Transcript [EN]: 5 AI Agents that can make you $1k/Day (Full Demo)

imagine going from zero to $1,000 per day how by creating AI agents that deliver high value services like expert level email newsletters content marketing or even SEO these services are so valuable that each client will pay you5 to $10,000 per month but what's crazy is you can now deliver these high-end services in minutes using clever AI agents and if you think this is just for tech wizards think again we're breaking down exactly how you can build MCP enabled AI agents this new method essentially gives AI superpowers allowing it to browse the web scrape social media and perform complex tasks just like a professional would so today you'll learn step by step how we've built five such agents automating everything from those high ticket newsletters to in-depth social media analysis to content idea generation SEO reporting and even scripting your YouTube videos we'll show you how to build these agents and how you can make serious money from them by launching brand new lucrative services or even by saving these costs if you want to run them in your own business so if you're ready to turn AI into a serious revenue stream let's get into [Music] it okay so the first opportunity we're going to be doing is an email marketing manager who manages an email newsletter and now this is something we have personal experience on Gail as we've been running our own email newsletter for over 10 years now and you do pretty much all of this singlehandedly and it's also to very high standards so what's the secret here yeah and I spend about two to three hours a week to write an email i kind of open my inbox here so you guys can see these are the replies I'm getting from people these are people that engage with the newsletter read this and essentially like it and ask me questions or whatever right and it's like you know lots of like thanks you just to show you like this is not AI slop people like this stuff and engage i think that is the difference between what we're talking about throughout this whole episode today and most other kind of AI use cases is they give you pretty like cookie cutter like here's how to make a shitty article or post on social media but this is something which has driven us a lot of revenue over the years uh and engagement drives serious engagement people actually like it right like Simon he's the CEO of a big company right for example he replied here we get good people from this I even have our email marketing tool just to show the open rates are pretty healthy overall like everything you can see everything's green and great and good 30% open rate on the list to tens of thousands of people i would argue that's pretty good especially given the clickthrough rate as well is also very high to these things so that just gives you an idea that this is not slop and so the way I do that is I prepared the newsletter there's still a bit of a human walk in there but I'm going to show you what I give to AI so I walk in notion and basically this is actually the full one so let me give you the last one where I literally just wrote a bunch of so I'm just putting random bullet points of what goes through my head of like what should be talked about in the intro not well written essentially this is a brain dump right you're not even structuring this yeah I don't care so I'm like tbh this is the first three months of the year catchup like this was a long intro like sharing about our current state etc and I'm just kind of like dumping sponsor i didn't even write it because I let AI write it i'll show you how I do this later i'm like main content okay AI is cool i use it a lot da da da it's just a bunch of bullet points it's not that long as you can see like this takes 5 minutes to make a note like this same for the links you know there's a section where I share links like cool links for the week or something and I put the link but I don't even explain what it is because I will have AI do all the work and open these links read what's behind understand what it is and write the description for people same for the tool of the week you can see I say tool of the week descript i'm like oh research it and then re and then write it why because I know AI can just go and do this for me so it's like I'm talking to AI a little bit like I would talk to like an intern or something like an intern i could see yourself like if you were to give this to an email marketing manager and say like hey go produce this email newsletter that would be their task so why don't you show us what's actually going on behind the scenes to get it from this to the end result and then we can look at how much that would cost us to do otherwise if we were hiring a freelancer or someone full-time or even an agency to do this for us because that's quite interesting the amount of money that we've saved here yeah i mean I just want to say like these notes are essentially my expertise right it's like I know some of this stuff that AI won't find but everything else I can Google I just let it do it basically so then I take all these notes to typing mind and I have an agent that's pre-written so in this case I don't even have to hold on for those not familiar with typing mind do you want to just explain what that is typing mind let's say it's kind of like an advanced version of chat GPT that allows you to connect a bunch of tools and allows you to connect every single AI model that there is on the planet basically so I'm able to switch the model i'm able to like you will see in this case I give it access to tools like perplexity like fire crawl etc so that the AI can do all the research I ask it to do it's a better more professional chat GPT let's just say that and I can kind of like instead of being locked to Netflix we're using Pirate Bay and we can just get all the content in one place it's not we're paying for this it's a professional tool and completely above board but but the good thing with typing mine is there's no subscription so you buy one off and then you just pay your API costs and that's about it basically and so like because I have pre-prompted it and I'm going to show you the prompt in a second i just say like use these notes and write the newsletter and I just dump my notion notes that I copy paste you can see like literally not well formatted i don't give a and then AI essentially has a planning phase where it just kind of like essentially digests what I give it thinks about what it's going to have to research so it finds the knowledge gap that he has like oh I don't know about this script he didn't give me the information i'm going to have to research this i need to research open router and their free tier i need to verify some information about OpenAI and I need some details about the sponsor section because it wasn't provided right and so it thinks about which search queries is going to do it rethinks of the style let's just break down a few things here you've then selected Claude as the model specifically because it's good at writing or because it's good at figuring out all these questions and using these tools is good at writing but also at tool usage model for tool use so because I'm asking it to use all these tools you can see after that like it's starting to use perplexity to do some researches so it's like oh descript video editing tools features and benefit and it just like gets the result from perplexity that essentially fills the context with like what is the script and what it does and what are the good stuff about it i didn't have to do it myself you're letting one AI use another to improve its ability to answer your original question is that right yeah and then also I know that it's going to be able to find it so I don't bother in my notes i save time that way the same way as you would give the low-level task to the intern you'll be like go figure this out but the top level expert stuff I would probably put in my notes because I know it's not going to be obvious in the googling or something like this that's essentially what's happening here and so like yeah I knew it can find the features of the script you can see it like went on the homepage it went on the pricing page it read some reviews it did all of that and then it just gathered all the information for itself very good and then after that it searched for like the open router free limit because that's in the body of the newsletter and I don't think I precised it maybe I forgot to put in my notes so it kind of identified that