Automate ANY task using AI Agents in n8n! (full system)

0:00By the end of this video, you will have

0:01a full framework for using AI agents to

0:04automate any task in N8N. And to whoever

0:07is watching this, I guarantee that right

0:09now you have tasks that an AI agent can

0:11do both better and faster. And the

0:14longer that you delay learning how to

0:16build these agentic systems, those

0:18wasted hours will begin to pile up, and

0:21it's actually holding you back from your

0:22growth. Now, this framework fixes that,

0:25and it's a system that I use in order to

0:26implement AI agents into my business and

0:29my personal life. Now, this is the

0:31system that we are going to be building

0:33today. I'm not only going to show you

0:35how to build this system that's actually

0:37going to provide you immediate value,

0:39but I'm also going to give you a

0:40framework for developing systems like

0:42this whenever you want. And so, let me

0:45show you how this works. What this

0:47automation does is it watches for new

0:48YouTube videos on all of my favorite

0:51channels, AI Foundations, Productive

0:53Dude, and N8N. What happens is whenever

0:56one of these channels uploads a new

0:58video, it automatically gets put into a

1:01database. And then in a matter of about

1:0310 seconds, it grabs the transcript of

1:06that YouTube video that just got

1:08uploaded. It runs through an AI agent,

1:11which has access to my goals, what I'm

1:13doing in my personal life, what I want

1:15for my business, and then that agent

1:17makes a decision. Should I watch this

1:20video or should I not? Is it signal? Is

1:22it noise? And then it sends me an email.

1:25So, for instance, I could execute this

1:27workflow and send in some test data from

1:29my YouTube channel. As you can see, it's

1:30going to get that transcript, and in a

1:32matter of seconds, my AI agent is

1:34already reading that transcript. In the

1:36instructions of my agent, it has the

1:38goals of my business and of my life, and

1:42it's going to make a decision. Should I

1:43watch this or not? I no longer have to

1:45get on YouTube, because what I can do is

1:48I can put the people who are actually

1:49providing value on a feed trigger, so

1:53that whenever they upload a video, it

1:54gets the transcript, and then the agent

1:56decides, is it even valuable enough to

1:58watch? And it sends me an email just

2:00like that. Now, let's go take a look at

2:02my email. As you can see, just now it

2:04gave me an email, new video from AI

2:06Foundations. It gives me a quick summary

2:08of the video. It gives me an ROI score

2:11of a medium, and it gives me the reason

2:13it gave me that score. So, ROI, if

2:15you're not familiar, means return on

2:17investment. What is my return on

2:18investment after watching this video? It

2:21says it's medium, and then it gives the

2:22reason. And then it gives me the link to

2:24the YouTube video. So, I don't even have

2:26to go to YouTube to find it. My

2:28automation puts it in the database,

2:30grabs the transcript, and then my agent

2:32decides on my behalf. And it's not even

2:34like it's deciding on my behalf, it's

2:35just giving me the option and laying

2:37everything out with the summary, the ROI

2:39score, and the reason. Now, in this

2:41video, I might be moving fast, and

2:43that's because I want to teach the

2:44framework, not necessarily how to build

2:47the AI agents themselves. Now, you're

2:49going to be learning things, especially

2:50if you're new, and I'm going to break it

2:52down as simple as possible. But, if you

2:54want a full 40-module agent building

2:56course, I recommend joining our AI

2:59Foundations community, because in the

3:00classroom, if you scroll down, we have

3:02an entire course on building agents with

3:04N8N. And this course goes over literally

3:07everything that you need to know in

3:09order to build out these agentic

3:11workflows completely by yourself. We

3:14have over 40 modules in here,

3:16drag-and-drop templates. And

3:17furthermore, if you ever get stuck, you

3:19can go to the support category and type

3:22in a question and then a description of

3:24your problem. One of our five team

3:26members, including myself, will help you

3:28out, giving you custom Loom videos and

3:31custom guides in order to get you to

3:32where you need to be and solve any

3:34technical issue that you have with

3:35building agents. So, if that interests

3:37you and you're sick of going on YouTube

3:39to try to learn all of the techniques,

3:41and you want one structured spot to do

3:43it, I left a free video in the

3:45description below that goes over

3:47everything that our community has to

3:49offer. I just mentioned a couple of the

3:50things, but I recommend watching that

3:52after this video to see if you would be

3:54a good fit for our private community of

3:56AI enthusiasts. Now, the beautiful thing

3:58about this framework is it allows you to

4:01take any input in your life and automate

4:04it and let the agent make the decision

4:06for you. So, life is full of inputs,

4:09especially nowadays. I mean, right now

4:12you're watching a YouTube video, which

4:13is an input into your mind. You're

4:14deciding, should I implement this?

