Transcription
My name is David Andre, and in this video, I'm going to teach you how to build AI agents, even if you're a complete beginner. The popularity of AI agents is absolutely exploding. It's clear that 2025 will be the year of AI agents, but just like with the internet and Bitcoin, most people are going to miss out. So don't be one of them! Watch until the end and start building AI agents.
Real quick, I'm hiring a developer to help me build VL, so if you have full-stack experience and want to join the team, the link is going to be below the video.
All right, to get started with n8n, you have to go to n8n.io and click on "Get Started for Free." Once you log in, you'll see the dashboard. Just click on "Open," and in here, you'll see all of your workflows. As you can see, I have one workflow here, but I'm going to click on "Create New Workflow."
Now, you can do this even if you've never coded before because n8n is a no-code tool. So you don't have to be a programmer at all. Seriously, anyone can do this.
Okay, so the first step is choosing the trigger. This is perhaps the most important part of the automation because it determines what starts the automation. When you click on this, you can see multiple different triggers we have. It could be a manual trigger, an app event, a schedule—for example, every day, every hour—or a webhook call. There are so many options!
Think of n8n as make.com on steroids. In fact, here's how n8n's growth compares to make.com. As you can see, it's growing exponentially, and in the past week, it has been searched three times more than make.com.
Now that you know how to set up n8n, let's start by creating a Telegram trigger because I think that's one of the most popular triggers. Again, you can choose anything else, right? You can choose WhatsApp, Google Sheets, Slack messages—any automation you want to make, you can make it with n8n. Let's just choose Telegram, and we're going to do "On Message."
Don't get overwhelmed by this UI. Once you understand it, it's actually very simple. On the left, it's what's coming in; the middle is setting up this current block, and on the right, we see what the data looks like after this block runs.
So we need to connect our Telegram account. If you've never done this, click on "Create New Credential," and in here, we need the access token. Once you have your Telegram account created, the thing we have to do is chat with the BotFather. So click on this and click on "Open Telegram."
The BotFather is basically a way to interact and manage bots. Let's just follow the official tutorial from the n8n docs. Let's type in "new bot."
Boom! All right, the new bot—how are you going to call it? I'm going to call it "Google AI Agent," and you'll see why in a bit. Good! Now choose a username for this, so I'm going to name it "New Society Google Bot." Boom! There it is.
Now we have this link. If we click on that and click on "Start," we can chat with this new bot. But what we actually need is the access token right here, so copy that. Do not share this with anybody; obviously, this is private. I'm going to change it before uploading the video.
Once you copy this, let's go back into n8n and paste it in right here. Click on "Save," and now let's wait one second. We should see a popup: "Connection tested successfully." Beautiful! We can go back, and by the way, anytime you do this, I recommend you click on "Save" so you don't lose your progress.
Now we have the Telegram trigger, and let's actually test it out, right? Let's go test step. It's listening for a test event, so if we click on—if we switch back to Telegram, I'm going to say "Hello, world!" Boom! There it is.
So it correctly found our message, and we can see our text right here: "Hello, world!" So we have the trigger working. We can see that if we send a message to Telegram, n8n successfully recognizes that.
So what is the next step? Well, actually, it is to set up a switch. That way, we can use both text messages and voice messages. Okay, so let's do this. I'm going to do "Switch," and then in here, we have to set up the values.
So here we're going to have two different routes: one for text and one for voice. I want to make sure our AI agent works with both. So first, we're doing the text. I'm going to rename it to "Text," and we need to make sure that this value right here exists.
So we need to do "String" and "Exists." If this exists, it's a text. Now I'm going to test the trigger again, but with a voice message so we can properly wire that up.
So click on "Delete." This deletes the test, and now when we click on "Test Workflow," it's waiting for us to create a test event in Telegram. So let's switch to Telegram, and I'm going to say, "This is a test message 1, 2, 3, 4." Boom! There it is.
So now let's go into "Switch," and as you can see, this is different. We have the audio, and we have the file ID. So let's create another route. This one is going to be named "Voice," and we just need to make sure that the file ID equals "String Exists."
