How To Build AI Agents [Full Course 2025]

The AI gold rush is in full swing and is showing no signs of stopping in 2025 and beyond. And the hottest skill set that you can learn inside of AI, even if you're a beginner who don't know how to write one single line of code, is building AI agents. And in this complete tutorial, I'm going to break down what are AI agents and how you can build one even if you are a complete beginner. I'm also going to show you some real-world examples and use cases and show you how to price it if this is an additional service that you want to add inside of your AI agency. And if you're new to my channel, hi, my name is Helena, and on this channel I talk about everything AI and automation. So if you want to stay up to date with the latest AI technologies, please make sure you like and subscribe. I also have a free AI automations course that you can sign up for. The link will be below as well. All right, so recently I listened to this Y Combinator podcast, and what they said was that the AI agent market is going to be way bigger than the SaaS And this is because for every SaaS tool that exists, there will be a corresponding AI agent. So AAAS, or agent as a service, is going to be a much bigger market than the SaaS market. And you can get started now by not knowing even how to write one single line of code, which is super exciting because anyone can now participate in this market. So let me give you an example of what they mean. So let's suppose that you bought LinkedIn Premium, which is a SaaS that gives you access to everyone that's available and has a profile on LinkedIn so that you can reach out to people in your target market. Now, having this tool by itself is not going to yield you business results. You actually have to use it, reach out to your prospects, engage with them, etc., to make the most out of this tool. So what if you can have an AI agent that you can delegate all of these tasks to? So it's no longer enough for business owners to just have tools or SaaS. They also want AI agents that will be able to operate and use these tools to get the most efficiency and productivity for their business. Now, if you think AI agents is a buzz term that's going to go away, think again. When we look at the Google Trends, we can see that it is a term that has been exploding in the last 18 months. So what exactly are AI agents? Now, before we get into the definition, we have to talk about what AI agents are not. Now, if I just went to ChatGPT and let's suppose I type something like, write me an essay about the War 1812. That is not an agentic workflow because it's executing everything in one go. You are getting the task completed in one iteration without any adjustment, revisions, or refinements. Now conversely, if I was writing this essay using an agentic workflow, I will have multiple specialized AI agents that's going to write a piece of this essay. So I can have a research AI agent, then I'm going to have a different AI agent to write the outline, another AI agent to write the actual essay and another AI agent to edit the essay. And they can work together in a feedback loop until the best essay is produced. So these AI agent teams actually function similarly to human teams, the same way that human teams have specialized team members. So for example, if I was in a marketing department, there's going to be a videographer, there's going to be an SEO blog writer, there's going to be a graphic designer, etc. The AI agents are also specialized in the task that what they do. This is how you get the most results out of these AI agent teams. There are also horizontal AI agents as well as vertical AI agents. Horizontal AI agents have a broad range of skills, whereas vertical AI agents are very specialized in terms of one task for one industry. And I also want to point out that AI agents are not the same as workflows. Workflows are what you can build inside of Make.com or Zapier. In fact, Anthropic actually has a very good definition here. Workflows are systems where LLMs and tools are orchestrated through predefined code paths, and agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks. Now, if we take a look at the search volume for make.com in blue here, which is my favorite platform to build these workflows, and we compare that to N8n, which is the leading platform to build AI agents, we can see that it has been overtaking Make.com by a significant margin here, and that shows you the potential of AI agents in today's economy. Now let's take a closer look at how workflows differ from agents. So an example of this could be a customer email came in, and this LLM would determine if this email is about sales or is it about customer service. And depending on what's said in this email, it's going to route this to the appropriate OpenAI assistant, which is going to produce the output. Here's another example of a workflow. You have an input that goes in, and then it's going to go into various different LLM calls, and all of the results will then be aggregated and you will get your output. An example of this could be social media. Let's suppose that you want to repurpose a blog post into multiple social media posts, the first LLM could be a Facebook post writer, followed by a LinkedIn post writer, followed by an Instagram post writer. And all of these posts will be aggregated together and emailed out to you for approval. Those are 2 examples of workflows. But I just want you to know that there are literally endless amount of combinations that you can combine LLMs and tools and APIs together to make infinite amount of possibilities. Now let's take a look at what an AI agent