4:16Should I not? What do I need to learn

4:17from this? Which, as you saw in the

4:19beginning, we're going to learn how to

4:21actually make a system that works for

4:22us. We have emails, we have notes, we

4:25have travel information, bank

4:26statements, subscriptions, team

4:29communication. We have news updates that

4:31we're deciding constantly, should we

4:33apply this to our life? Should we care

4:34about this? Should we not? We have ideas

4:36that are slipping our mind because we

4:38just have so many things going on, phone

4:40calls that are coming in, Zoom calls we

4:42have to get on, direct messages we have

4:44to answer to, invitations we have to

4:45answer to. Everything in life is an

4:48input. And chances are you are usually

4:51the one making decisions on all of these

4:53inputs. You've got to make the decision

4:55whether to apply the knowledge. You have

4:57to make the decision on how to respond

4:59to certain things. And by the time all

5:01of these decisions and all of these

5:02inputs in your day-to-day operations are

5:04complete, you've already used up 80% of

5:06your mental bandwidth on the day, with

5:08decisions that could have just been

5:10completed with AI in the first place. A

5:12lot of the time, you don't even need to

5:13be the one making decisions. AI can make

5:16the decision for you. And I'm going to

5:17show you the system that allows AI to do

5:19that for you. And that does not mean

5:21that you're completely removed from the

5:23loop, by the way. I like looking at AI

5:25as an enhancement to your existing

5:27workflows. That's where it thrives best.

5:30And so, the system is pretty simple. We

5:31take our inputs, we put them into a

5:34database with an automation, and then we

5:38train an agent on three things. Number

5:40one, how do we use the database itself?

5:43The agent needs to understand the

5:44database and the information that it has

5:46access to. Number two, the agent needs

5:48to understand our goals and the reason

5:51for automating this. Like, why are we

5:53automating information into the database

5:55in the first place?

5:57And number three, we need to give some

5:58instructions and guidelines for how to

6:00respond. And so, with all three of these

6:02things, the agent can begin to make

6:04decisions on our behalf that we don't

6:06need to be there for. We can just say

6:07yes or no. We don't have to think about

6:10all of the process that goes into making

6:12a successful decision, because we can

6:14set that up in the agent. And I'm going

6:16to show you how to do this with YouTube

6:17videos. We're going to automate YouTube

6:18videos into a database. We're going to

6:20grab the transcript during the

6:22automation. We're going to feed that

6:24transcript to the agent. And then that

6:26agent is going to use these three

6:28things, and it's going to help me make a

6:30decision. It's going to score the video

6:32based on the transcript, and it's going

6:33to say, "Drake, should you watch this or

6:36should you not, based on the goals that

6:38you've given me?" And I've done this

6:40with everything. I've done this with

6:41Zoom call transcripts. So, I've gotten

6:43summaries of my meetings. I've done this

6:45with ideas to where I set up Siri. All I

6:47had to do was say, "I have an idea." And

6:49what it did was it transcribed my voice,

6:52sent it to a database, and all my ideas

6:54were stored in that database. You could

6:56do this with anything in your life, but

6:58I recommend starting with something that

6:59we're all doing right now, which is

7:00watching this YouTube video. Step number

7:02one in all of this is update your N8N to

7:04the latest version so you can get the

7:06data tables field right here. In order

7:09to do that, you can open up your

7:10sidebar, go to the admin panel. When

7:13you're in the admin panel, go to manage.

7:16And here, where it says N8N version,

7:18make sure that you're on the latest

7:19stable. After you've done that, we can

7:21go to data tables, and we can create our

7:23database, the thing that we're going to

7:24be sending all of the YouTube

7:26information to. And I'm going to click

7:28in the upper right-hand corner. I'm

7:29going to hit this drop-down and select

7:31create data table. And this is just like

7:33creating a Google Sheet. It's very

7:34simple. I'm going to call this YouTube

7:36testing, and then hit create. And from

7:39there, I'm going to create my fields.

7:40So, I'm going to add all of the fields

7:42that we need,

7:43starting off with video_title,

7:46so we know the title of the video. We

7:48can leave these two created fields for

7:49us.

7:50I'm going to type in URL, author.

7:53I'm going to type in publish_date,

7:56and I'm going to make the type a date

7:58time. Add column. And I'm also going to

8:00have a transcript field, so I'll add a

8:02column and type in transcript. And then

8:04I'll hit add. We can leave that a

8:06string. And now our database is ready to

8:08throw information to. So, first we need

8:11to create the automation that actually

8:12watches my YouTube channel or any other

8:15YouTube channels that I want to have on

8:18an automated timer, so that whenever

8:19they upload, I can get notified if the

8:22video is actually valuable to me. And

8:23this is putting me and putting yourself

8:26not at the whims of the algorithm, but

8:28it's giving you control over what you're

8:30consuming. That's a big problem right

8:32now is we don't even have control over

8:34what we consume. They throw videos on

8:36our feed, and our mouths start watering

8:38because we just want to watch them. So,

8:40adding this in your day-to-day system

8:42will already save you so much time. So,

8:44in the upper right-hand corner, I'm

8:46going to hit create workflow. I can

8:48rename this to like YT automation test,

8:52whatever. You can name it whatever you

8:53want. And the first step we need to add

8:56is an RSS feed trigger.

8:58And remember, we're building the

8:59automation right now and automating data

9:01into our database so our AI agent can

9:03use it. So, I'm going to add that first

9:05step and type in RSS. And where it says

9:08RSS feed trigger, you want to select

9:10that. And what this does is it

9:12automatically pulls in a feed every

9:16minute, hour, day, week, month, any

9:19custom time that you want. So, I can

9:20just watch this

9:22every minute, so that right when

9:23somebody uploads a new video, I can get

9:25it. Or if I want, I could also just do

9:27every day. And in order to get YouTube

9:30channel feed URLs, what you need to do

9:31is you need to go to the YouTube channel

9:33you want.

