This Is How You Actually Build an AI Agent (Step-by-Step)

AI agents are easily one of the most exciting  parts of AI right now. And they don't just   respond like typical AI chatbots like chat GPT.   They could actually take actions for you. And   they are getting more powerful every single day.   My goal with this video is to explain what AI   agents are in the most simple way, even if you're  completely non-technical. Show you how they work.    Show you what they're used for with practical  examples. And then we're going to go ahead and   build one from scratch, very step by step. Now,  you don't need any coding skills to follow along,   and I'm confident by the end of the video, you'll  not only understand exactly what AI agents are and   what they could do, but you could build your very  own AI agent. Now, let's get into what AI agents

are first. So, AI agents are a computer program  that can actually think and reason and act on   their own to complete a task. They're autonomous,  right? They could actually take steps on their   own. Now, these are totally unlike traditional  computer programs. those have to follow a very   specific set of instructions that usually someone  else wrote for them. AI agents could actually   adapt to new information. They could learn from  experience. They usually have memory and they   could figure out a best way to actually achieve a  goal. Now, what makes AI agents so different and

so groundbreaking? Well, the best way to break  it down is by comparing it to the other tools   that are available right now. So, for example,  let's take chat GPT. Chat GPT will take in a text   prompt your question and it will output an answer,  right? But it actually can really take actions for   you. It's just a back and forth conversation.   Very useful, but not an agent. And then there

is something called automation. And there are  plenty of automation tools that I've used for   many years before AI came out. And those can take  action, but they have to follow a specific set of   predefined rules. they actually can't reason and  they can't figure out the best path forward. They   will literally go step by step on rules that you  have set for them. Those are called automations.    So agents in a way combine all of those things  and more. So they could have access to the power   of models like chat GPT. They could have access  to steps and tools that you typically have in   automation tools and they have access to reasoning  memory and they bring it   all together to actually achieve a goal. Okay, let  me break down how agents actually work in a very

step-by-step way. So the very first step is you  want to give your agent a goal. So for example,   help answer customer service question. That's the  role of the agent. The second step is it will go   and gather information that he has access to. So,  it will look at what's happening in the chat. It   will look at customer messages, any order history,  and it will put all of that together. Now,   step three, it will make decisions based on what  it's learned. It will figure out what the next   step is. It will answer questions. It will ask  for more details if it needs it. It will arrange   a refund. It will actually take that step to  arrange a refund, for example. So, not only can   it reason and understand what it needs to do next,  it will actually be able to do that. And that is   step four. It will actually take the action. An  AI agent does what it thinks is best to hit his   goal and it will literally do those steps to get  there. Now what makes an AI agent? It typically

has three parts to it. The very first part and the  core of an AI agent is the AI model it is using in   the background. So you could use models that chat  GPT uses for example you could use claude Gemini   whatever your favorite large language model is  that could be the model that is powering your AI   agent. I'll show you when we build one how to pick  from a list of available large language models.    Now, the second part that makes an AI agent  useful is the fact that it could have memory. So,   if it remembers past interactions, it will  actually be able to make good decisions,   right? If it's a back and forth and it forgets the  conversation after a few chats, then it's not very   useful. It's not really a good AI agent. And third  are tools. AI agents can use tools to do all kinds

of different things. So, for example, they could  retrieve data from your Google Drive. They could   go ahead and search the web for you. They could  send emails using Gmail for you. And they could   chain actions together to get a lot more done than  just taking one action inside of one app. Now,   one of the ways I actually wrap my head around  AI agents and what they could actually do is   with a free resource that HubSpot created, which I  will link in the description below this video, and   it's a perfect companion to this video. It covers  many of the same core concepts in written format.    So it's really easy to reference later and refresh  your memory. It's a 42page ebook which goes beyond   what I've covered here with section that break  down specific use cases across marketing, sales,   and operation with multiple examples in each  category. Plus, there is a step-by-step guide   on how to build a smart human AI collaboration  strategy for your business along with common   pitfalls to avoid and best practices to follow.   And it comes with a second free download called   How to Use AI Agents in 2025. This one's a  practical checklist you could follow and it   will walk your organization through each phase  of adoption. It's a hand-on tool to make sure   your implementation is smooth, strategic, and  effective. And I think the section about common   pitfalls to avoid is worth mentioning because  I won't have time to get into it in this video,   but it's an important part of understanding how  to work with AI agent and definitely worth a read.    The entire AI agent playbook is free to download  using the link in the description. Thank you for   HubSpot for sponsoring this video and making such  a practical resource available to my audience.