looks like and how it differs from a workflow. So in this example, a human is interacting with this autonomous agent. Depending on what the human says, the LLM will take the appropriate action, and you can define the skill set that you want to give to this AI agent. I'm going to show you how to add in all those tools in just a moment here. Then you're going to have some sort of feedback. The most important distinction here is that the AI agent is able to pull out the skill set or the tool that you have given it and use it appropriately depending on what inquiry is. And the other important distinction is this feedback loop. All right. So with an autonomous agent, there's always an action as well as a feedback loop. And as humans, we can interact with these autonomous agents through some sort of interface. This could be a chatbot, a voicebot, a button, or a form on a website. Now, the use cases for AI agents are endless, but let me give you a few examples to get your creativity juices rolling so that you can see how you can apply this technology inside of your own business. You can use these AI agents as a personal assistant to help you book appointments, manage your calendar, send out meeting invites, and keep track of your deadlines. In marketing, you can use AI agents for blogging, creating social media posts, translating materials into different languages, and creating personalized outreaches to every single prospect. In sales, you can use AI agents to automatically follow up with your customers. For example, if you have a new lead that just came in from a Facebook lead ad, you can have an AI agent that calls and texts that customer instantly. And in customer service, you can have specialized AI agents that answer routine customer service inquiries, product questions, and triage their tickets to the right department. Now, if building AI agents sounds exciting to you and you want to start offering this as a service that you can start making money from, there are 3 different pricing models that you can employ. The first one is a flat fee, so you would charge the business owner a flat fee to the AI agent for them and they own the agent afterwards. This pricing could be anywhere from $2,000 to $10,000 depending on the complexity of the bot. You could also charge them a maintenance package afterwards to ensure that the bot is functioning as expected. The second pricing model is to rent these AI agents. So you could create a framework for these AI agents and customize it to the client that you're working with, and you can rent it to them for a few hundred dollars every single month. And if they stop paying, then they stop getting the agent. This lowers the barrier to entry for them, but they don't own the bot. And if you already created one of these AI agents, it's not very difficult to clone it and make the necessary changes so that it is customized for that specific business owner. Now, lastly, the third pricing model is to pay for usage. So for example, if this is a voice AI agent, then you can charge them for every single the minute that this AI agent speaks to their customer on the phone. Now, if you're using the third pricing model, be sure to understand what all of your costs are as well as your API costs and charge a big margin on top of that so that you don't go under. All right, so now I want to walk you through how you can build your very own AI agent, and I'm going to show you how you can do this step by step using a no-code tool called N8N, which is super powerful. So once we're logged in here, we just go up here and click on Create Workflow. What I'm going to be showing you how to build today is an AI email agent, but I just want you to note how to use the platform and the nuances of it so that you can build any kind of AI agent that your heart desires. So once we land on the dashboard here, the first step is to define what the trigger is. This is what's going to activate our AI agent. So if you've been building automations on Make and Zapier a lot, you will actually see a lot of similarities here. When I click on Add first step, I can choose the different triggers available. I can choose the trigger to be a specific date or time or something that happens inside of one of these apps. Now for my example, I'm going to choose Gmail. So whenever I have an email coming in, this AI email assistant is going to help me reply to that email. The trigger is going to be on message received, and if this is your first time connecting to any of the modules, you will need to give N8n permission to be able to access your account. I have already done so, but if this is your first time logging in, click on Create New Credentials and simply sign in with your Google account. And after you have done so, you can then choose the poll time. So how often do you want this AI agent to be activated? You can choose 1 minute, 1 hour, 1 day, etc. I'm going to choose 1 hour for this example. And once I'm done with all of these parameters, I can simply click on Go Back to the Canvas, and now I have my trigger configured. The next step is to create our AI agent. So I'm going to press the plus button here, and I can see all of the different actions I have available. Again, similar to Make and Zapier, but what differentiates N8N is the advanced AI capabilities. So when I click on this, I can see that I can create AI agents. I can extract information, I can create sentiment analysis, I can create summarization chains, I can create text classifiers. So these are things that you can quote unquote hack on platforms like Make or Zapier, but N8n just makes it so much more simpler. I'm going to select AI agent, and