9:34For instance, if I'm on N8N, I want to

9:36go to their channel and select their

9:38videos tab. And then you want to

9:40right-click anywhere on the screen, and

9:42then hit view page source. And it will

9:44bring up a screen that looks like this.

9:46Once you're here, you want to hit

9:47command or control F on the keyboard,

9:50and it will pull up this little search

9:51bar for you. You want to type in quote

9:54RSS in all caps, and then another quote,

9:57and that will pull you straight to this

9:58link. You can just copy that link like

10:01this, and then go put it into your RSS

10:03feed trigger. So now,

10:05if you put that feed URL, it pulls in

10:08the channel ID for you, which is

10:09important, because now it's watching

10:11that channel every day to find the

10:13latest content that it's posted. So if I

10:15hit fetch test event, it's going to pull

10:17in N8N's latest YouTube video. So if

10:20you're having this feed run every

10:22minute, it's going to be basically

10:24real-time pulling in of data to your

10:27workspace. And so now I'll rename this

10:29N8NYT,

10:31and now I'll go do this for a couple of

10:32other channels.

10:34And this is the step where you want to

10:35go through and you want to pull in all

10:36of the channels that you like watching

10:38that you get value from. And so now I

10:40have three YouTube channels. Let's say I

10:41like watching these. The next step is I

10:43need to create a node that combines all

10:46of their data into one unified field. So

10:49I'll go to the upper right-hand corner,

10:50I'll type in set, and I'm going to use

10:52the set node to map all of these no all

10:56of these trigger nodes to, so they all

10:58can be named the exact same thing. So

11:00I'm just going to connect all of those

11:02triggers to this set node, and then I'm

11:05going to start naming the fields. So you

11:07can get some test data in by just

11:09executing one of the workflows, and then

11:11double-clicking, and then you can find

11:13it, and then just start dragging in the

11:15fields like this. We can rename them a

11:17little bit later. I'll drag in publish

11:19date and author. These are the fields

11:22that we put in our table, so these are

11:23the fields we want to set. And now I

11:25want to map it to the actual field names

11:27that we had of our table, so we can be

11:29organized. So for the video title, I'll

11:31name it video_title.

11:32The link, I believe I called that URL.

11:35Publish date, I called that

11:37publish_date.

11:38And author, I left author. So now we

11:40have all of this information.

11:42So no matter what channel comes out with

11:44a new video, it's going to just set it

11:47with these names for each one of the

11:50fields. So it doesn't matter who the

11:51author is, it will be called author.

11:53Doesn't matter what the video title is,

11:54it will always be called video_title,

11:56and it will populate dynamically here.

11:58And so now we can set up a little search

12:00filter to make sure that the same video

12:02doesn't get in the database twice. We

12:04can do that by selecting this plus

12:06button, typing in data table, selecting

12:10that, and then hitting get rows. We

12:13could rename this to like

12:15search

12:16and filter.

12:18And where it says data table, we want to

12:20choose the table that we created, so

12:22YouTube testing in this case. And then

12:24we want to match all conditions, and

12:27then add a condition where URL

12:31equals, and then we can map the actual

12:34URL right there. So what this is doing

12:36is it's saying if there's a URL that

12:38matches the one incoming in my database,

12:41then we don't want the automation to run

12:43because it's already been processed. So

12:45what we can do is after this data

12:47search, we could test that out right

12:49now. As you can see, we get no output

12:51data, which is good. But if we want the

12:52automation to run, we'll have to go to

12:54settings and always output data.

12:57And so now when we run that, it's not

12:59just going to say no output data, it

13:00will just return empty. So if I select

13:03this plus button, and I type in if, I

13:06can add this to the canvas, and it

13:08automatically connected for us. And what

13:10we want to do is we want value one to be

13:14not empty. So is not empty. So what this

13:17is saying is if the value that we type

13:19in is not empty, it will be true, and it

13:22will go down the true route. And so if

13:25json.url,

13:27if this is not empty in the search and

13:29filter node, then we can go down true.

13:32So I want you to copy what I'm doing

13:33here. I want you to turn this to an

13:35expression,

13:37and then I want you to type in double

13:39brackets,

13:40money sign, json.url,

13:44because that's what this field name will

13:45be called in this search and filter

13:48category here.

13:49You'll see what I mean. So if I actually

13:51run one of these after setting it up

13:53like that, it should go to the false

13:54route since

13:56it is empty. And if it is empty, we want

13:58to continue the workflow down here. If

14:00it's not empty, we'll just stop the

14:02workflow. And so since it's not in the

14:04database, now what we need to do is we

14:06need to create a row in our table. So

14:10I'll hit the plus button, type in data

14:12table, and then I'll hit create row or

14:15insert row.

14:16I can choose this from the list, so I'll

14:19just go to the YouTube testing, and then

14:21I can start implementing all of these

14:23things in right here. For now, I can

14:25delete the transcript since we don't

14:27have that yet, we'll add that later.