Okay, so what can an AI agent do for you? You  could build simple AI agents or you could get   much more complex and build what are called multi-  aent systems. Okay? And I'm going to give you some   ideas here and then we'll go ahead and build one.   So a very common one that people like to build is   a social media content planner. It will suggest  and schedule post using automation platforms and   chat bots. Now AI agents for customer service  is one of the most common ways people are using   AI agents, but they could actually do a lot of  things for you beyond just answering questions.    So they could obviously answer FAQ questions, but  they usually could solve issues, too. For example,   they could get connected to Stripe and handle  billing questions. A lot of different ways to   build those. One of my favorite AI agents is a  personal email assistant. An agent that could   go categorize and filter all my emails for me,  get rid of all the spam for me, draft replies   to emails that I need to reply to. Really easy to  build that with apps like N8N. You could also have   automated lead management agent, right? So, if  you have a business and you have leads coming in,   you want to track those leads, you want to  send follow-ups in a timely manner, agents   could take care of that for you. You could have  an agent that scrapes and summarizes web pages,   right? So, they could go ahead and browse a web  page, extract all the information and summarize   it for you and make it really organized for you  and easily accessible. AI marketing agents are   great. These could automate all kinds of content  creation, social media post, and they will   incorporate AI to actually create something really  useful and practical. So you're not just posting   noise on these social media platforms. Now, I'll  also link this page in the description below this   video. So if you want to see over a thousand  other examples, there is a platform called N8N,   which I'll introduce you to in a second. But they  have this page right here. If you click on AI,   you'll see all kinds of different examples of what  AI agents could do. And when you click on any of   them here, it will kind of show you how the AI  agents is laid out. And if it's free to use, you   could just click here and it will be added with a  platform called N810, which we'll get into right   now. Okay. Now, let's go ahead and build our very  own AI agent. And the easiest way I've learned how   to build an AI agent is with a platform called  N8N. Now, one of the main reasons I like this   is the visual interface. That's my favorite part.   You could literally drag and drop these things,   move them around, and it's a really nice canvas.   It also has ton of different integrations.    So when you're building these agents, you use  tools. And those tools could be your calendar,   you could be Gmail, it could be things you have  inside of notion, Slack, WhatsApp, any type of   communication tool you use. They typically  have an integration for it. That's very easy   in plugandplay. And there's more advanced things  you could do to actually attach these workflows or   these AI agents to things that are not included,  but they are adding more and more all the time.    So, most likely you will find an integration  that's easy and plug-andplay. And what's great   is they actually have a free trial right now  and it's very affordable. It starts at just $20   a month and give you access to a lot of usage  for that price. And if you're more technical,   this is actually an open-source platform as well.   So, you could technically install it locally on   your computer and run it. I did make a different  video about that is more technical, but the   NHN website is actually really easy to get started  with. I'll put a link in the description to that   as well. Now, let me quickly show you what we're  going to build and then we're going to build it

completely from scratch. I'm going to build an AI  agent that is going to be my personal assistant.    And what it's going to do is it will have access  to all my calendar events. So, it will know all   the meetings I have coming up. It will access to  my email. So, it could actually read the emails   and it will tell it exactly which emails for it to  read. It's going to be just the emails I'm getting   from clients. And then it's going to be able to  send me an email telling me a proposal for a time   for a meeting based on everything I have going on.   Now, this is just one example of an AI agent. So,   if you're not using this exact same one, the  building blocks of this one will be relevant to   a lot of different agents that you'll build. So,  even if you don't build this exact same one, which   I think is going to be very applicable for you,  too. But if you don't build the same one or don't   need the same one, you'll see that a lot of the  building blocks will work through every AI agent   that you'll build. Okay, I just made this full  screen and I'll make a new AI workflow AI agent