here I'm going to select tool agent from this dropdown. And for the source of prompt dropdown, I'm going to select define below. In the text here, this is where I'm going to put in prompt or instructions that I'm going to give to this AI agent. I can also create a system prompt. So the system prompt is what's going to be executed for every single time that this AI agent is going to run, and the text is only going to be specific for this specific run. All right, I have filled in my prompt here, and it says you are a helpful email assistant to Helena Liu, and your job is to reply to the user's email in a friendly and professional tone. If the user is asking to book an appointment, be sure to check my calendar. If the user's email is in another language, please translate it to English. You want to write the emails as Helena. So this AI agent is going to write my emails for me. And in the text here, I wrote, here's the user's email, and I filled in what that email will be. And I'm able to do so because when I ran the Gmail trigger and this API returned the information that I need, I can select the appropriate parameter that I want to drag and drop into my prompt here. Okay, so after I have filled in all my parameters for my AI agent, I can also go over here to settings to see if there's any additional configurations that I want to make. In this case, I do not. So let me go back to the canvas. Now that I have the base of my AI agent configured, now I need to attach it to a chat model, memory, and the different tools that I want this AI agent to have access to. For the chat model, here are all the ones available to me. So I can use DeepSeek, which is super hot right now. I can also use Anthropic, or I can use OpenRouter, or I can use OpenAI. For our example here, I'm going to use OpenAI chat model. And to get your account created, you just need to put in your OpenAI API key. To get your API key, you can can go to platform.openai.com/api-key. And you want to treat your API key really as your password. This is not something that you want to share with others because it's connected to your credit card. The more usage that there is on the API key, the more that you have to pay. Now, once you have your account connected to your OpenAI account, you will need to choose the model that you want to use. In our example, I'm going to choose 4o Mini, and there's no other parameters that I need to configure here. Once I have my OpenAI model created, I can also choose if I want this AI agent to be connected to a memory or tools. A memory will just allow this AI agent to remember the past interactions you had with it. So by default, you can use the window buffer memory. It is not as powerful, but there's no other configurations that you need to make. You can just drag and drop this window buffer memory into your AI agent. Now once you select this, you want to actually go down on the session ID and select Define Below and give this a memory key. For the memory key in this example, I think it's appropriate to use the Gmail ID of this email because this is what allows us to track that this is the same interaction, the same email that we're replying to. So by default, the memory is going to be set to 5, which means that this AI agent will remember up to the last 5 interactions. You can increase this, but it only works to a certain amount because this is the window buffer memory. If you do need more memory, then you will need to purchase one of these additional memory storage that is available here. Now for our example, 5 is perfectly fine. So now I just linked the window buffer memory back to the memory of our AI agent. All right, so the last thing we need to configure here is the tools. This is the quote-unquote skill sets that we want our AI agent to have. Now if I press the plus button here, we can see all of the various tools that we can connect to our AI agent. We can also search here at the top. I want this AI agent to be able to access my calendar and tell me when I have availability, so I'm going to select the Google Calendar tool. And in the parameters, I can define how I want this AI agent to use my Google Calendar. So I can select whether I want it to set automatically or manually— automatically is fine— and I can select whether I want this AI agent to be able to create events, get events, delete events. So for this example, let's suppose I just want this AI agent to get my next 10 appointments in the next week. So I set it up to get many, limit being 10, and from now to a week from now. Now there's no limit to how many tools I can add inside of my AI agent. So let's suppose that I also want to have this AI agent be able to read my emails. I can simply search for the Gmail tool and link it to my account, give it the specific operation that I want it to do. In this case, let's say is to read my last 50 emails. I can just configure that and add that to the tools inside of my AI agents. Just to show you the scope of what these AI agents can do, you can make custom API calls. So any task that can be done on Zapier or Make, these AI agents can now operate. You can get it to update something in your Asana, in your ClickUp. You can get it to calculate things using the calculator tool. You can get it to update things in your GitHub, in your Gmail, to, I mean, the possibilities here are just endless as I'm scrolling through this. As long as the tool has an API, you can now give this skillset to your AI agent. And if I was doing this linearly, like in a workflow, I would have to say, go here, go here, go here. But by giving this AI agent tools it can use based on what you wrote in the prompt here, the AI agent