14:29Now we need to map all these fields. So

14:31I can connect it to the false route, and

14:33then I can double-click into here.

14:36I'll go back to the edit fields node

14:37that we created earlier, and I'll just

14:39start dragging and dropping these fields

14:41where they match. So URL can go to URL,

14:43video title to video title, publish date

14:45to publish date, and author to author.

14:48So now, if I save this, and I open up my

14:51data table in another tab, as you can

14:54see, we have no data in here right now.

14:56But if I run this since we have insert

14:58row, and I go to AI Foundations,

15:01just like that, it adds a new row of

15:03data. If we hit the refresh button, as

15:06you can see, we now have my video title,

15:08my URL, the author, and the publish

15:10date. But now we need to pull in the

15:11transcript as part of the automation. So

15:14in order to do that, you're going to

15:16need to go to a website called RapidAPI.

15:19And this is free. You can do this up to

15:21100 times per month for free.

15:23So I just typed in RapidAPI, and I'm

15:25going to go to rapidapi.com. You can

15:27sign up for an account, it literally

15:28takes 5 to 10 seconds. Then you can go

15:31to the API hub. And what I want you to

15:33do is type in YouTube transcript,

15:37and then what you can do is select that

15:38top one with this logo by Solid API.

15:43This is going to allow you to grab

15:45YouTube transcripts very quickly, and it

15:47is free for 100 requests per month. In

15:51order to use it, you'll have to

15:52subscribe to it. You're going to see a

15:53subscribe button up here. You don't need

15:55to enter any card information, you can

15:57just subscribe to the free plan without

15:58any card. I'll show you how that works

16:00real quick. For instance, if I go to

16:02this fresh LinkedIn profile data,

16:05there's a subscribe button in the upper

16:07right-hand corner. I can hit subscribe

16:09to test. And for the YouTube one, there

16:11will be a free plan that you can use,

16:13and you can just hit subscribe to this

16:15plan, and it's a one-click subscribe.

16:16You have zero card information, and it's

16:18ready to go. And once you subscribe to

16:20the free plan, you want to go to this

16:22get transcript with URL,

16:24and then you want to copy this cURL over

16:27here on the right-hand side, and then go

16:29back to your N8N. Then what you want to

16:31do is you want to add an HTTP request.

16:33So I'll hit add, go HTTP request, and

16:36then I'll hit import cURL, that thing we

16:38just copied over here. This is a cURL,

16:41and it builds out the entire HTTP

16:43request for us. So if I import that and

16:46paste it in, and import it, now it's

16:49using this request, and we can

16:51dynamically put in any URL we want to

16:54pull in the transcript. So in order to

16:56do that, I'll just delete out that

16:59predefined value it had, and I'll make

17:01this an expression, and drag in our

17:04video URL from the insert row node. I'll

17:07drag that in, and now if I hit execute

17:09step,

17:10in a couple of seconds, the YouTube

17:11transcript will be all ready for us. So

17:14this is from my most recent video. And

17:16so we could rename that module if we

17:17want to like get transcript. And now

17:20that we have this in here, we want to

17:21insert the transcript into this row. So

17:24we can actually update a specific row by

17:27using the ID from this. If I click in,

17:30as you can see, each time you actually

17:33create a row, it comes with an ID. So we

17:35want to make sure that we update that

17:37same row that we've been messing with

17:39this entire process since each row

17:42equals one new YouTube video. We want to

17:44make sure we get that transcript to the

17:46exact row it needs to be in. So I'll add

17:48this,

17:50and I'm going to type in data table.

17:52What I want to do is update row. If I

17:54hit update row, I can select the data

17:57table, which is the same thing we've

17:58been using, and then I want to add a

18:01condition, and I want to match all

18:03conditions. I'll add a condition,

18:06and we can leave this on ID number. We

18:08can leave this on equals. And what we

18:10want to do is we want to hit the

18:11drop-down on insert row, and drag and

18:14drop this ID to that value. This will

18:16ensure that we are updating the row that

18:18was created at the beginning of this

18:21run. And then I can just delete

18:23everything in here except the

18:25transcript, and then I can just go pull

18:28in the transcript, and drag and drop it

18:30in right there. So this can be adding

18:33transcript. And then what I'll do is

18:35I'll click out back to canvas, and I'll

18:37hit this little play button. And what

18:39that did was it quite literally added

18:41the transcript to this row. So if I hit

18:44refresh, you're going to see that entire

18:46transcript pull in to our data table.

18:49And I have it all right here in N8N. So

18:51now we can use this entire thing within

18:54our AI agent in order to have it make

18:56decisions on whether we should watch

18:57that video or not. And so this is the

19:00automation of the input to the database.