here step by step. I'll just create a new one.   And inside of your AI agent, this is what you're   going to see. It says click here to add your first  step. And there's lots of different ways where an   AI agent could start. So a very common one, for  example, is trigger manually. You have to click   on something to start the AI agent from actually  activating. On a schedule is one of my favorites   because maybe every single day at a certain time  you could go ahead and create something. And this   is the one I'm going to use actually for this one.   And another very common one is on a chat message.    So it will actually work like chat GPT. You give  it a message, but it will have everything that   you set up after that message and it will initiate  that way. So for this one here, I could choose on   a schedule like this one right here. And it could  be set based on days, minutes, seconds, weeks,   or months. And then this will be every single day.   And it could run at midnight, for example. Or in   this case, it will make sense to run at 10:00 a. m.   And if I just click away, that schedule is set.

Okay, so that's your very first step. You decide  how you want to trigger it. You could actually   add multiple different triggers. So I'll show  you that in a bit. Then I'm going to press the   plus sign here. Now the next one is we need to  add an AI agent. So if you check for AI agent,   this is the one we want. And let me just click  away just to show you what the AI agent looks   like. The AI agent will have access to a chat  model. That's the AI model that we're going to   give it access to. It will have access to memory.   We could actually change this how much memory it   has access to. Obviously long enough for it to  remember a chat if you were using it for customer   service, for example. And ton of different tools.   And this platform, N8N, has access to hundreds of   different tools that you could tap into pretty  much just about every tool you use will be one   of these integrations here. So, we're going to go  ahead and add those three here. Okay. So, let me   click back on it here because I clicked away. This  is the initial page that you typically get. And   all you have to do over here is you have to decide  how this is going to work. So, source for prompt,   this is your user message. It could be based on a  chat. So if you use the trigger to be a chat like   how a chat GPT conversation starts you could  choose that or you could choose define below   and with define below you just give it a system  prompt that will go over here. So this is how   you want the chatbot to actually behave. This  is kind of an important part when you go ahead   and set these up. So let me go ahead and paste  one over here to show you how that goes. Now,   I'll briefly walk you through these, but one of  the best ways to get these type of prompts is just   to explain what you're trying to do with chat GPT  and ask it to give you a system prompt here that   you could add inside of N8N. And all I did was  I said you are a client status assistant. And I   could even describe more about what it does, but  I described in detail what it could actually have   access to. So these are the different tools like  my calendar, list every meeting, find every free   slot and so on. And then I give it steps. So call  the get event tools and then call the get client   tools. So those are the couple of tools that we're  going to set up in a little bit here. So I just   jumped ahead, but again, chat GPT could help you  come up with something like this. And you could   just manually type in exactly what you're looking  for and have chat GPT actually flush it out. Just   tell chat GPT to give you a better system prompt  from the one you have. And I like to say build   an HTML email with the sections. And this will  all make more sense here as we build it out. So   typically you could come back and add this in  a little bit after you know all the different   tools that you're going to add to it. But it is  a system prompt that is required. Without it,   it's not going to do a good job at all. Okay.   Now, if I click away, the next thing we want

to do is we want to add a chat model. Right? So  the first step was just clicking over here and   setting that prompt for us. Now the chat model, we  have lots of them available to us. So for example,   OpenAI is obviously the most popular one. So you  could choose that one here and then you could   choose the different types of large language model  that comes from OpenAI. So if you want the default   one that chat GPT uses for example, right now  is GPT 40 latest. If five comes out, it might be   that one by the time you watch this video. And if  you want a more cost effective one, the mini one,   this one right here, 40 mini, this is the most  cost effective one. So I'm going to choose that   one right over here. And over here, you'll have  to add your credentials. And this basically lets   you find out where you get something called  an API key. If you're not familiar with this,   this is something you get directly from OpenAI. So  then you could actually use OpenAI's chat models.    This is not your regular chat GPT account that  you need in order to do this. So if you Google   open AAI API, it will usually give you a link  to this page right here. And this page will let   you log in and get an API key. So once you log  in, if you go to the dashboard page right here,   there is something right here on the left side,  API key. And then you will go ahead and create   a new secret key over here. give it a name and it  will create you a key which you will only be able   to see one time and then you will take the key  right here. Copy that key number. Don't show this   to anyone because anyone that has this key will  be able to access your API and the API will need   a credit card attached to it and it will charge  you based on usage and that's why I recommended   a cheap model there. And then you paste that over  here like this and then you could go ahead and   save it from on top. Okay. So, so far inside of  our AI agent, we double cllicked over here. And