So if this exists, we know we have a voice message. If this exists, we know we have a text message. They cannot exist at the same time. So now our switch is currently outputting two different scenarios: a text scenario and a voice scenario.
In case of a text output, we actually want to use "Edit Fields." To set this up, I'm going to do a text output again. Okay, so let's do another test. Click on "Test Workflow." I'm going to send "Hello there," and now we can actually work with this.
The reason why you want to test it each time is that you want to drag and drop stuff. You can copy-paste the code from my video, but that's a very bad strategy. What's much better is if you wire it by dragging the actual field itself because then you have it guaranteed to work.
Okay, so we need to add a field for the name. I'm just going to do "Text," and for the value, I'm going to drag in the text right here. Boom! That's good.
Now, what if it's a voice? Well, we have to process it a bit. So if it's a voice, click on "Plus," click on Telegram, and we actually need to find the voice file. So click on "Get File," and in here, "File Get," and we need to get the file ID.
Now, again, as you can see, we have no data right here because we currently tested a text message, not a voice message. So let's delete our testing. By the way, when you click this, "Delete," it just deletes the current data in the test. It doesn't delete anything about the setup, so don't be afraid to do that.
So let me delete this test, and now let's rerun it, but with a voice message again. Okay, so it doesn't let me run it with this field empty, so I'm just going to put "Test" in there. I don't know if that's enough. Oh, seems to be enough. That's annoying.
Anyways, let's test the voice message again. "This is a test voice message. Hello world, David here." Okay, boom, boom, boom! And now, okay, so if we double-click it, as you can see, now we see the data because it's no longer a text output; it's a voice output.
So every time, make sure you actually run the test so you work with the data. So anyways, we need the file ID in here, so let's delete this. Obviously, that's incorrect. Okay, so that's my bad. Make sure to connect the file ID, not just the ID. We have to drag in the file ID.
Okay, beautiful! So let's try it again—like actually rerun this. Boom! And now it's executed successfully. But when we find the file ID, we have to do something with the file. That's where OpenAI comes into place.
So click on OpenAI, and then click on "Transcribe a Recording." First, we need to connect an OpenAI account. So if you didn't do this before, click on "Create New Credential," and all you need is an API key.
Now, if you've never worked with API keys, stop panicking. It's actually super simple. All you need to do is go to platform.openai.com, and on the left, click on "API Keys." In here, you can actually log in with the same account you use in ChatGPT.
This takes 20 seconds. Once you're in here, click on "Create New Secret Key." I'm going to name it "n8n Transcribe." Create secret key. Again, do not share this with anybody. I will revoke my key before uploading this video. Treat them as passwords.
Okay, so once you copy this, go back into here, paste it in here, click on "Save," and wait one second for "Connection tested successfully." Beautiful! I don't think you actually need to do any changes in here.
Anyways, let's test it out. Click on "Test Step." Executing note. Boom! So this is the text. "Thanks so much for listening, my friends. I'll see you." That's not what I said.
Okay, maybe actually my setup was wrong in Telegram, and it was using the wrong mic. Now it's using my proper mic, so let me try again. As you can see, the audio waveform was silent, so it didn't pick up the audio. So that's my bad.
Let's click on OpenAI. "Okay, David here." Test says beautiful! So now it's correctly transcribing because the audio is correct.
Okay, so that's why you need to test at every step; otherwise, you just can run into issues that later on will be much harder to debug.
Okay, so we're successfully handling both text and voice. Now we're ready for the main part of connecting the AI agent. So let's click on "Plus" in here. Do "AI Agent." Boom!
So this is by far the main module, and we actually want to connect both into the input. Okay, let's go back to Vectal. So we did set up the switch.
Okay, one more thing. If you want this whole automation and you want to just copy-paste it, you're like, "Oh my God, David, this is too complex. I don't have time to set this up." All of that will be in the New Society.