19:02As you can see, we have three database

19:04nodes that we used this entire time, and

19:07native N8N tables make this so much

19:09easier than it used to be. But what we

19:11can do is we can feed all of that

19:13information we just received, which

19:15could be happening on a timer, by the

19:17way. You can set this up to go every

19:19minute, every hour, day, week, and then

19:21you can just activate the workflow, and

19:22it can be running for you while you

19:24sleep, and then emailing you whether the

19:26video is good to watch or not. And so

19:28far, we're doing really good. We've

19:31gotten down this first half of the

19:33process, where we've taken an input,

19:35which is a YouTube video, we've scraped

19:37the transcript, and then we've uploaded

19:39it to a database. So now, all we need to

19:42do is attach an agent to it,

19:45help it understand how to use the

19:46database, give it some of our goals, and

19:49give it instructions and guidelines for

19:51how to output its prompts to us. And

19:54it's going to act on behalf of our

19:56goals. So, if a video comes in that's

19:58completely crazy, it's not going to tell

20:01me to watch it in the alert. So, let's

20:02build this agent now, and there's a

20:03couple of things you should do when you

20:06build an agent. Number one, what I like

20:08to do is go through a five whys

20:10analysis. So, why am I actually doing

20:13what I am doing? I've created this GPT

20:16called the root cause GPT, and what this

20:19is going to allow you to do is discover

20:20the root cause of why you want to

20:22automate what you want to automate. And

20:25this is really important. It goes

20:26through a five question loop that asks

20:28you why five times. So, that when you

20:30hit let's automate, you can truly

20:32discover the reason behind why you want

20:34to automate that thing. So, if I hit

20:36let's automate and I'm doing this

20:37YouTube transcripts as an example, I can

20:40go through this process. As you can see,

20:42it says, "Great, let's kick this off.

20:44I'll ask you a series of short why

20:46questions up to five layers deep to

20:48uncover the real root cause behind the

20:49automation you want." "What input do you

20:52want to automate or capture?" I can just

20:53say, "YouTube transcripts."

20:57And I can send that off. And now it's

20:59going to go through the why analysis.

21:01Why number one, it asks you, "Why do you

21:03want to automate YouTube transcripts?"

21:04Or why do you want to automate whatever

21:05you want to automate? And then you can

21:07just answer honestly, and it's going to

21:09really uncover the true reason behind

21:11why you want to do what you want to do.

21:13So, I don't keep wasting time on content

21:17that doesn't help. I could send it off.

21:20Then it gives me why number two, "Why is

21:21avoiding wasted time on unhelpful

21:23content important to you?" And I'd say

21:24because I need to spend my time creating

21:26instead of consuming. I'll send it off

21:28and it will ask me my third why, "Why is

21:30it important to prioritize creating over

21:32consuming?" I can put because it's what

21:34helps my community members and viewers

21:36learn most. Then it asks me, "Why is

21:38helping your community members and

21:39viewers learn the most important thing

21:41for you?" I can just keep responding to

21:43these whys and uncovering the root cause

21:45of my problems. And some of these

21:47questions get honestly pretty hard to

21:48answer. So, you just want to keep on

21:50answering.

21:51And after I answered the last why, it

21:53says, "Got it. Thanks for going deep."

21:55So, if I put this together, your core

21:56goal isn't just automating YouTube

21:58transcripts. It's really about freeing

22:00up time from low-value content

22:01consumption so you can focus on creating

22:03teaching material that helps your

22:05community learn faster, which is

22:07absolutely correct, and I'm amazed every

22:10time I go through this system. Then I

22:12can say, "Yes, that sounds exactly

22:16correct." And I can send that off. And

22:18then it gives me an entire brief that I

22:20can use when creating my AI agent. And

22:22it gives you your root goal, which is

22:24really important because this is what

22:25you want to give to the agent in its

22:27system instructions.

22:29So, if I go back to my N8N workflow

22:31here, what I want to do is I want to hit

22:33this plus button and type in AI agent.

22:36And when I select this, what you're

22:38going to notice is you have something

22:39like a prompt, and you also can add an

22:41option and create a system message.

22:44These are your custom

22:46instructions for the agent. This is

22:49where you show the agent how to act on

22:51your behalf, and this is where you input

22:53the information that the agent needs to

22:56use in order to act on your behalf.

22:58So, for source for prompt, I can change

23:00that to define below. And for the prompt

23:03user message, I can make that an

23:04expression, open up that window, and

23:06then I can just put YouTube video title,

23:10and then I can pop in the video title. I

23:13can enter it down a few times, put

23:14YouTube video author. I can enter in the

23:18YouTube video author, and then I can put

23:21YouTube video transcript,

23:25which is really the most powerful part

23:27of this entire thing because we can have

23:29AI look through the entire transcript to

23:31see if it's worthy and worth our time

23:33based on our goals. And our goals will

23:35be put within the system message. So,

23:38right off the bat, we can copy our North

23:40Star goal that was created from the

23:42transcript automation. It's up top, and

23:44you can scroll down, and you can even do

23:46a PDF download if you want at the very

23:49bottom. In the next step, what you want

23:50to do is you want to make the system

23:52message an expression and open up this

23:55window, and just put your root goal up

23:57top. And we're just going to kind of

23:59move this around a little bit, but we

24:01just want to make sure that this gets in

24:03here because this is truly the reason

24:05that we want to automate these YouTube

24:07transcripts. And the cool thing is is if

24:09you go to the description and you use

24:11this GPT, which is free, by the way,

24:13every one of us is going to have a

24:14different reason why we are automating

24:18YouTube transcripts. Some of you may

24:20just want to learn a bunch of

24:21information a lot faster, and you don't

24:23have time to watch all the videos. Some

24:24of you just may be business

24:25professionals who want to take some

24:27concepts or some business models from

24:30the YouTube videos and extract those. It

24:32doesn't really matter what your goal is.