inside of our AI agent, we gave it a prompt.   So, that was the first step. The second step   was we added a chat model. Again, there's a lot of  them available. And if I double click over here,   we chose a model from OpenAI, but you could use  Gemini, you could use Claude, you could use Llama,   all kinds of different models you could add here.   The next thing we need to add is a memory. So,   let me press plus here. And this simple memory is  really easy to get started with, but they do have   more complex memory options available. But if you  search for simple memory here, this one's pretty   straightforward. All you have to do is decide how  long of the conversation you want it to remember.    So how many past interactions the model receives  as the context. The context is how much from the   input and output it remembers. By default, it's  set to five. Now, for what I'm using this for,   that's enough. But if you're using it for like a  customer service chatbot, you want something much   higher than that. Maybe 25, right? But in this  case, five is good because I'm using it as an   assistant. And it doesn't need to remember a long  conversation. Now, by the way, from time to time,   it's good to just save this right here. This  will save everything you've done. And you'll   see it says saved. So every time you make a new  adjustment, this turns red. So you could save your   workflow so you don't lose anything. And this is  a nice visual layout here. So you could organize   things nicely that way. Okay. Now this is the  most important part of the whole process. This

becomes an AI agent when it gets access to tools.   So it could actually do things for us, right? So   it has AI accessible here through OpenAI. It  does have memory, but it also needs tools. So   if I press plus on tools, you'll see that this  has access to ton of different tools. These are   just really popular apps that you probably use  already where all your information sits on. And   one of the best ones right now for what I'm using  this for is Google calendar. There is a Google   calendar tool. So if I select this, this will pop  up with this page. Every tool usually has a page   like this where you have to give it access. So the  first thing I want to do is credential to connect   with. So you want to create new credentials over  here. And then you will have to connect this to   your Google account where your calendar is. So I'm  going to do that here. Okay. And it's connected   here to my Google calendar. And that's all I had  to do. So now I could click away here and this is   now my new Google calendar that I just set it up.   Now everything else here could seem a little bit   complicated, but it's pretty straightforward.   It says what calendar do you want to get this   from? So from the list of calendars, I'm going to  choose one that is my calendar here. And when it   comes to a start and end time, a real simple one  is you could start with this fixed option right   here and literally choose a date and time. So  let's say I'll just start it from now. And then   you could set an end date for yourself. And then  again, this could be based on that type of date.    Change it to fixed. And let's say it's a week  from now. Now, this is a very beginner way to   do it. Another way to do it is with these things  called expressions. Now, with expressions, it   gets a little bit more advanced, so I won't really  go into it in this beginner tutorial. But if you   have set it to expressions, if you type in these  double brackets like this, you could do something   like now. Okay? And this is always changing. This  is why it's an expression, not a fixed thing. So,   you could do now. And then over here, if I change  it to expression, it will be now plus one hour   from now. Right? So, this is one hour from now.   Or I could do one week from now for example. So   instead of hours I could put week and this will  become one week from now. Right? It gets more   complicated. So you could start with everything  that is fixed and then over time as you get more   advanced you could learn about expressions and  you could always ask chat GPT to give you N8N   expressions and it will give you these where  you could just type them paste those into these   boxes. And then I'm going to change one last thing  over here for operation. I don't want to create a   calendar. I actually want to get many events. So  retrieves many events from a calendar. I wanted to   check every day what things I have on my calendar.   So get many is what I want. And then everything   else is set up. So just to recap, you connect your  Google calendar here. Then you go ahead and change   this to get many. Choose the right calendar here.   And then down over here, you want to make sure   that you change this to after right now. So that  would be the time based on fixed or expression   looks like exactly like this. And then before  would be one week from now. So that's how I set   that or you could set a fixed date here if you're  not going to want to change this all the time.    And I'm going to rightclick on this and I'm going  to rename this. Okay. I'm going to call this get   events tool. That way when I actually describe  in the system prompt what I want, I could use   this exact name instead of Google calendar tool.   both would work, but this tells me exactly what   I'm trying to do here, and it's easier for me,  especially when you add a lot of different tools   and when you stack agents on top of agents.   I like to just simplify it this way. Now,