and started to do some research and you can see that for example it found that in a free tier you get 50 free calls but if you have more than $10 balance you get a thousand free calls so that was kind of like a nugget that we got for the newsletter found it on is that something you knew before or that's something i discovered it as I put together the newsletter it's actually true it's like if you check our bank statement you'll see a top open router after so great again filled the gap did the research and I have all the sources here so if I wanted to you know verify the work I could just open these URLs and look at this basically this is all really possible because of the reasoning or thinking models that we've talked about before on the show right untrue we can do it with a non-reasoning model it's just like tool calling is different from reasoning this is so this is different so it's like a model can call a tool then do an answer then call a tool then do an answer essentially just chain prompts right is it not thinking about which tool to use though in this case it is in the prompt so usually process so for agents you need to tell them in the prompt so we're going to go to a prompt in a second but I tell it use perplexity here's the tools you have here's the list of the tools here's when to use them in the process blah blah blah you prompt for it because otherwise agents can do crazy things like I've had an agent do a 100 tool calls because I didn't specify that was like a $7 mistake basically in in API calls that can happen so you you guide them you want to talk about how it interacts with tools or are you covering that later then let's just finish going through what it does and and what the first output so again it searched for the perplexity it searched for a studio and you can see it did not reason actually I used the non-reasoning model reason why wise because for writing you don't really need reasoning and then you can see that the output I get is okay the sponsor it made it up because I didn't give it anything what people got here was mostly what they got in the end except it was a bit too long so I kind of get the first draft but it still was pretty good I think and then I get the first draft I just make a list of things to change basically I like it but like change these three things and I just basically this is like you're giving feedback to your editor who just done the first draft and for some stuff I'm like oh I don't like the way you did the math cuz this was like math related like you know was like cost of API etc i was like a I don't like how you calculated that so I did some research on like you know what's the current price for the permillion token for this API to recalculate the value etc that it keeps going and so basically you can do that and when you're happy you basically is this good enough to send like after that first round of revisions it depends it depends sometimes literally after one round of revision I just send it I think last week that's what happened uh some weeks this one was a little bit longer reason why I disagree with the math basically so it's like I disagree with the math and it just kind of like we had a few rounds but it's pretty quick like it's a long chat but I don't write very much i think it was like four rounds and I sent it total time spent about an hour basically um between writing this and then just going back and forth rereading again I didn't read everything every time it would output i would just reread the problematic sections so that's pretty much like the way and then also on typing mine you have this canvas so you can actually click on this button and if you wanted to rewrite a sentence you could actually select this enter your prompt and be like make it shorter whatever and then it would just kind of rewrite only that sentence that's really cool well it's kind of like leaving a comment in a Google doc and then AI is just automatically fixing it for you and then it goes back to chat format so you can alternate between doing canvas and doing chat format now let's go to the prompt if you want to see how this works so when you say going to the prompt we're talking about the kind of system prompt behind this because you gave it a prompt initially which was but there's no prompt really i said write a newsletter the reason you were able to give such a short instruction is because you've explained what the newsletter is here mhm yep so like that's the thing it's like a pre-prompt when you're going to do repetitive tasks you make like one of these agents and you just use it again and again and that's the reason I can do the newsletter in an hour and a half let's just pause there for a second because the word agent I feel gets used in a variety of different contexts and I don't really know exactly what people mean when they say an agent so when we're talking about an agent what do we mean here it's either a pre-prompted chatbot that has very precise instruction but I would say in 2025 an agent will be a chatbot that has access to tools as well search that can open URLs that can scrape stuff as we'll see later that can do a bunch of stuff and then just give you better results because it's more encode into real life data so in this case like this is an agent because it's pre-prompted it has access to complexity and it has access to fire crawl which allows it to open web pages and read the content on these web pages so that's pretty much it so I have a pretty long well structured prompt and to be honest I'm looking at this prompt I'm like I do better ones now you know so it's like I'm going to update it but like first thing is explain what the top level goal is it's like hey we're writing a newsletter here and our main goal is to just give practical info to people we focus on AI and automation it talks about me as well a bit is engaging the sprinkle with G breton's signature dry humor and punchy and conversational and then sponsor section the same way you would introduce a new employee to their task after that it has instructions again it has three phases it has the first phase which is the planning phase which you know it's thinking about so it's like complete it's thinking about the flow of the newsletter it finds which highlights like which actionable insights are going to be shared it finds the gaps in the notes so that it thinks about like hey I need more information about this to fully write about this etc it thinks about the search queries to fill the gaps it will think about how to make it sound like me then it will rethink the whole structure after it's done with this that's the first phase the second phase is research so again it has the search queries so based on the gap analysis do your searches perform the search you have access to an AI powered search so feel free to ask multifaceted question the point is because it's per flexity it can ask like a prompt it doesn't need to ask a single keyword the same way you would use Google for example so it's like I tell it hey it's powerful use it that way you know okay so just to zoom in on that then you have to tell it how to use the tool and then you also have to connect the tool somehow as well yeah we'll talk about connecting the tools after if you want and for this example we're not even using MCP so it's very easy then I tell it every time there's a URL mentioned on the notes open it open it read it and understand what this is about um again it fills it context right these models have lots of space in their context just fill it up without any effort from me i can see though that the instructions are a little bit more basic than a human would use you could just say "Oh go use this perplexity tool." And they would probably figure it out whereas you say "Oh if you find something you have to open it." It's very kind of you know handholding instructions here yeah because otherwise you let the agent decide and it's not always going to go the way you want so you need to put guard rails around the process so that it mostly follows what you want it still has a little bit of decision power but again the more decision power you give it the more random the output is going to be do you have to give it a maximum number of attempts that it can try and use perplexity so it doesn't end up you know using it a thousand times and costing i think it has it later but let's see in this case it doesn't do it too much because it kind of brainstorms the content gaps first and as long as there's not like a thousand content gaps it's