24:34This is where this thing gets very

24:35custom, and where we can all start to

24:37learn a lot by building this AI agent

24:39out. So,

24:41in the system message, I use a framework

24:43called the IOE framework. I made a full

24:46video on it going over how to create the

24:48best instructions to make AI sound

24:50exactly like you and follow exactly what

24:52you tell it to do. I'll leave that video

24:54on the upper right-hand corner if you

24:55want to watch a full guide on creating

24:57instructions. But for now, let's just

24:59quickly create these so you can watch

25:01how this works in real time.

25:04And you can go as in-depth as you want

25:05here. So, I'll give it a quick roll, and

25:07then I'll just say,

25:09"You are

25:11a YouTube transcript agent." And then

25:13I'm going to tell it what it will

25:14receive. I say, "You will get a YouTube

25:16transcript video title and author, and

25:18your job is to decide whether it's worth

25:20Drake's time to watch the video or not

25:22based on his goals." And then I can put

25:25Drake's North Star goal, and that's why

25:29this is important. So, now the AI agent

25:31has context into what my North Star root

25:33cause goal is for automating these

25:35transcripts. And here you honestly need

25:37to think about what is going to have the

25:39biggest return on investment in your

25:41life when consuming content on YouTube.

25:44And why do you actually watch YouTube

25:45videos? Do you like taking little

25:47pointers from them? Do you like watching

25:49full guides like this and following

25:50along in-depth? This is where you need

25:52to let the agent know exactly what you

25:55want it to base its decisions around

25:59when telling you whether you should

26:00watch a video or not. This is like going

26:02inside of your brain and creating set

26:05clear instructions for it to follow. So,

26:07under here I'm going to add a knowledge

26:09section, and this is basically going to

26:11be things that I want it to know. I'll

26:13add a couple of these pound signs in

26:15order to signify that these are headers.

26:19And I can even in the role put, "Use the

26:21knowledge section in order to tailor

26:24your answers and responses." And now I

26:28can just start adding little knowledge

26:30points. I can put knowledge and then in

26:32parentheses information about Drake, and

26:37then I can put

26:38"He runs an AI community where he

26:41teaches evergreen AI skills." You can

26:44even put things you're not interested in

26:45here. I could say something like he

26:47isn't interested in local N8N videos.

26:49Maybe I'm just somebody who wants to

26:51stay on the cloud. I can put like

26:53Drake's main automation tool is N8N.

26:58I can put something like he loves full

26:59agent builds and replicating them for

27:01his community. So, maybe if a video

27:03comes in and it's like full N8N agent

27:06build for research or whatever it may

27:08be, it will recommend that video to me

27:10since it now knows that I love that

27:12because I like replicating them for the

27:14community. That's a high return on

27:16investment task. I could keep adding to

27:18this knowledge section, and I recommend

27:20you do as well in order to actually make

27:22this agent very valuable in giving it

27:24insight into what your goals are, what

27:26do you like, what don't you like because

27:28it will seriously change the output of

27:31this agent anytime a transcript goes

27:33through it. So, the next section I want

27:35to create is the output format section.

27:38How should the agent output its

27:40response? And how should it use that

27:42output format every time it sees a

27:44transcript? So, I say, "You are to

27:46output answers in a specific output

27:48format each and every time. Here is the

27:50consistent output you are to use." And

27:53instead of consistent output, I'm going

27:54to put, "Here is the JSON object you are

27:58to use." Because this is how you get

27:59consistent outputs is by using JSON

28:02objects, something that we teach in our

28:04agent building course. So, I'm going to

28:07do one real quick, and it's very simple.

28:08Don't be scared by this. I'm going to

28:10type in two curly brackets and enter

28:12down, and then I'm going to put quotes.

28:14I'll put summary, and then I'll arrow

28:17over, put a colon, and then two empty

28:22quotes with a comma. And so, this is

28:24basically saying summary and then

28:26response. And now I'm going to add

28:28another object, so I want to add that

28:30comma. I can put ROI

28:33_score,

28:36and then again, colon,

28:38open quotes, and then a comma since I

28:40want to add one more thing. I can just

28:42type in video_compliments,

28:46and then I'll arrow over, colon, and

28:49then double empty quote. This time with

28:52no comma since it's my last object. So,

28:54these are going to be the responses that

28:56the AI agent generates every time. It's

28:58going to give me a video summary, an ROI

29:00score, and how this video compliments

29:03what I want to do. And so, I can tell it

29:05how to use this. So, I described quickly

29:07how to use this output format here. I

29:09say, "You are to generate a good overall

29:11summary of each transcript going over

29:13the main concept and what happened in

29:16the video. Furthermore, you are to give

29:18an ROI score of high, medium, low, or

29:20horrible depending on the return on

29:23investment this video would have for

29:24Drake depending on his goals. Lastly,

29:28you are to provide video compliments.