So inside of the New Society, if you click on "Classroom" and then "Templates and Presets," you can get all of my automation, AI agents, code prompts, and everything else that I've made in previous videos, and also exclusively for New Society.
So specifically in pre-built automations, you can find this one: "n8n Google Agent." So you can literally just download this and paste it into your n8n, and you don't have to spend an hour building it.
So if you want access to all of my automations as well as tons of exclusive content, make sure to join the New Society. It's going to be linked below.
Okay, so the next step is adding our AI agent, and this is the best part about it. This is what makes it better than Zapier and make.com. This AI agent is very powerful, and it's unfortunate because you can only appreciate it after you build your own AI startup like I did with Vectal and after you try giving an LLM different tools, memory, and stuff like that.
Luckily, n8n makes this super easy compared to what it's like when you actually build it with code. So here's what we have. We can use Azure OpenAI if you want OpenAI models with privacy, Anthropic, Mistral, or Gro if you want super fast responses, or OpenAI, obviously, and Google Gemini models.
I'm going to use Anthropic because right now, the new Anthropic is my favorite model. But be careful because by default, this is the old Anthropic. The new Anthropic is 1022, so make sure to select the new Anthropic.
Now we have to connect our Anthropic account. Mine is already connected, but if you don't have it connected, click on "Create New Credential," and as always, we need to paste in an API key.
So you need to go to console.anthropic.com. That way, it's easier for you to access. Then click on "Settings" at the top, and on the left, click on "API Keys."
Now again, if you don't have an account, you can use the same account you have on ChatGPT. If you don't have ChatGPT, you can create an account in like 30 seconds. Super easy!
So once you log in, once you go to "Settings" and "API Keys," click on "Create Key." Workspace: I'm going to do "Default," and I'm going to name it "n8n AI Agent." Add.
Again, do not share this with anybody. I will delete my key before uploading this video. Okay, so the reason I say that is because always there's like a few people in the comments, "Oh my God, he showed his keys!" Guys, I'm not an amateur.
So once you paste this in and click on "Save," you'll see "Connection tested successfully." So now our AI agent is using the world's best AI model, at least currently.
When you click away, you can actually see it very nicely. You can see the chat model is using the Anthropic chat model. Let's click on "Save."
So now there are two more things we have to give it: tools. This is what separates it from a simple chatbot, right? Where you're just chatting with the AI for an actual powerful AI agent, it's the tools and memory.
But honestly, memory—that's an entirely different video. How to create vector databases, how to take thousands of words and hundreds of documents and put them into a vector database, and how to give them to your AI agents so that AI agents can have infinite memory.
By the way, if you want me to make a video on that, you know, connecting AI agents to a vector database, comment below.
Anyways, what really makes AI agents powerful is the tools, right? So let's actually first skip this, and let's set up the output. The reason for that is I want to see if our AI agent is working.
So for the output, we're going to use Telegram as well. Again, you can use anything else. You can use WhatsApp, Gmail, Slack—whatever else you want. Do not be an NPC, and do not blindly follow what I'm doing. Instead, actually adapt it to your use case, to whatever you need to automate.
Take this and modify it. So I'm going to go with Telegram again. Telegram, and I'm going to say—actually, we need to do "Send a Text Message."
Send a text message. Okay, so make sure this is the same account you connected at the start. Message: "Send Message." Chat ID.
Actually, before we set these two fields up, let's first run the AI agent. For that, we need to set in the system prompt. So underneath "Prompt Source," we need to change that to "Define Below," and let's drag in the OpenAI output in here: json.text.
Okay, now underneath "Options," we can actually add a system message, which is the system prompt. So you are a helpful assistant. I'm going to say, "Always respond in full caps." That's just for testing, right?
So let's do "Test Step." Okay, "Test Step." It says it executed successfully, but it's not doing anything.
Okay, so when stuff isn't working, sometimes it's best to delete the data and try again. So let's do "Test Workflow." "Which day is after Monday?" Super simple question to test if this is working.
So text—it's a text field. Boom! Beautiful! So now when I click on "AI Agent," we can see the output: "Tuesday comes after Monday." Blah, blah, blah. It's full caps, so it does see our system prompt.