not going to do it but I've had tool calls like literally run in a loop and I had to stop it because it would just destroy our credits basically um you tell it like hey it's fine if you take your time in a research just make sure you do everything and then organize it and make a content plan right and then once it makes a content plan it's just the writing part right focus on the style focus on all of that etc and then I gave it like an output format so it's mostly like the newsletter is a set format so there's you know a subject line preheader a hook and an introduction the main body with the value then there is like the tool of the week is the links that we want to share and then this prompt that makes people ask people to reply basically so that's pretty much it and then in general if you're doing any kind of repetitive content having a structure is helpful but especially so with AI like you're much more likely to get what you want when you clearly define the structure of it here and that applies to any type of content you're creating yeah then it has some rules so like you know 600 to 900 words in total sometimes we go a bit over but roughly it sticks around that and then again a little bit of information to write good newsletter like you need a strong hook like follow with something personal be genuine and then even examples of sentences like holy etc like you know it say stuff like that so that it just sounds like me then it's like you know focus on actually giving value to people actionable stuff etc there's also like visually how it looks you know like single sentence paragraphs do use line breaks all the kind of stuff that I would want in a good newsletter same with language patterns i actually gave it transcripts of me on the podcast and then I just kind of wrote a lot of sentences I use like I made this mistake you're better off you want me to etc kind of capturing the essence of you yeah of me exactly that's the point so it's like it gets like it's not perfect but it's closer to it than I don't have to edit very much basically then it explains you know the subheadings have emojis like you know key insights at the beginning is a bit like TLDDR the beginning of every sentence etc etc formatting guidelines voice a lot of things that I want the AI to think about and this is for this newsletter if you did something else you would do it differently but you can look at this and after that what's important is at the end I gave it examples of past newsletters like full newsletters how many examples do you include two because they're pretty long already i don't attach any file anymore i used to attach like these big swag files to these agents i don't anymore i find one or two good examples are good enough basically so that's kind of like the prompt for it and then in typing man you have this thing called plugins and in plugins you can connect a bunch of tools there's this MCP tools that we're going to be talking about in a second but what I use here is I use this Perplexity one so you just I'm not going to go in the settings because you would see my API key but you go on perplexity.ai you get an API key and then you put it there basically by putting an account i think you get five bucks when you sign up and I use this one called fire crawl essentially allows the chatbot to request the content of any web page in markdown format so without all the in the HTML and so on and so that is that particularly beneficial for AI to have it in that format does it just get less confused or you can do it cheaper by just like pulling all the HTML no you can but then you're going to put a bunch of HTML into your prompt which is going to cost you tokens and Firecal gives you 500 free scrapes per months so for a newsletter for anyone who does like manual work like this it's plenty you never use that it's only if you're going to if you're going to start automating heavy data you're going to run out so and just quickly explain why are we using perplexity search here as opposed to any other type of search is it just because it summarizes things so well or is there any yeah it just reads like you know 10 15 sources and it just gives you the info that you're asking and then the AI can think of like the highle thing they want and then a sub AI basically does some it's like an agent using an agent which is cool and it's pretty cost effective if you use the pro model it's like two three cents per query um I have Perplexity Pro with my Revolute account so as part of it I actually get $5 of API per month so I pretty much never pay for Perplexity for like this kind of usage at least once you scale up with automation NA10 etc different story but for typing mine it's plenty already you can choose which model you use on Perplexity so you can ask for deep search if you go into settings you can see here I have Sonar Pro but you could put Sonar deep search and it would do like multiple searches and give you a much bigger report if you wanted sometimes I switch it you just change the model name here and it will give you the different perplexity on the other end pretty handy pretty powerful imagine an AI agent that can do like multiple deep searches read all of it and write your content based on that that's powerful so if you were to put a price if you had to guess how much it would cost to hire someone to do all of this newsletter work for you per month what would you say how many newsletter per months once a week well let's say one a week as we're doing at the moment did you research it yes okay it's quite difficult let's say US worker i would imagine like maybe two $300 per newsletter so let's say $300 per newsletter times four 1,200 times 12 so that's like 24k plus two time 200 time 12 so 1,200 like 14k 15k okay well the average annual salary of a US email marketing manager is over $100,000 including compensation okay on Upwork globally they make $25 an hour in the US a junior would make $30 to 50 and a senior would make $100 to $120 per hour if you go to an agency this would cost $1,500 per email written minimum and 5 to 10K per month for the whole end to end setup and that's a weekly newsletter at best you start going daily and producing a lot more then the numbers go up significantly so this is an expensive thing to get two hours two three hours a week for this you know there are people teams of people whose job it is to do what you've just shown that you do in like couple hours a week so yeah it's kind of interesting and again it has results right people like it this is a newsletter people like reading the engagement has gone up a lot since we've gone with this format has insights from us again I guess if you hired someone to do it they would need to collect insights from people that might be the most annoying but again I would probably make some kind of AI prompting them and then they would just reply would be transcribed and I would just use this as context there's a way to automate that part too quite a bit i think interviewing the kind of person or the knowledge expert or even one a month I've seen some content agencies do and then just they pull out multiple things from that okay well that's quite valuable definitely and it's not super hard to build like a lot of people have email list a lot of businesses have email list and don't email them engaging stuff so I feel like if you wanted to sell that to businesses that would be quite strong and that could definitely boost their sales replace the sponsored section with like a promo for whatever the business is selling and then sell that to an ecom like you know I'm thinking about again this mental sleep guys because we kind of take them all as an example like sleeping masks they could definitely do like some kind of newsletter with new research on sleep and optimizing your productivity and that kind of stuff i can feel they would get lots of engagement and they could mix their promo with it and that would make them money great business model i think we can talk more at the end and we can compare the different business models together but let's talk about the next one now which is going to be around content formats and researching that but first let's have a quick word from today's sponsor Surfer SEO they've been a crucial part of our content optimization toolkit over the years helping us to create more comprehensive content that ranks better in search engines what's exciting is how much Surfer has grown now with over 150,000 monthly active users and powerful new AI features to streamline