Daniel from the Founder Institute builds a working AI agent in under ten minutes using the Command platform. He connects Exa search and OpenAI to create an agent that gathers startup funding news, then shows how detailed prompts and guardrails reduce errors and improve output quality.
What if you could build an AI agent in just 10 minutes? Well, guess what? You actually can. Daniel here from the Founder Institute, and today I'm going to show you how to build an AI agent using this new tool called Command. I know what you're thinking. What the heck is even an AI agent? I think the best way to describe an AI agent is an AI that can complete tasks without needing step-by-step instructions and at the same time use tools to connect data and information. If you're someone with a great idea and you want to turn it into a product, we have a Wipe Coding Bootcamp coming up. I highly recommend checking that out.
Okay, to kick things off, you'll head on over to command.new. And in command, you'll see that it's pretty much like any other Wipe Coding tool in that you have a text box where you'll enter in your prompt. And then you have a couple of suggestions for prompts. So for my prompt today, I'm going to go super basic. What I want is an agent that collects and organizes startup funding news from across the world. As somebody who works with startups, I always want to be on top of everything that's happening. So I feel like this agent would be super useful for me. So what we'll do next is hit the arrow button. So like with a lot of other live coding tools, you can actually see Command building out your agent in real time. There are a couple of things to look at over here. You'll see that there are a few API keys that will be required for this particular use case. And the thing with Command is according to your use case and the output that you're looking for, you'll obviously have to connect a bunch of different tools. So Command will actually recommend these tools to you. You can of course change them within the prompts. For example, if I don't want to use OpenAI, I want to use a different one, maybe Gemini, I can let the agent know. Okay, so right before you are actually given your final product, you will have to hit deploy. And when you hit deploy, you'll be asked for two API keys. And for our use case, again, it's Exa and OpenAI. Exa is a great tool for search. So to actually get these API keys, you'll head on over to their website. For Exa, it's dashboard.exa.ai/api-keys. From here, you'll head on over to create key, create a key. I actually already created mine because I don't want everyone to see my API keys here. And then with OpenAI, you'll pretty much do the same thing. Platform.openai.com, create a new secret key, copy that, and then move that over to command over here. And then you hit confirm and deploy. Now that our agent is ready, let's test it out. I want news on AI startups. So then I'll click collect news and then we'll wait for it to give us our results. And there we go. We have our result. While the content is perfectly fine, I think it could be way better organized. And for that, I honestly recommend going back into to when you're initially coming up with a prompt to have those guardrails in terms of like the type of content that you want to see and way it should be organized. And if you go back into our chat with the agent here, you can see that it actually had to work through a couple of errors as well. And I think for the sake of preserving credits, you want to maybe have more details in your prompt. And that's definitely a mistake that I made. But yeah, I think for the purpose of what we want to do, this looks pretty, pretty good. And one One of the really cool things that I like about Command is if you head on over to Agent Flow right on your screen, you can actually see the workflow that the agent follows in giving you the result. And if we zoom in a little bit here and we can actually see it completely. So it begins with the workflow, it looks at various sources and then it filters down what it needs to actually display. And after all of that, you essentially end up with the actual output that we received. And I think this is also a great great way for you to maybe think more deeply about the tasks that you want to accomplish with an agent. And for that reason alone, I think just the insight that you're receiving makes it a definite worthwhile investment.
I hope you found this video useful. And if there is another tool or use case you want us to test out, let us know in the comment section. And if you're confused about whether you should be wipe coding an agent or just using ChatGPT, I honestly recommend going for what works for you the fastest. And sometimes that's ChatGPT and sometimes that's an agent. Do check out our Wipe Coding Bootcamp. I'll see you in the next one.