we're going to add another tool. This time, we're  going to ask it to go and read our Gmail. So,   the Gmail tool is what we want. And same kind of  thing. We want to link it up to our Gmail. So,   it worked the same way as before. You create  credentials. It already knows my Google. Now   with this one, I do need to change the operation  again. I want to get many. So that will get many   emails from my email inbox. And I have limited to  50 right now because this will work every single   week. But you can increase this limit as well.   And under filters, this is how I actually want   to do it. I want to do it based on label names.   And I'll explain this in a second. Right now,   if I choose label names, it's going to see all the  different labels I have in my Gmail account. So   the one I want is clients. Okay. And I'm going to  set it to that and I could exit out of that. Now,   let me open up my Gmail to show you why that's  working the way I want. Now, what I've done   inside of my Gmail here is I've created a label  which you could create just pressing the plus   sign and creating a label and then selecting any  email and putting it inside of that label. Now,   they also have an automation in Gmail. So, for  example, if you wanted to do it in the settings,   if you go to all settings, they have this options  for filters, and you could actually type in your   client emails here and create a filter and create  an automated rule. So, they all get that client's   email copy. So, they'll just directly go over  here. You don't have to manually do it. So,   that's how I set it up. That's on the Gmail side.   Or you could just manually tag everything with   this client's email and then pick things that  way. So that's why when I go to my Gmail, I had   that label by email. All my client emails directly  get that label here. Right now I'm using a test   account, but that works in my regular account as  well. Now I'm going to go ahead and right click   on this one too and I'm going to rename this get  client email tool so it stays consistent when I   use the system instruction. They all have similar  names. Now, as I mentioned in the beginning, you   could actually have different types of triggers  in the beginning. So, I'm going to go ahead and   click this little section right over here because  I want to add another trigger just to show you   that you could add multiple triggers. So, this is  set to a schedule, but for the sake of this video,   I'm going to trigger this manually. So, I could do  a test, right? So, I'm going to move this up here.    I'm not going to wait till tomorrow. And right  over here, if you drag this, you could connect it   to the AI agent like this, right? So, your trigger  could happen right when you click this, right? And   you could test the workflow that way or you could  trigger based on that schedule or based on chat.    Number of different ways the triggers could start  and you could have multiple stacked on top of each   other. And let me go ahead and save this right  now. Just remember to save every time you add   something. Now I'm going to add one more tool over  here. So this is going to be a Gmail tool again.

So I'll check for the Gmail tool one more time.   But this time I'm going to have it send. So before   we got the emails so it could read the emails, but  now we're going to send an email to ourselves. Now   the subject line again you could set it to fix  so it's always the same thing. You could use   expressions or you could just select this option  right here. Let the model define the parameter so   the AI will decide what the subject line is for  us. And I'm going to do the same thing. The email   type is HTML so it could be formatted nicely.   And I'm going to choose this option again for   the actual message. Okay. So that's the next one.   And again I made sure I linked my Gmail account.    So I'm going to choose that over here too. And  I'll move that down here. And I'll rename this   one over here. Send update message tool. Okay,  now we have everything set up over here. Let me   save it one more time. We could go ahead and test  it, right? So to test it, I could just start with   this one that I added. When you click test, I'm  going to test this flow. And you see it's going to