But the error is in the Telegram message, and that's because we didn't wire it up. So let's do it now. For the text, we just need to drag in the output from the AI agent because we want the exact same text as the AI agent, right?
And for the chat ID, we actually need to scroll down to the switch, find chat, and then the ID right here. Boom! And let's actually test the step. Boom!
Let's go into Telegram. Do we see anything? Not yet. I'm going to run this again. Still didn't set anything.
Okay, so let's delete it again. "Test Workflow." "What is 4 + 4?" Right? Let's see. Boom! There it is!
So we got our first successful completion. Now, this is not done yet because we still need to connect the tools and actually make this powerful, right?
So, okay, so now one thing I should probably explain—let's click on "Save" first—is that we're in testing mode, meaning the workflow is inactive. Right? If I were like—once you're happy with it, we were going to trigger this on, and it's going to be active always.
But now when you do test, you need to delete it and do "Test Workflow" again. So that's why now I didn't get a response. And boom! "June comes after July." So there it is!
All right, so let's delete it again. Actually, we can keep it, but what we need to do is we need to add the tools for our agent.
So as you can see in Vectal, I have the steps. By the way, if you're someone who likes to be super productive, make sure to check out Vectal. This is a task management tool that actually saves you time.
It has built-in AI that automatically sorts your tasks and helps you complete them. For example, I can say, "I've completed the AI agent task." Obviously, it's a simple request, but let's say you have too many tasks.
You want to move half of them to tomorrow, or you want to reorganize them. Like, for example, today I want to focus on business only. It will decrease the priority of personal tasks and increase the priority of business tasks. It can do a lot more, so go to vectal.ai and try it yourself.
All right, so the next step is adding tools to our AI agent, and this is really what makes the agent powerful. So the reason I name it "Google Agent" is because I'm going to add some Google features, right?
First and foremost being Gmail. So let's click on "Tools" and click on "Gmail." Let's start by giving it access to the inbox, right? And giving it access to all the unread messages.
So for the operation, we need to change it from "Send" to "Get Many." That way, you can get multiple emails. Limit you can leave on 50. I mean, it's going to increase your token cost if you have a lot of unread emails, so maybe consider changing that to like 20. It's up to you.
Leave "Simplify" on, and actually at the top, the first thing you need to do is connect your Gmail. So again, if you don't have it connected, click on "Create New Credential" and click on "Sign in with Google."
Here, choose the email account you want to sign in with. Once it says "Account connected," we're good. So once you have the account connected, scroll all the way down, and we need to set up the filters.
So label names or IDs—in here, we need to select "Inbox." So scroll down and click on "Inbox." And the second thing is "Unread." So right now, it can get your unread emails.
Now, if you don't want just unread emails, you can obviously set it up to anything else. That's entirely up to you, but for this use case, I'm going to have emails that come into the inbox, meaning it's not going to be some promotional material, and then only unread emails are going to show up.
So that's good. Let's exit out of this. So now the chat agent has the tool, and it should be able to use it. Now, in terms of the prompt, we actually have to specify that a bit more.
So in here, I'm going to change the system prompt to actually something more custom. And by the way, if you want access to that, again, it's in the New Society, in "Templates and Presets," underneath "Prompts and Instructions."
Right here, if you scroll down, you'll find the "n8n AI Agent System Prompt." So again, if you want access to all of my system prompts, make sure to join the New Society.
All right, so once I've pasted this in, I'm going to save again, and now it has knowledge of the Gmail tool. That's the main thing. In the prompt, we give it knowledge of the tools it has and how to use them.
So let's test out the execution. I'm going to delete this. I'm going to do Telegram, and actually, first, we need to click on "Test Workflow."
And I'm going to do a text this time. It's good to alternate between voice messages and text. I'm going to say, "What is my latest unread email?" Enter.
And let's look at—okay, so we go into AI Agent. Boom! It starts, activates the tool, goes back into the Anthropic model. Man, this is so beautiful to watch!
