agent works, why this is beneficial, um, and so on and so forth. And if you want to understand quad code more deeply, if you want to like take these questions to me, uh, you can head to the link in the description. Uh, essentially it is where we run live calls. So, we have live calls going. We have the troubleshooting and systems call where you bring your questions about cloud code to me and I'll answer them for you on a call. We have like complete cloud code course, all the plug-and-play systems, everything is in here. Um, everything you've seen in this video will be in there. Um, but I don't want to uh speak more about it. So, would love to see you in there. Would love to see you on a call. Otherwise, thank you so much for watching this video. So, I hope it was valuable and I'll see you in the next.