29:30This is how the video compliments or

29:32doesn't complement what Drake wants to

29:34do it or his interests. So, what I'd

29:38recommend is adding a little bit of

29:40goals or a little bit of what do you

29:43actually want out of a YouTube video

29:45when you come to watch? What are you

29:47interested in? So, under my North Star

29:49goal, I can even add a few more goals. I

29:51can put other goals from YouTube videos.

29:58And then I could just start entering

29:59some things that I want out of YouTube

30:01videos. A couple things I added are

30:03wants valuable insights that can help

30:05his AI community become irreplaceable,

30:07wants videos that are new updates so we

30:10can stay atop of the trends, things like

30:11that. And now what you want to do is you

30:13want to copy this JSON object that you

30:15created. Click outside your AI agent,

30:18select require specific output format,

30:22add the output parser.

30:24Then you want to select structured

30:25output parser here. And then replace

30:27this JSON example with the example that

30:30we gave. And now what we can do is we

30:31can add a chat model to this. So, I'll

30:33add a chat model. And I can select

30:36OpenAI in the bottom left in the bottom

30:38right-hand corner. Then you can connect

30:40your credential. It's very simple with

30:42OpenAI. I'm sure you've actually done

30:44that before. I'm just going to put this

30:46on one of my accounts. And then I'm

30:48actually going to put this on GPT 5

30:51mini. I recommend using a good model for

30:54this because remember, this is going to

30:55be your brain. Transcripts take up a ton

30:58of context, so it has a lot to look

31:00through. And I just wouldn't cheap out

31:02on the model or go for speed here

31:04because it's going to be making big

31:05decisions for you on any automation or

31:08agent that you build, not just this one.

31:11And so now we can really test this out.

31:12I'll go to my data table and I'll delete

31:15whatever we have in here. So, I'll

31:16refresh just so we can run through all

31:18of these as examples.

31:21I'll delete.

31:22And then what we can do is we can go run

31:24this automation. So, I will execute the

31:27Productive Dude one. This Productive

31:29Dude one looks like an N8N automation in

31:32same shorts automation tutorial. So, it

31:34should recommend this video to me since

31:36I love full agent builds, but we will

31:39see if it does. It's going to review

31:40that transcript. It's actually reviewing

31:43that transcript right now. Looks like it

31:44already added it to our database with

31:46the the transcript in there. So, if I

31:48refresh our database, as you can see,

31:50the transcript's already in here. The

31:53agent is just reviewing it currently.

31:54So,

31:56uh as you can see, it's completed now.

31:58And it created in our structured output.

32:00So, every time it will generate summary,

32:03ROI score, video compliments, which is

32:05beautiful. So, I'll click into our AI

32:07agent and see what it created for us. It

32:10says ROI score high. So, it really wants

32:13me to watch this video. It says, "This

32:15is an end-to-end N8N workflow

32:17walk-through that builds an automated

32:18short-form content generator, goes over

32:21everything he did in the video, shows

32:23step-by-step node setup in N8N, gives me

32:26the pricing of everything, all the tools

32:28that Carter used in that video." Then it

32:30gives me why it compliments my goals. It

32:32said, "Why this compliments Drake's

32:34goals? Full agent build. It's a complete

32:36replicable agent automation pipeline.

32:40Exactly the type of end-to-end agent

32:41Drake likes to replicate and teach." So,

32:44it's actually looking at my goals and

32:46recommending this video to me based on

32:48what I like. So, we're kind of creating

32:49our own algorithm in a way, but we get

32:51to control the algorithm. And so now

32:53that we have all this information

32:54pulling in, I mean, I could keep

32:56testing. I'll go send an N8N YouTube

32:59video right now, which I think the N8N

33:01video is install N8N locally. And if you

33:05remember in my goals, I said I don't

33:07want to do local N8N stuff. So, it

33:09should turn me away from this video.

33:11While this is generating, I will just

33:13add a quick Gmail node to the end of

33:16this. You can easily connect your Gmail

33:18account and send a message to an email

33:21that you want. If you want to create a

33:23new credential, you can sign in to the

33:24Gmail that you want to uh send emails

33:27from. And you can send this to any

33:29email, which is really cool. But now

33:31we'll create a little message that

33:32notifies us whenever one of these

33:34channels creates a new video. Let's see

33:37what it said real quick, though.

33:39It says, "ROI score low. It's about

33:41basic local install. Useful as a quick

33:44reference for beginners. Recommend

33:45skipping unless you need a short how-to

33:47clip to onboard beginners in your

33:49community to use as a simple install

33:50guide." So, very cool. So, then I would

33:53just wouldn't watch that video from that

33:54channel. But now we can notify ourselves

33:56so we have some sort of notification

33:57system so we can stay off YouTube but

33:59still get notified when our favorite

34:01creators make content. And we'll get

34:03suggestions from our agent whether we

34:05should do this or not. So, now under

34:07send a message, I'll send this to my

34:09test email. For the subject, I'll put a

34:11little emoji. I'm going to put the

34:14siren. And then I'm going to put new

34:16video alert. And then in the email type,

34:20I'll select text. And then I'll create

34:22the body of the email. So, I want to

34:23make this an expression. And we can use

34:26all the data in this expression that

34:28we've picked up throughout this entire

34:29workflow. So, I'll drag in the channel

34:31name that created the video. I'll go to

34:33add transcript and I'll pull in the

34:35author. I'll say

34:38uh author just uploaded a new video. And

34:43so, in the email, it'll look like this.