And now we get a response. So actually, if we check this out, here's my email right now from Google: "No reply security alert." Oh yeah, so it is this.
Okay, I thought it would be this one, but I didn't realize we got a new email in the meantime. So yes, it correctly found this email.
Okay, let's do one more test, but this with voice. So let's delete this. Let's start "Test Workflow." "Summarize all of my unread emails into a concise message."
Okay, let's see. So we transcribed the recording. Okay, boom! AI Agent. Now it chose to do the tool. Now it goes back. Oh my God, this is so crazy, guys!
You have no idea how difficult this is to code. If you want to do it in code, setting this up in code will take you multiple hours. With n8n, it's like 5 minutes, and you have the AI agent set up.
It's so beautiful! I wish you guys—you can only appreciate how beautiful it is if you actually try building it with the Anthropic API in Python.
Man, anyways, we got the response, and it summarized all 400 emails: "Security alert, Cognition AI, YouTube notification, and another security alert."
And there it is! Actually, you know what? Let's mark this one as read to simplify it a bit. So now we should only have these three. But as you can see, it accessed all of them.
So that's one tool. Okay, it can already do a lot more than ChatGPT, right? Like now, anytime I go into Telegram, whether it's on my phone or on the desktop app, I can just have it summarize my emails.
But you know, we're just getting started over here. The agent is about to get a lot more powerful.
Okay, so one tool we have: the Gmail. Actually, we can probably rename this. No, I'm going to say "Get Many." Rename.
Okay, now the next tool I'm going to give it is to send an email. So again, Gmail, and instead of—yeah, actually, operation "Send."
So in here, n1 is actually telling us what to do. Use the expression from AI placeholder name for any data to be filled by the model. So let's copy that and let's paste it everywhere what the AI model should fill out.
So to: subject, email type: HTML or text. Let's do text and let's put it everywhere—literally everywhere.
And okay, so that looks interesting. Now I'm going to also rename it from Gmail to "Gmail Send." And we have to update the system prompt of our AI agent, giving it information about this new tool.
So I'm going to say, "You also have access to the Gmail Send tool, which allows you to write emails and send them to anyone."
Okay, let's see if this is enough. Actually, I'm curious if that's enough. Let's click on "Save." Now added a new tool, and I'm going to delete this.
And I'm going to click on "Test Workflow" for the email since I need it to be exactly—I want to do a test. I'm going to do a text message. That way, I know the email address is exactly correct.
So I'm going to say, "Send a friendly lunch invitation to daviddavid.com." I think that could be enough. So let's do "Send." Let's see what happens.
I'm going to wait on my phone. I have it loaded up here. "Send email." And we have an error.
Okay, I see the error right here: invalid something invalid. Okay, boom! No, actually, let's see. I have it on my phone right here.
Well, let's see what happened. Telegram, I apologize, but I'm experiencing technical difficulties. Gmail sent up—it seems to be an issue with how the email parameters need to be formatted.
Okay, so let's see what the error is exactly. Invalid email address. Okay, let's double-click this. What is the issue? Placeholder name to subject.
All right, so maybe this is a good place to actually use the n8n assistant. So ask assistant: "Analyzing the error." Let's see the error you encountered. Blah, blah, blah is not resolving.
While email to resolve the issue, ensure that the expression used in the tool is previous name, previous node name, item.
Okay, so the issue here is actually quite fundamental, so pay attention. The mistake I made is that I put "from AI placeholder name" into all of them, and we have to change that.
We have to give it a clear description of what should be in this, right? So in here, I'm going to say "To." I'm going to give it a description. I'm going to say "Raw email address of the recipient."
And this is actually an instruction for the AI. So now it doesn't work because we have the incorrect parameters. On the next test run, this should work.
So, okay, so we fixed the "To." Now we need to fix the subject in the same manner. So delete everything in the parentheses here. We just need to do what it did here, but we need to give it a description.
Subject: "This is going to be a clear and concise subject line." Okay, and then for the message, let's click on "Body."