run through here. And if you get any error message  here, I specifically set one up here so you could   see what it does. If I double click it, it gave me  a message that there is an error in the memory how   I set up the memory. So if I click over here, it  was actually set to connected chat trigger node,   but we didn't have a chat trigger, right? So I  need to say define below over here and just type   in some text here as a key. So for example, I'll  just call this daily briefing. And now I'm going   to click away and I'm going to save it. Now that  should take care of that. So, let's test it one   more time. And now it's going through. The AI  agent is working. Everything is working. I'm   getting the checkbox. Anytime I don't get the  checkbox, I'll get this error message here. So,   I'll go ahead and pause this for now. So, you  could always pause the workflow if you get any   errors. And I'll go ahead and double click to see  what happened here. And it looks like I had the   wrong email here because I did switch my Google  accounts and it was stuck on the old email. So,   I did switch that back. Okay. So, we'll go ahead  and test this one more time from on top. Okay, now   that got the check mark. Everything is getting the  check mark over here. And soon it should send an   email to me. There we go. We got the check mark.   Okay, now this is the email that was sent to me,

right? So, it found the three emails, these three  test emails that we sent to ourselves, which under   the client tab. So, if I go back here, these were  the three emails. It read through those emails.    Then it checked my calendar which is using Google  calendar to find out all the different things   I have scheduled in the next week. So these are  three meetings I already have and then it proposed   a time. This is where it reasons for itself  to make sure it proposes a time that doesn't   interfere with any of the previous meetings I  have. So it had to analyze all the previous chats,   bring itself up to date, then check my calendar  and then reason within itself and then find a time   where I could propose to a client. So now I could  simply just copy and paste this, draft an email,   which I could also set to be part of that AI agent  workflow, but right now I just wanted to be more   as a personal assistant that checks all existing  emails and my calendar and propose a time. And you   see because of that HTML formatting, it gave it  to me in this really nice format over here. Now,

I'm just going to go back here to show you a lot  of that happened over here because of this text   over here, which was my prompt. And it's set to be  user prompt. I'm calling it the system prompt. But   this is the box that it goes under here. And that  is the behavior that you set for your AI agent to   follow. And you could really flush this out again  with the help of chat GPT. And if you just want to   pause it right here, I basically told it what type  of tool it has access to. And then I told it these   are all the different steps I want you to do in  that order. Right? So it's like step one through   five exactly how we describe the same steps to a  personal assistant. Now the agent is just doing   the work of that personal assistant for me. But I  still have to tell it exactly how I want the work   to be done. Okay. Now I'm going to go ahead and  save this one more time. Now this has the schedule   trigger. So this will run on that same time every  single day and I will have that in my inbox every   single day. So I don't have to manually check  all those client messages if they require me   to set up a meeting with them. And remember on  the N8 website if you just click on AI there is

over 1500 AI different automations and they all  work in a similar way where there's an AI agent   in the center of it all and it does all kinds of  different things. So, if you search by popularity,   for example, you could see these are some AI  agent chat bots that I'm making videos about,   different ways to use WhatsApp and other social  media platforms, scraping web pages, all those   different things. You could click on them. And now  that you have a good understanding on how a basic   one is built, you could see a much more complex  one here that even combines multiple agents and   stacks them on top of each other. And a lot of  these are free to use and they will be added   automatically to your account. And if this is the  first time watching any of my videos, we have an

entire platform called Skill Leap. And with Skill  Leap, you get access to over 20 different courses   that are very similar in the teaching style that  I used in this video, but covers all the top AI   tools and all the different AI techniques that  are available right now. And we partner with   other creators to bring you things that I'm  not an expert in. So AI coding, for example,   Kevin from Futurepedia has a great course on using  video tools to create short films and commercials.    And again, me and Andrew and other instructors  have well over 20 different courses and we release   new courses every single month. And you get access  to this under one subscription and it has a free   trial. So, with the free trial, you could watch  a course, make sure it's a good fit, and once   you do subscribe, you get access to all our new  courses, you get access to our community, and you   get access to all our resources, certificate,  and much, much more. I'll put a link in the   description if you want to learn more about that.   Okay, if this video helps you understand AI agents   better, please drop a like and let me know in the  comment what kind of AI agents you're thinking   about building. Thanks so much for watching  this video and I will catch you on the next