Build Your First AI Agent in 10 Minutes — No Coding
Publicado no IA Dots ·
Agentes de IAAI Signalidioma-enInglêsyt-7ReHiROdPbE
Vídeo em Inglês, publicado pelo canal The Tech Girl.
Video transcript
Transcript from available captions; it may contain errors.
AI agents are everywhere right now, but I think we have made them sound way more complicated than they actually are. When someone says build an AI agent, you probably imagine coding APIs, automation tools, and one of those giant workflows with 20 different boxes connected together. But your first AI agent doesn't need any of [music] that. In this video, I'm going to show you one of the easiest way to build a genuinely useful AI agent using Claude. We're basically going to describe what we want in normal English, give the AI a job, test the result, improve it, and then make it run automatically and no coding at all. Before we build anything, let me quickly explain what an AI agent actually is. Normally, when you use AI chatbot, you ask it a question. It gives you an answer, and that interaction is basically finished. An AI agent is slightly different because instead of giving the AI one question, we give it a job or an outcome that we want it to achieve. The AI can then work through the steps required to complete that job. Depending on the task, it might search the internet, read multiple sources, compare information, work with files, use connected apps, organize everything, and then give you the final result. For our first agent, I want to keep things really simple, but still build something that I would actually use. We're going to create a daily AI research agent. So, every morning, I want this agent to search the web and find the five most important AI stories from the previous 24 hours. I don't want every tiny funding announcements or random AI update. I want it to figure out which stories are actually important, give me a short summary of each one, explain why each story matters, and include links to the original sources. Then, once we know that our agent is doing a good job, we'll make it run automatically every morning and then save the finished report straight into Google Drive. And obviously, you can replace this example with almost anything you want. You could create a stock research agent, a competitor research agent, a job search agent, a travel search agent, a content idea agent, or even an agent that checks information from your connected apps and tells you what needs your attention. But for your first one, I would highly recommend choosing one very clear job instead of trying to build an agent that does [music] everything. So now I'm going to build this using Claude co-work. If you have used Claude before, you're probably familiar with normal chat where you have a back and forth conversation with Claude. Co-work is designed more around giving Claude larger tasks and letting it actually work through them for you. So open Claude, preferably on your desktop, switch from chat to co-work near the message box, and now we essentially have our workspace for building this agent. The first thing we need to do is explain the job clearly, and this is where people sometimes make agents unnecessarily complicated. You don't need to write some massive technical prompt, just describe what you want the finished result to look like. For example, I could say, "Research the five most important artificial intelligence stories from the last 24 hours. Prioritize major AI product launches, new AI models, important research, and significant announcements from companies like OpenAI, Anthropic, Google, Apple, Meta, and Microsoft. For each story, give me a short summary, explain why it matters, and include links to reliable original sources. Avoid duplicate stories and minor announcements. That's already enough to give Claude a very clear job. Did you notice that I mainly described the outcome? I'm not telling Claude, "First search the website, then open this website, compare these three articles, then create a document. Claude can figure out many of these intermediate steps itself. Now, we let Claude run the research. It has the goal, it understands the output I want, and it can work through the intermediate steps necessary to get there. >> [snorts] >> Before we move on, I want to quickly thank HubSpot for sponsoring this part of the video. If you enjoyed building this agent and you want to experiment with more advanced ones, HubSpot partnered with Futurepedia to create a completely free build your first AI agent course. And this is actually a really useful next step after what we just built. There are 18 hands-on videos where they walk you through building different AI agents using tools like N8N. And it's not just explaining what an AI agent is, you're actually building things. For example, you will learn how to connect an agent to Gmail, Google Calendar, and Google Sheets, build a personal assistant, create an agent that turns YouTube scripts into LinkedIn posts, and short-form scripts, and even build a customer support agent trained on a website. They also give you the actual system prompts and explain the really simple brain, memory, and tools frameworks that you can reuse when building your own agents. My favorite part is that it's very hands-on. You're not just learning the theory behind AI agents, you're actually building useful agents step-by-step and connecting them to tools that you already use like Gmail, Google Calendar, and Google Sheets. The whole course is free, so I'll leave the HubSpot link below if you want to check it out. Now, let's get back to the video. Once Claude finishes, look at the results. Your first version probably won't be perfect, and that's completely normal. You can improve the agent. I can say, "Keep each story under 100 words. I can revise further and customize it with all my requirements. Now, the agent becomes much more useful because it understands the actual workflow that I'm trying to accomplish. The next thing to understand is tools. Agents become much more powerful when they have access to tools, but again, this sounds more technical than it really is. A tool is simply something the AI is allowed to use. Web search is a tool. Google Drive can be a tool. Gmail can be a tool. Your calendar can be a tool. Different services and connectors can give an agent access to different information or actions. For example, if you connect Google Drive, you could create agent that works with information stored inside your drive, rather than requiring you to manually copy and paste everything into Claude. If you connect an email service, you could build workflows around finding messages that need your attention, summarizing certain emails, or helping you prepare responses. For our AI news agent, though, we honestly don't need many tools. Web research is enough, and I would actually recommend keeping your first agent this simple. Don't connect 10 different apps just because you can. The more access you give an agent, the more complicated the workflow becomes. Start by asking, what does this agent actually need to complete its job, and only give it those tools. At this point, our AI research agent already works, but there is one problem. I have to remember to come into Claude and ask it to do the research every morning. So, the final step is to automate that part. Inside Claude co-work, go to the schedule section, and create a new schedule task. I'm going to call mine daily AI news brief. Then, I can use essentially the same instructions we just tested, but I'm going to add one more thing. I've connected Google Drive, so I'll tell Claude, "Every morning, create this AI news brief and save the finished report to my daily AI news briefs folder in Google Drive with the date in the file name. Now, to run this task every day automatically, I can again simply say run this workflow every day morning at 8:00 a.m. automatically as a scheduled task. So, every day Claude can do the research automatically, create the report, and save a new dated copy directly into my Google Drive. I don't even have to remember to come back and run it myself. You might think couldn't I just paste this exact prompt into normal Claude every morning? Absolutely, you could. The difference is that then you're still responsible for starting and managing that process every single morning. Here, we have given Claude the goal. We have given it instructions about what a good result looks like. It can use tools like web research to gather the information. It can work through multiple steps to complete the task. And then, we have told it to repeat that task on a schedule. That's really the easiest mental model for an AI agent. It has a goal. It has instructions. It has tools. And it can work through multiple steps to complete the job. And if the task happens repeatedly, automation makes it even more useful. One thing to keep in mind before connecting everything, be careful on how much control you give an agent. There is a big difference between letting it research or summarize information and letting it edit files, send messages, delete data, or take actions inside your accounts. Start with low-risk tasks like research planning organization and drafting. For anything that can make important changes or communicate on your behalf, keep a review step in place until you're confident the workflow is reliable. And that's really all you need to build your first AI agent. Let us know in the comments some of the interesting agents that you have built or going to build. I'll catch you in the next one. This is Sanjana. Bye-bye.
Assistir ao vídeo original no YouTube
Assuntos: Agentes de IA. Publicado no YouTube em 2026-09-05.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.