So actually, the AI already knew how to do this; we just didn't format it properly. Okay, so for the body, I'm going to say "The body of the email itself," and end it with a quotation mark.
Okay, so now if we click out and save, delete this test, and let's test it again. "Test Workflow." Switch on Telegram. I'm going to copy the same exact email.
"Send a friendly lunch email to David Andre." Boom! And Anthropic should choose to send the email right here. Boom! Correct!
Okay, let me check my phone. Okay, boom! There it is: "New AI Society." Okay, so the body is all right. So it did send the email.
However, we have an issue: the subject is not parsed, so it sent the whole thing, right? All right, all right, right? So we need to look into what happened here.
Oh, I need to switch it to expression. That's my bad. Let's run this again. Save. "Test Workflow." Let's do a different one.
Let's do "Send a party invite to my good friend David Andre at daviddavid.com." Boom! Let's check the phone. Still writing, still writing.
Oh, there it is: "Party invite." Now it's correct! Guys, can you see this? This is beautiful! Party invitation!
Let's make it legendary! But man, now it wrote the email and formatted it correctly. So now let's give it more powers. Let's give it Google Calendar as well.
Okay, so let's click on "Tool" again. Plus, do "Calendar," Google Calendar tool. And in here, let's start by having it read events.
Okay, so if we go into Google Calendar, I don't have anything in here, but let's see. Let's maybe try. I'm going to do like a dinner here.
I'm going to do "Dinner with my YouTube subscribers." If you want to be invited to this non-existent event, make sure to subscribe.
Okay, so let's save that. Now, if we give it the ability to read, so operation is going to be "Get Many," and then from list, select "Calendar."
Oh, okay, okay! So as you can see, we did not add the credential. I made the same mistake the first time. We should always start by selecting the credential.
So let's click on that. Create a new credential. Sign in with Google. Account connected. And now inside of here, we should be able to select "New AI Society."
So this is my main calendar for this account. Maybe we can do like "After" and then select date and time. Could be like start of this month or whatever—something like that.
Yeah, the expression should be like this. Let's see if that's good. You know, maybe you don't want to load super old events.
Okay, so this looks correct, and I mean, the easiest way to test that is just to test that, right? So I'm going to delete this, and actually, let's update our AI agent.
We need to tell it that it also has access to the calendar. So let's put that here. You can also access the calendar read tool, which lets you see all of the events on the user's Google Calendar.
All right, so let's save that and let's test this workflow. I'm going to say, "Do I have any calendar events for today? If so, what are they?"
Okay, let's see. So we're going AI Agent. Hopefully, it calls the calendar read. Oh, it did! Boom! Okay, that's green.
Let's see. Let's go back. Let's go back in Telegram. Boom! There it is: "Dinner with my YouTube subscribers from 8:00 PM to 9:00 PM Dubai time."
Let's check that, and it's correct! So now it can access our calendar, which is amazing. But what's even better is if it can create events.
So let's give it that right now. Calendar, Google Calendar tool. Instead of—yeah, operation "Create." Beautiful! Select calendar: the main one.
And then the start—okay, so this should be AI. Again, we have to make sure that this is filled with AI, and we have to do the same thing as we did before.
So let's copy this. Boom! And paste it in here. And we have to instruct it, right? So let's switch to expression and delete the stuff inside of the brackets, parentheses.
And since this is the start date, we need to instruct it how to format it, right? So we're going to do "Quote Start," quote, and then a clear description: "This is the date and time when the event should start."
Okay, and let's copy this expression and paste it into the end. Boom! And instead of "Start," obviously, it's going to be "End."
And when the event should end—beautiful! And inside of additional fields, I'm going to add "Summary."
And in here, let's switch on expression as well. We need to do the title of the event. So title—change that to "Title."
And the description—change it to "The title of the event." Calendar create. Okay, looks good!
And again, let's click on "Save." And the best way to test it is to test it. So, okay, we tried speaking.
I want to say, "Create actually, you know what? Let's try to create an event in the past." So I'm going to say, "Create a 15-minute event starting today at 5:00 PM."