your SEO workflow their coverage booster automatically fills content gaps by adding relevant facts and subtopics making your articles more authoritative and complete with just one click surfer's NLP powered term suggestions focus on semantic relationships helping you naturally include related terms that improve search rankings without keyword stuffing for teams real-time collaboration allows multiple people to work and comment on the same document simultaneously plus integrations with WordPress Google Docs and even Zapier make optimizing content effortless and if you want to maintain your brand's authentic voice while leveraging AI Surfer's custom voice feature lets you generate content that sounds like you keeping authenticity intact and that's something that's really hard to get right so kudos to them for adding this check out surferseo.com to see how they can help you create better content that ranks thanks again to Surfer SEO for sponsoring this episode so let's talk about content format analysis you've got a really cool example here of something you built on Twitter using one of these i heard you mention the term MCP server do you want to maybe explain what that is before we go any further cuz I know there's a lot of people out there are going to be confused by that so I got a graphic actually that I found on medium.com that explains what an MCP is and host with MCP client that's basically your typing M chat right that's the thing that actually uses the MCP server and you connect to something called an MCP server the simplest way to explain this is it's API for AI so it means that it allows AI to use tools that you would probably normally log in to get access to so an example here was Perplexity but HFS just launched one for example so you could use HFS with it i will show an example with search console on it so what that does is AI can go in these tools can run queries on it get data and put that into their context and give you an answer that's based on the reality of things as we will see in this example we're analyzing a bunch of tweets from a guy who's doing really well like Greg Eisenberg he has a really strong podcast as well he has lots of engagement i was like "Okay how do I write tweets like him that get that many likes?" Again that's based on his personal brand it's not because I copy his stand that I'm going to get as many likes immediately but there's maybe something to learn here right if I asked Chad GPT how do I write tweets like Greg Eisenberg it's just going to be based on like his training data on what it knows about the guy which could be outdated searching it might not have access to Twitter yeah and it doesn't read his actual tweets when it answers to me so it's very vague it's kind of like what it remembers of him like the same way as if I ask you to remember something from 10 years ago versus actually going on Twitter reading the tweets one by one doing some pattern matching and understanding looking at how many likes they get and then doing some pattern matching on like what's working best for example and so like that's what we've done here let me show you the output literally the prompt is super basic i'm like "Hey analyze the last 100 posts from Greg Eisenberg on Twitter over the last three months give me his top 10 as well as a full guide of what kind of post/elements seem to be getting the most traction on the platform and teach me how to build post that get similar traction and then I say use sequential thinking that's another MCP that I'm going to show you how it works and so what happened here is that instead of thinking about this guy it actually went and scraped his last 100 tweets i have all the content in my LLM here through an app called Appify actually I used this software which we're going to use a lot through this podcast and that's basically a scraping app like you just go on there there's a bunch of scrapers you can see there's for Google Maps for Instagram for Tik Tok for also for YouTube for for anything you can think of any platform they have over 4,500 different scrapers so if it exists online there's probably an Apple which will be able to scrape it like social media you'll find everything you can think of like Reddit YouTube Instagram Twitter blah blah blah one of the real interesting things here is that each LLM kind of has some of them have unique access to places like Reddit some have Twitter some have Meta but none of them have them all so this is one way at least to reach into those walled gardens and build context into whatever AI agent or process you're building right and even if they have access like you on trap they have access to Reddit right but it will read like two or three threads and then it will just tell me something it's not going to go deep like we just did here pulling a 100 posts reading all of them and analyzing them there's no way to do this unless you're scraping with something like API for API is actually you pay per usage so actually for this Twitter scraper for example you pay per event scraping a tweet is like 5 cents so that was still not that cheap that was like five bucks basically to scrape 100 tweets i connected this through this MCP server so I connected to MCP server of API here that connected to the scraper got me the data gave it back to my LLM and I processed it that's what happened and so that's what MCP servers allow you to do is connect tools to your LLM just to explain this in the simplest possible way like the human process that this is replacing is instead of you going to Greg's Twitter reading it analyzing it maybe copy pasting some of the data figuring out what the best or most engaged tweets are AI is just going to scrape all of that feed it into whatever process you're going to run whatever you're asking it and it'll have that context knowledge so it can work on that so it's replaced that manual work that you were going to have to do not only that but I always get distracted by some tweet when I do this it's crazy on social media like so difficult to work because you just see cat videos or whatever all the time at least it doesn't happen this is much cleaner as an interface and it's like I'm focused here i like that the draw the drawback is I don't see the images but I'm going to get the links to you see my result i actually get the links to the tweets so I can just click on my result and these are real links to the stuff that actually work so then I can go check the images on the stuff that worked but I didn't have to scroll his whole timeline it just found me the best ones because it doesn't just have the content of the tweet it has how many likes it has how many bookmark it has etc and so it was able to essentially identify you know which ones it did some thinking here it was like oh I'm going to find the top 10 tweets using metrics like bookmark likes retweets and replies and then I'm going to start making my guide based on that so it did all of that it found the top 10 tweets very handy again very simple to analyze someone's social media that way oh yeah and it got lazy it didn't do it so I was like no actually do the rest of the stuff and it's like it did it and it made me a full guide on like the stuff that works practical useful frameworks big list build forward building in public that kind of stuff I got all of that and made me a full guide basically which I could feed into an LLM that writes tweets for me later now the one thing that we need to talk about is how would you connect MCP to typing mine a little bit i'm not going to go too much in detail because it's quite nerdy but it's not that difficult either and I'm not going to do it live because I've already done it i don't want to break it but basically you go in typing mind go in your settings you go in MCP here you see all the list of the servers you have connected so you see I have all these servers connected already you click on edit server because you have nothing at the beginning and you need to click on setup connector when you click on setup connector you'll get this window and you click on this device and then it will tell you to open your terminal app um which is an app that you have a little bit scary if if you start to get into coding but you just follow the instructions here and then it's going to give you that line this npx typing And so you will just copy paste this in the terminal you don't need to know what it does it will ask you do you want to do this you will click yes or Y on the terminal and then that's it and after that you