34:46Whatever the channel that uploaded is,

34:47it will have it right there. And then I

34:49can enter down a few times and I can put

34:52the video is maybe I'll put quotes and

34:56then drag in the video title here. And

34:58now we'll enter down and put here's a

35:02quick summary.

35:04And then I can put the summary down

35:06here. So, I'll drag in the summary from

35:08the AI agent, which it gave us a summary

35:11because we did the consistent output,

35:13the JSON object. So, we can now drag

35:15different things that the agent

35:16generated from a single response. I'll

35:19drag in that summary just like that. And

35:21then I can put your return on investment

35:26by watching this video

35:29is add some suspense there. And then

35:32I'll drag in the ROI score and enter

35:35down.

35:36So, now it'll say this, "Your return on

35:38investment for watching this video is

35:39low" since it's the uh

35:41quick step-by-step install guide with

35:44Docker. And then I'll put here's why.

35:48And then I'll put in the video

35:49compliments. So, this section tells why

35:51it's a good fit or why it's not. This

35:53gives me just the general score, the

35:55summary, the video title, and the author

35:58of who uploaded the video. So, now it

36:00will send me emails

36:02whenever this automation starts, which

36:04is anytime one of these channels uploads

36:06a new video or on a schedule. Another

36:09quick little hack, if you don't want N8N

36:11attribution at the bottom of your

36:12emails, you can go to add option, append

36:15N8N attribution, and turn that off. And

36:18now

36:19we can test out this full workflow. So,

36:21I'll go to my database and I'll refresh

36:24it and delete everything real quick out

36:26of this table. Hit refresh.

36:30Delete everything out of the table. And

36:32we'll run this from the top. So, let's

36:34say AI Foundations comes out with a new

36:35video. We have this set up every minute,

36:38so it's watching my channel literally

36:39every minute. So, it'd be good for

36:41competitor research as well. I'll

36:42execute workflow to send off a test.

36:45And it will get that transcript for me

36:47and start reviewing my video, which my

36:49latest video is actually a promo for the

36:52AI Foundations community. So, I'm not

36:54sure how it's going to handle this one,

36:55but we can see how it handles it when it

36:57gets sent to my email. So, I don't have

36:59to be here for this to happen. I could

37:01just activate it and let this run. But

37:03as you can see, it emailed me new video

37:05alert and then it gives me channel name.

37:08So, AI Foundations just uploaded a new

37:10video. The video is and then it gives me

37:12the video title. Gives me a quick

37:14summary. It says, "My return on

37:15investment by watching this video is

37:17low" because in my goals, I put that I

37:19only want videos that are full agent

37:21builds, videos that are going to allow

37:23me to build things for my community.

37:25But it gives me an exact quick summary

37:27and why it aligns with this return on

37:30investment score. What I could also do

37:32is I could add the the link to the video

37:34in here as well. If I open this up and I

37:36just say something like link to video,

37:40I could paste in the video URL

37:43by dragging it from the add transcript

37:46node from earlier just like that. And so

37:49now, if I execute that step real quick,

37:52it will send me another email just like

37:54that with the video URL down here.

37:56Perfect. And you could customize this.

37:58You could make headings, forms. You can

38:00do so many things in here. It's

38:01absolutely insane. But this is the

38:04system. This is how to go from input to

38:08agent making a decision for you. What

38:10we've just done is we've eliminated the

38:12need for me or any of you to get on

38:14YouTube if you don't want to anymore.

38:17And I recommend using that GPT, the five

38:19whys root cause GPT I left in the

38:21description for free. It is truly going

38:24to help you understand why you do some

38:25of the things you do. And if you start

38:26using this system and replacing a lot of

38:28the inputs that you have to make

38:29decisions on on a daily basis, I know

38:32that you can save at least 2 hours a

38:34day. And if you do that, it saves you a

38:36month out of every year. If you can

38:38figure out a way to save 2 hours per

38:41day. So, that's all I have for this

38:42video. If you enjoyed, please drop a

38:44like. Let me know in the comments, what

38:47is the next input that you want to

38:48tackle? And I hope I still see you on

38:51this channel after the automation we

38:52built together today, but at the very

38:55least, at least put me on your RSS feed.

38:58That way your email gets notified

38:59whenever I upload. I would highly

39:00appreciate it. Like I said, I have a

39:02full free video in the description going

39:05over everything our private community AI

39:07Foundations has to offer. Probably the

39:09most expensive thing you can do right

39:11now is not investing in everything you

39:13can do to learn and master the skills of

39:16our time. Think about if you would have

39:17learned SEO or email marketing back in

39:20the 1990s and early 2000s. You'd be

39:23sitting on top of an empire. So, these

39:25are the AI skills that you can learn

39:27today with us at AI Foundations. So,

39:30hope you enjoyed this video. If you did,

39:31please drop a like and subscribe. I

39:33would highly appreciate it. And with

39:35that being said, I'll see you in the

39:36next one.