Or maybe I could say, "Track the last 15 minutes in Google Calendar by adding a new event named 'Recording for YouTube.'" Let's see if it can do this.
Oh, actually, I didn't do "Test Workflow," but whatever. Let's click that. All right, so successfully create event. Boom! There it is.
Let's see the response. "I've added the event." Let's click on Google Calendar. Refresh, and it's not here. Why not?
Oh, it might not have knowledge of today. Okay, wait, wait. What do we name it? "Recording for YouTube."
Oh, it put it January 2024. Oh my God! Okay, so that's crazy. January—wow, it's last year.
So the issue is that—okay, so the issue is actually in the AI agent. It doesn't really have knowledge of what the date is.
So let's click on the AI agent. We have to update that. It should have it in the—now why doesn't it know it? Maybe we should actually repeat this.
Let's copy this. I'm going to put it in here. Go Calendar. Actually, oh, we didn't put in the calendar create.
Okay, that's one bad. Let's put it in here. I'm going to rename this to "Calendar Create."
Let's scroll down to the system prompt and let's update the knowledge of that. So yeah, I'm going to say, "Use the calendar create event to add any new events into the user's Google Calendar."
Important here is today's actual date. Then if you click on expression, we should see date time 2025.
Okay, maybe we can put more after this. "Make sure to set the start and end times accordingly."
Okay, so hopefully that is clear enough. We have to convince the assistant that it's 2025 because its knowledge cut-off is in the past.
So let's save that and let's try again. Okay, so I'm going to delete it. I'm going to do it. Actually, I'm going to name it "Building in n8n."
Okay, let's do that and let's do "Test Workflow." Boom! Oh, actually, one more thing I should have specified.
Yeah, okay, so now it might work. Wait, let's go back to today. Oh, there it is! Oh, it's very specific: 5:16 to 5:31. That's really good!
Okay, that's really good. One more thing I should have done, though, is inside of the AI agent, I should have specified the format.
Because in here, I need to specify this type format. Okay, so maybe actually we can use ChatGPT.
Take this date and give me the general date and time format this is written in. Answer in short, and then I can specify that furthermore in the system prompt for—oh my God!
I meant like, yeah, I can specify that in the system prompt so that it knows it exactly. Okay, actually, we can copy this message right here and include that in our AI agent.
So let's pop that out, and in here, here is today's actual date. Make sure to set the start accordingly in this format.
Okay, I'm going to paste this in. Delete this. It's redundant. Year, year, room.
Okay, year, month, day, T, second, time offset. Okay, maybe that actually is not needed, to be honest.
Anyways, let's save, delete, and I'm going to create a new event in my calendar for today from 6:30 PM lasting 45 minutes.
And it should be reading. Okay, let's see if we can do that. There, the workflow.
Okay, the AI agent chose the right tool. Okay, it basically always does that. And let's see.
Okay, the tool executed successfully, so let's hope this is correct. Reading—oh, this is not 45 minutes, though. The end time is correct.
Okay, maybe I confused it. Actually, maybe the—what is in here? Start? Why is the start? Did I mess this up?
Anyways, let's save that. Delete. "Test Workflow." Okay, event was created. Let's see.
Oh, okay! There it is! It works! It works! Is it? Yes! It's beautiful! Beautiful!
So yeah, that was my issue. I set up the calendar wrong, and that's why you need to do a ton of testing.
See, the secret of building with n8n is testing at every step. You literally should test, test, test! Like, no matter anytime you change, do a test.
And it's the same programming, to be honest. But yeah, like in here, it's super clear. You need to test every tool you add.
You need to test any prompt change because you never know what happens, right? And especially the more complex your AI agents and automations become, the more testing you have to do.
So now our AI agent is actually quite powerful. It can read all of our unread messages. It can send responses to them. For example, it can read the calendar, like get our events for today or for other days, and it can create calendar events.
Now, again, if you want to add more tools, just click on the plus icon next to "Tool," and literally, it's almost endless. Like, they have so many things, right?