will have a file that looks like this like I'm going to show it on my typing mind yep you can see it here you have this JSON here i'm not going to scroll down because I have a bunch of API keys here but the point is that you connect your servers that way some of them will require API keys so for example if I go on to the API one and you will find the one that says NPX and you will just essentially copy the one that you already have the example that you have here you will paste it into like Gemini or something and then you'll be like "Please also add this server and you'll paste this as well." And it will just update your file for you and you'll just paste it here if it asks you for an API key you can get an API key on your API thing so for example you see it says like your API token here i would just go in my API thing and I would go in my settings and there's an API key there basically and that's pretty much all you do so you give your API key to Gemini you give it that bit of text it's going to edit it you paste the text Gemini gives you here you click save changes MCP is set up the one that one text file that you got under model context protocol inside typing mind that is basically a list of all the MCP servers that has access to including your API keys okay yeah and so when it works you can essentially see the active MCP servers are here and when you're in your chat you will see them you can activate them here basically you do have to turn them on each time is that right well you can put them by default like some models are actually broken with them so like Gemini for example does not work with most of them right now hopefully it would be fixed it's a bug so I activate them right now that's specifically in typing mind or that's and other things as well okay and A10 is different so in typing mind that's why I'm using GPT4.1 i found is kind of the best one at using all these tools so 4.1 4.1 mini are the ones that I would be using all of that to say that was pretty cool i guess social media managers do this right they kind of identify what's winning with there's a specific job for this it's called a social media analyst okay in the US the average this is the average salary per year is $72,000 crazy how high salaries are in the US now actually Upwork so the freelance platform globally $40 per hour for uh social media analyst in the US you're talking about $30 to $40 an hour for a junior and $1 to $150 for a senior US social media analyst which is I'm going to start taking jobs on my pocket and like even more crazy an agency charges and I didn't even know this was a thing but an agency that just does strategy reporting and content format testing frameworks okay well charges between $2,000 and $10,000 per month so they're not actually making any content for you they're just essentially advising you which is pretty crazy i'll do the content too I guess well all of that to say like yeah it's pretty powerful the reason why these agencies still get to charge that in the age of GPT is because it doesn't have that context of actually going through all the posts etc but that bridges the gap that's where the MCP and the API is like super powerful and I think you have an edge over most people who use AI if you do this like I think most people don't do this it's not widespread and because as you said it's a little bit scary to set this up for some people it's like they won't do it but honestly the marketing kind of like arbitrage here is crazy given the value actually so again you want to analy you want to get going on social media you want to build yourself some style guys that you can also use for chat bots you want to identify what works on the platform you know it's not hard to go on Perplexity to find these people as well right be like "Hey who are the top 10 influencers in your niche?" Just go and check them scrape their stuff on the platform you want to actually optimize for and build style guides and then use that to train another i mean you could even make an agent that does that right also access to perplexity and then you give it perplexity give it all of that it will research and you give it a top level idea or like some kind of insights or whatever or like you have a cool photo that you can attach with and you just have it write the content and it will pick the format it will research it it will write it for you you can quickly edit it and post it what I like about this as well is like because you kind of work on a chatbot you still have this back and forth and this control on the final thing whereas when you fully automate it on something like an A10 that's when it kind of becomes a slop sometime because you don't have the revisions this is the right balance between automating a lot but at the same time retaining control and making sure it's quality let's move on then to a slightly more practical example cuz the last one was more about analyzing competitors and bringing insights there but if I want actual content topics like ideas how can I build something here okay so I have another example here where I had AI scrape Reddit as you can see it basically went on like this one subre I gave it to be honest I should have given it multiple subreddits but it went and scraped the top thread on our chat GPT Pro which is supposedly a bit more of an advanced subreddit and it has the votes it has everything and you can see it has the content in there as well i didn't in this case i just kind of looked for which topics did well and kind of like how many votes etc and based on that I had it kind of brainstorm YouTube video ideas so for example a lot of people are reporting problems with the memory and also chats disappearing there's multiple threads talking about that like you know chip is using chunks of a conversation there's a bunch of these and so that is an interesting thing maybe that's not the best YouTube video in this case it's not really actionable writing custom killer instruction for charge GPT real user pro tips so there's a bunch of threads here that it founds with people sharing how they put custom instructions to charge GPT to get better answers basically and so you can go through these and make a video from this that's one of them people comparing GPT search with perplexity seems to be one thing that people are asking hallucinations questions like there's a lot of drama around the 03 and 04 mini apparently they have seen it more than before so that's another topic and then wanting to see real workflows basically like people asking about that etc so that's kind of like the idea I came up with based on essentially scraping a bunch of these threads reading them and giving me ideas and again I use this ready scraper light that costs again it's very cheap like you can see I scraped a lot and I'm at $6 usage so that's the whole month so it's not too bad but like it's per GB I'm not exactly sure how much that would have been but I think this query is probably like a few cents basically to scrape a bunch of these and again I think I could have gotten better results if I put like 10 subreddits for example the point is to put a volume of data and fill the context but you can see I've used a lot of tokens here i've used 1.3 million tokens so it goes fast when you put this stuff but thanks to the caching it got much cheaper so use case of this would be maybe if there were a whole heap of subreddits in your industry that you wanted to monitor you could run this process maybe once a month or once a week and just constantly get new ideas out of it for what's trending what's people are talking about imagine if you do your like you do your reals right now right imagine if we could make a system that essentially does this daily finds what's trending researches the topic writes you a script pings you on notion and then essentially you just have to quick edit and I was thinking about that as you're shown through this example it's a perfect use case i think though we would need to dive a bit deeper and actually pull out what people are talking about in the comments as well because that's like where people are asking questions and struggling that's often where you get some good ideas from i agree so I would tweak this one this one can scrape the comments too like if I go here I read it it can also scrape comments so what I would need to do is make kind of like a two-step thing where it finds the threads first gets the URLs on the the subreddit and then after it has the list of URLs for each interesting thread goes and pull the