is able to decide what is the best tool it can use to accomplish the task that you want. Another really great feature is adding humans in the loop. So after this AI agent writes the email, I want it to actually send it to me for approval. And so to do that, I'm going to press the plus button again. I'm going to select humans in the loop, and I can decide whether I want this AI agent to send me a message on my Gmail email, Slack, Discord, or Telegram. I'm going to select Telegram for this example. Now, if this is your first time linking to your Telegram account, you actually have to message the BotFather account to get your API key. I have already done so. And now inside of my parameters for my Telegram, I can put in the original email as well as the response that the AI has written. Okay, I have filled in my Telegram parameters with the email sender, what the original email said, as well as what the AI reply will look like. And again, these The green parts that you see are outputs from previous modules that I can drag into the parameters of this module that I can use to build on top of my agentic workflow. So for example, if I want the output that just came out from the AI agent module, I can just drag and drop this output onto my canvas here. Now I have it twice, so I'm going to erase it once. And when it's green, you know it's working. When it's red, you know that there's some sort of error. Now once this is complete, I can go up here and click on test step to ensure that everything is working. And when I opened up my Telegram, which is named Helena's Assistant, I can see that this is just an email from my GoHighLevel account telling me that one of the subaccounts have been paused. And we can see that the AI has replied here, thank you for notifying us that there's been an account closure, looking forward to your guidance. So this is not bad. And you can see here that there's actually a button underneath this message that's approved. And we get that from the response type. It is by default selected to approve, and I can actually add 2 options here to approve and disapprove. So when I approve the email, I just want it to be sent out immediately. And if I click decline, I want nothing to happen. So the AI agent is not going to send that email, and I'm going to take a look at it at my own time. Okay, we know that we have configured our Telegram module correctly because there's now a green check mark here. Now the last step here is to add in an if-else router. So if we approve the message, it's going to send that reply to the user, otherwise it's going to do nothing. Alright, so let's press the plus button again, search for the if-else router. And what we want to do in this case is under the Telegram, we can see under data there's approved or not approved. So we're going to drag over the approved here. And this is going to be a Boolean That is true. So if it's approved, then we want to send the reply. If it's not, we're going to do nothing. Now let's click back to Canvas again. We can see now there's 2 routes under the if-else router. So if it's true, we're going to press the plus sign and we're going to select Gmail again. And it tells you all of the actions you can do here. In this case, we want to reply to a message. In this account. And let's just get the message ID by going to the Gmail trigger and we can see the ID right here. And for the email type, I'm gonna select text. And what we're going to reply with is the output that the AI agent has written. And that's it. So let's test out this step and see if it has a green check mark. So it's working properly and that is it. So just within the last 15 minutes or so, we have built a complete AI email agent. Isn't that like incredible, guys? So let's go ahead and save everything. So N8N does not save things for you automatically. And now let's actually try to test the entire thing and send us an email. So in this email example here, I wrote an email to myself and I'm just asking myself, what is your next 3 upcoming availabilities for our follow-up meeting? And I'm going to press on test workflow and send out this email to have this entire automation run. Give this a whirl. You can see that it is now running, and we can see that this AI agent is actually choosing this calendar tool because of what I said in my email. And again, like, there's so many tools that you can add to this. I can add in calculators, I can add in Google Analytics and spreadsheets, and list just goes on. And we can see now that we are in the Telegram module. And if I open up my Telegram, we can see the email sender, myself, the email content. This is what we wrote. And this is the AI reply here. Thank you for your email. Here's my next 3 availabilities, because it's looked in my calendar and actually determined this. Really cool, right? And if I hit approve, it's going to pop this up after I approved, it's actually going to send out the email. So now let me go back in my email and show you what this looks like. And you can see the AI has replied to this email for me with what I have just approved here. I hope you're just as mind-blown and geeking out about this as I am. This is like one of the biggest tech breakthroughs in AI that we have seen in a very long time. I hope you'll give this a whirl and try to build your very own AI agent. Let me know in the comments below what kind of AI agents you are planning to take. And before you go, please make sure you sign up for my free AI automations course below as well. And if you got any value out of today's training, please make sure you like and subscribe and I'll see you next time. Thanks so much for joining me today.