So you can do calculator, you can do Airtable, you can even call other n8n workflows. That's where it gets real crazy. You can literally have custom workflows for anything and then just add them as tools to an agent.
You can do, like, let's say, you know, you want to add Google Docs, Google Drive, Google Sheets. Let's say set up APIs, Google Search.
So for example, if you do ChatGPT, you can give it Google Search, and suddenly you have ChatGPT with web search, which is even more powerful than ChatGPT.
You can do anything: Slack. Just look at the opportunities, right? Hacker News, Jira, Microsoft Outlook, MongoDB—even databases!
Like, do they have Supabase? They even have Supabase! Oh, that's actually kind of crazy.
Okay, did I could actually integrate this into Vectal in some way? Hmm, now we're talking!
As you can see, n8n is really the most advanced AI agent out there. Now, actually, let me show you what to do once you're ready.
So once you're happy with it, it's like, "Okay, I'm finished building. I'm finished testing." Click on "Save," as always, and then flip it to "Active."
Okay, so workflow activated. Don't show again. Got it! And now it's active. So we can actually delete this test. It doesn't matter.
And if we go into Telegram, it will work every time. So let me do it. Let me do it from my phone so you guys can see it truly works.
Open up the Google AI agent right here. "Quickly summarize what unread emails are in my inbox."
Okay, so let's say you're busy. You don't have time. Like, let's see if there's anything important. Boom! There it is!
You can see it on the screen. You can see it on the computer better, but this is the response. So it took like, what, four or five seconds?
And we have the summary of the three unread emails we have in here. So that's just one automation, and that can save you hours per year or even per month.
Like, see, the beauty about these automations and AI agents: you build them once, and they work forever.
And if you are considering, like, "Should I use Zapier? Should I use make.com? Should I use n8n?" The worst thing about n8n is the name. Other than that, it's great!
Like, the product is amazing. The AI agent is insanely powerful, and this is honestly the easiest way to build AI agents for yourself.
You can do this with zero coding experience. As you can see, we didn't write any code in this video, and you can still build very powerful AI agents that you can use from WhatsApp, Telegram, Slack, Messenger, Google Sheets, Gmail—anywhere!
You can integrate them anywhere. They have literally thousands of possible triggers inside of n8n.
So really, the ball is in your corner. There are no more excuses. There is an amazing tool that doesn't require any coding, that's visual, that's easy to use, that's very powerful, that allows you to build AI agents.
If you don't use this, if you don't build AI agents, in two years, you'll be like, "Oh my God, David, I'm behind!" You know, AI has changed the world.
It's up to you! You now have all you need to know to start building AI agents. Start automating your tasks.
If you have, let's say, a job, right? Show this to your manager. Show this—go as high as you can in the company leadership.
You know, go to the leadership levels and show this to them and say, "Look, we can use this to automate. Just, you know, let me do something. Let me automate some part of our workflow."
If it works, I would be happy to discuss a potential raise. Most leaders, most CEOs, will be more than happy to do that if you can automate something that's costing the company hundreds of human hours per month with an automation that's predictable and does the job just as well, if not better.
That's one of the best ways, fastest ways to get a raise. Of course, you can build these automations for clients. I mean, there are so many options, but where you should start is build them for yourself.
So whatever you need, you can build it with n8n. And honestly, I think I'll make more videos on n8n.
That being said, hopefully, you found the video valuable. If you did, make sure to subscribe. And more importantly, go to vectal.ai.
Currently, I'm hiring for the developer, and what that means for you, people who are using my Vectal customers, is that the tool is about to get a lot better.
Even though now I've been adding features and improvements like every few days, now it's going to be basically every day once I have a full-time developer.
Man, Vectal is going to become a real software contender to Todoist and Notion very, very quickly.
So if you want to be one of the first people using Vectal, make sure to sign up now. And if you don't, then well, you'll sign up in the future anyway because the tool is only going to get better.
With that being said, thank you guys for watching, and I wish you a wonderful, productive week. See ya!