comments and fills its context and then just basically makes the answer i think that would be a bit better i agree this is more of a proof of concept but you can see it works and it's just like this wasn't a good prompt this is very basic and that's why you can level it up basically and so another source of content ideas that people actually care about is YouTube comments like going into the comments of YouTube videos and again I've done the same thing here okay this one I wrote a better prompt actually and then I told it like scrape 300 comments per video like these are popular videos in AI and then essentially identify which ones could be good idea group them group similar comments under the same ideas and then make a list of ideas for each idea give me a oneliner where it's good the comments that actually support it and then give an idea of what a YouTube video based on this would look like basically and then I gave it that sometimes you see these tools fail so for example it tried to scrape both at the same time and it kind of errored it happens it did it one by one was that because it was trying to do it too quickly maybe or I think it timed out you can see here it timed out it happens it's not perfect to go but it's fine it understands that it needs to try again so it did it and again you can see for one video I have hundreds of comments that are pulled and it's like if I scroll it you can see there are comments here obviously lots of useless comments but that's the point it's like I don't have to read them and may I can read them for me and identify the questions people are asking the troubles they're having so it's very powerful on your competitor's videos or even on your own videos to identify what your audience wants to see so it did that and then used that sequential thinking which is basically a way to make a cheap model reason right so it decided to analyze the comments from the first video identify what is actionable and then do that and so it just basically thinks the same way as a reasoning model does except it's 4.1 mini so it's very cheap so it did all of that and then it basically come with a bunch of ideas like how to build your first agent to save time in your business again not the best headline but there's a bunch of comments that were like "Oh if you're a person in your 30s learning AI from scratch what course website or method would you use nate end tutorial for nontechnical user please i'd be interested i need a hand setting it up." So like people are asking for this it just gives you an idea of how to package it so like build your first AI agent in 30 minutes step by step for nontech marketers and business owners pretty okay title the second one was comparing top AI automation tools business Lindy versus N8N versus others again a lot of people like this was like a demo of Lindy i think that got lots of views and like the comments that inspired this was like honestly you can do most of what Lindy AI does for free with open tools tools like N8 how does this compare to another tool open source sounds great like the point is like people are asking compared to other tools so okay pretty good i think arguably better than the Reddit ideas so far do you think that's because there's like a lot deeper context in hundreds comments than just one one initial video and again you can run this as a multi-step where you give it a channel and it will find the top videos and then it will scrape the comments on these top videos and give you a lot of ideas the problem is because you put so much data you might even run out of context size even after a million context the API costs rack up fast when you put that much data into it and is this one of those situations where the context size limits start to become an issue cuz we're using OpenAI chat GPT 4.1 here is that right okay so I was thinking cuz like the Gemini ones are known to have the largest context like 2 million or something the 4.1 models are made for agentic use so they're good at tool calling and they have a larger context window um so they're pretty good 4.1 mini is okay not the smartest 4.1 is good but it can rack up like I had like queries like this that cost me $2 in API cost for example so again if you run 50 that can go quickly but again if you sell this for like hundreds of dollars like I'm happy to pay two bucks you know so same like you know idea real business use cases of a agents to automate repetitive tasks and the packages like 3A agents transforming business workflows automate emails lead and competitor analysis yeah it's okay i think like the ideas are decent so it's like this is quite powerful actually overall to and then it's pretty annoying like you could send this to read thousands of comments like you could just input 10 videos obviously you might run out per query so you might want to run them as a batch of like three four five videos with lots of comments but you can basically process a thousand comment per query do that in like three minutes and again let's look at the cost of it so scraping YouTube video URLs is $5 per thousand video URL scraped so that's fine and for the comment scraper it's $2 per thousand comments so again I mean the thing is that you're going to you can limit it to like you know the first 100 comments or maximum certain You can also say it in your prompt so for example in the prompt here I said scrape up to 300 comments on the video given and that's what it did if you look at the query it said max comments 300 and look I mean we're talking about like a few dollars here and there like the time cost to do this for yourself or a freelancer is infinitely higher so we're optimizing something but it's really not a problem if you're just running this a few times you know yeah you don't need to do it that many times again if you need to come up with 10 ideas I would spend like 10 12 bucks of full API cost with like my model my scraper my everything and save literally days of work and potentially this would be better ideas as well like you might do better you definitely capture stuff that a human would miss i actually had a look at how much a content strategist costs to hire as well and $81,000 a year is the average US salary $45 per hour globally on Upwork a junior person in the US $50 a senior person between a hundred and $250 per hour in the US in the US to do this agencies cost $2 to $10,000 a month probably around 5k average you know it depends on how much you're looking at here and that's just for the ideas right that's not script writing that's not making the video that's just to tell you what you should make a video of course there'll be a bit more detail around like and you know I'm sure that the reports are nice and and all that i mean it's not how to do that like if I make it on any I can have a beautiful picture of coming on the other side you know so it's all doable this is these are just like raw proofs of concept you know it's a very systematizable if that's a word business that you can build with this yeah there's a lot of side hustles too a lot of people like email me and like oh how do I make money with this stuff like honestly just take side gigs of these kind of things and then just automate it like that and sell it to people still pay pretty much full market price for these things and yeah it's like I mean 50 bucks an hour the problem is on upro track your hours right what do what do I do you either do hourly rate or you do like a one-off contract rate and I've noticed on Upwork recently more and more people want to do the one-off contract rate because they're not incentivized just to twiddle their thumbs and you know move jiggle their mouse once every 10 minutes to stay active they'll get it done faster basically and they can use AI to automate yeah so do that like negotiate one of prices for like the output and then just take these side gigs and you'll make more money than running content sites these days and arguably in a less painful more predictable way all right let's talk about the next one here which is going to be creating full YouTube scripts but before we do that let's have a quick word from our sponsor at Authority Hacker we've processed tens of millions of dollars in sales through Thrive Cart since 2018 and their new Pro Plus tier has us genuinely excited about what's possible for digital product sellers their latest update introduces multiple order bumps that stack complimentary offers at checkout which can increase your average order value by between 10 and 30% and their new tax inclusive pricing option shows customers the final price upfront helping to reduce car abandonment but what sets Thrive Kart apart are their practical innovations like checkout QR codes for inerson sales advanced UTM tracking for marketing attribution and even crypto payments options through USDC that are very exciting their platform is designed with conversion in mind offering customizable checkout templates and AB testing capabilities that let you optimize every element of your sales process everything integrates seamlessly with major teaching email and payment platforms and after evaluating all available options we've stuck with them for over seven years now because they simply deliver the best results so check out thrivecart.com today to get the best shopping cart experience for digital sellers thanks again to Thrive Cart for sponsoring this episode okay so how do we create an entire YouTube script using a process like this go okay so I mean I didn't do the full process here if I did it fully I would also insert perplexity in there but here I focused on the new stuff I haven't showed you yet otherwise boring you know I made an agent that watches YouTube videos to get inspiration okay um again if you want to do content like it's a good idea to look at like who's doing well what they're talking about and also to extract the gems out of the content there's always like one trick or one thing that you can pick up here and there prepare your script first of all that's how you make a script that stand out you make good content so I have this YouTube transcript extractor that I actually found on API as well again $5 per thousand transcripts it's fine you won't ruin yourself getting these is one transcript a whole video that's that seems really cheap okay and I've met an agent so here I have like a bit more of a very simple prompt that basically says "Hey you're an expert content researcher your job is to read YouTube video transcript based on the topic and create in-depth video outlines for the topic include all the best tips and tricks and teaching of the top videos organized logically." And I just instruction again i guard rail it right one pick the search term that you're going to search for on YouTube based on the query two search for the top 10 video three use the video's URL and scrape the subtitles for each video using the video transcript extractor tool that you have access to i actually put the the wrong numbers see I can't count because I changed the prompt so I should fix it but I think AI is smart enough to figure this out smart enough to realize what to do yeah that's what happens when you edit your prompts then analyze the subtitles then extract all the gems from the video and then organize the info and then put a full outline for the topic including all the interesting bits for the video is analyzed right again I could add perplexity onto that be like oh and search for extra data on Plexity add it to your outline etc you can put more stuff i could give it URLs tell it to read it it's unlimited you can combine these things how you want right and I put some constraint like make sure you use the scraping tools don't make facts up actually all the big AI companies in the system prom AI to not hallucinate so apparently it helps and always include URLs and tools when mentioned but in this case it's like for tutorial right and because it was pre-prompted against I didn't have to like prompt too much i said watch YouTube videos for NAN for beginners prepare mem outline use sequential thinking which is one of these MCPS that makes it think between steps basically the reasoning MCP I'm using and again it used this so it first search for these two queries which is N for beginners and tutorial for beginners probably there was some overlap here but that's fine it got the results so it got all the URLs and the videos basically so it got all these results then after that it used the transcript extractor so you can see put this first it tried 10 and it was too much it timed out so it tried five it got five and here these are the full transcripts so you can see you can read the transcript here you can see the transcript here it's all of it right and is that like you know YouTube generates the auto captions is it somehow pulling that file this is exactly what this is so it's kind of like a cheap way to transcribe the video it's not perfect but it's good enough for context for an AI basically yeah it has this five videos that it watch and then it starts reasoning on it and thinking about all this stuff so it's thinking about how it's going to organize things and then it gave me a full video outline and if you look at the outline I mean we use an A10 quite a bit first of all I understood that people want to build AI agents with an A10 so I like that i think if I just asked Chad to make me an outline for N10 it wouldn't do that and then after that the examples it gives are good you know like cloud versus self-hosted which is like a big thing how to get your API keys and access it gives the right examples of triggers they're not made up etc looking at the execution log it's real basically because it read all this stuff it's good enough and then you can see it actually has the link to the videos it used and then if I click on these links they're real videos right it's like they're not made up that's a 30-inute video that actually watch it's not hallucinated and then after that or once you have an outline I would ask you to use perplexity to enrich the outline if I did another one if I was implementing more tools but for the sake of speed here we're going to skip that part i generated a quick writing prompt on writing a script using this outline and yeah it has a pretty good script that is exactly talking about the stuff that we're talking about it tells you what to show and because it's based on all the real stuff it actually has like all the real comments and everything cuz the script writer can also read the transcripts because it's part of the chat history so it reads the transcripts of all the other videos and my outline and it makes my script for it and it looks good you know like oh look at these curly brackets click on test step which is exactly what you do actually when you do us if everything work you will see a success message appear on your Slack channel it's good actually so I think the kind of secret sauce here is not trying to do everything in one or is like breaking it down into stages which is exactly what you do when you're actually creating a YouTube video right you don't just start writing the script you make the outline and then it's infinitely easier to make the script and AI is essentially replicating that whole human process here which is crazy yeah and I imagine if you add perplexity on top you add perplexity deep research you can use that as well before it writes the script all the script writer is like "Oh I want to dig this deeper let me just research this and check that like read the NAT node documentation or something like this to to talk about this stuff." So that's the thing it's like now we're getting to a point where we can get these agents are replicating the human process of the content creation and that's why it's not anymore because it's not just using whatever training data the LLM has but it's just browsing the internet open stuff it's quote unquote watching videos and after that doing content which is how we create content too right yeah it's replicating the human process as I said and if you're someone who has tried in the past to create content with chat GPT without any of this fancy content text you know hey write me a script about such and such and it's been like ah it's mediocre no way this will never work like it has changed it's actually good now this isn't generic crap that you would just sell this is stuff that we would use to create a YouTube video that would get views let's look at the best practice start small and build incrementally you'd probably tell people to do that test before you put the workflow live document your workflow with clear names and descriptions use community templates and learning resources yeah but the committee templates are amazing in Nen and join the community that they have it's decent it has the real URLs as well you see it didn't hallucinate if I open this this is going to work right these are the docs of NM so it's all very very well anchored and that's because you put so much into the memory of the LM again we spent almost a million token building this outline because of all the stuff we've put in there but still we're looking at a total cost of $0.2 right plus scraper cost so 20 cents plus you know let's say plus a dollar for scraping it was less than that but let's say a $120 to create that whereas if you want to hire a full-time US script writer for YouTube it would cost you $75,000 a year that's crazy there's a mismatch here right and look I appreciate having someone high level to take the output of this and add that final finishing touches and the last 10% to it does make a big difference and I still think for a long time there will be huge value in that but if you're a junior script writer who isn't aiming much higher how long do you have left yeah not a lot you just need to have taste now you just need to kind of like steer it and you need to have taste and you need to learn how to build this because you can watch this podcast but if you're not actually learning the MCP stuff or whatever like this is not going to be useful to you you need to go and do that it's not hard it's like the most scary thing is like putting that line in the terminal after that AI can do the JSON file for you you can do all of that and that's like a test of skill right now the people who do not learn this will be left behind there's no way that you will make it in five years learn this so I think for any job where you're using a computer this right here is the single most important thing you can learn and do right now bar none and it's cheap like look it's like I spent like 10 bucks preparing this podcast in all API costs and everything and that's me wasting a lot of stuff and not being careful it's like how much does education cost in US again remind me $150,000 minimum plus whatever interest on your loan yeah it's crazy yeah so and we're also in this unique period right now where most people don't realize that this can be automated and so they're still paying full human rates for it and you have probably a few years until this really adapts like if you remember early chat GBT days when people started using AI to create written blog post content there were writers earning hundreds of thousands a year i mean people knew it was crap but they just wanted like SEO SEO content but they were making lots of money because they were still charging their full rate for it but they were producing it in like a few minutes it's the same here now with scripts so with agents in general build yourself agents that can actually use tools and it's like the rest of the world has not caught up and and you have an advantage here you can sell a script for about$1 to $2,000 for a good one i would say you probably need to polish it a bit like Yeah would for that price like you'd probably need some editing of what comes on the other side but you could still shave absolutely but this is 80% of the work is probably done by this yeah so we've got one more example and here we're actually diving back into our SEO roots and looking at an SEO analyst job to sort of analyze search console and some things here tell us what have you got for us here so this one was like a bit rough if I didn't run it multiple times but I have an MCP that I found here on GitHub a lot of these MCPS are running on GitHub you need to be a bit careful for this because they can be run by anyone and so your data transits through this service steal your API key with this or maybe so for example the API we use like this is run by API they already have your API key like there's nothing to steal they just want you to use their service and pay them for it these kind of GitHub ones be a bit more careful this one has good reviews and it's been reviewed by people from what I've Googled but be careful when you use a public MCP servers that you find on GitHub but kind of handy again you find this little piece here you get you ask Gemini to add it to your file and then you need to actually set up a project in Google Cloud Console it's a little bit annoying but there's some documentation that you can find here i would just follow that basically it will take you like 10 15 minutes to do it's free and you set that up and what that means is now my typing mine can use GSC right so actually like you see for example it has access it was like listing sites and it has access to a toy hacker and then what that allows me to do is to start asking questions like oh how is the traffic doing versus last week for example it's like oh for the week of April you got 9,300 clicks the week before you got 9,200 so you decreased by 136 clicks right but then you can go a little bit deeper so you can just chat with your data it's like oh how is it evolving overall it starts doing a bunch of queries and thinking and telling you overall the traffic is stable with minor fluctuations week to week without significant downward trend and then after that just of reading this I'm like this looks eerily similar to the type of client report you would send to an SEO client each month yeah and then I was like okay let's level it up a bit let's make this reason a bit on the data do a deep analysis of the last three months and give me a full report of what's going on so I started to think where should I search for and I did a search of like I want the whole site for these dates and then I want to aggregate this oh I'm going to run also this one with like a thousand rows there was only 12 last time for some reason again these calls are free so it's fine not a big deal but it basically just kind of thought and search for data and then they started telling me like your impressions are up but your clicks are down like your city again a overview is rolling out that's what's happening it just gives me a whole situation of what's happening and I'm like okay what are my top growing and shrinking pages and again it goes and gives me sorry I don't think it did it here actually yeah I was like oh it didn't do it so I was like I need to remind it of the tools because this wasn't pre-prompted so I was like used to control syncing and the tool now it did it it pulled the data and I had the data so I could see which pages are growing which pages are shrinking and then I was like "Okay great this page is growing." Analyze it identify opportunities and then use Brave Search and Fire crawl to see who is outranking me and what do they have on their page that I don't have that I should be adding on my page to start ranking basically base your recommendations on what you've observed on the top ranking pages so very much like what an SEO would do right analyze a page analyze the pages that rank above decide and so what it did is it again it thought about it it got data for the URL then it actually did the search like best travel affair program travel a programs and it got the results basically like the top five on Google i got some rate limit exceeded actually on some stuff then it opened the pages that outrank me with fire crawl to actually read the content and then it basically said oh you're ranking between two and six for these queries like best driver affilate programs here's the top competitors and then action list and it just tells me what people are doing again not perfect because testimonial carousel is for booking.com I think which is like they're promoting their aff program again it needs to be reviewed a bit you could tweak this yeah I would need to prompt it a bit better etc the point is it just went on all these pages identified what was on these pages that outrank me and started making me a list of things like top tools and network section that's okay how to monetize guys apparently like that's probably something that might work for this page a visual comparison table like apparently some people might have that and then it just like some schema markup etc in there but the point is like I can now have an AI model it's not perfectly prompted here because this was just like a quick test of the MCP but imagine if I ran like a proper reasoning model and I made it sync a little bit more etc and yeah if you show examples of what you're trying to of what the output is etc and because my prompts you see they were a bit dirty compared to the other ones um like the potential is here and again HFS has now an MCP as well so I could give access to HFS to that but so you could check which keywords the other pages rank for that I don't rank for and start making recommendations based on that and you can have essentially an agent that does a lot of the basic SEO work of basically optimizing of your pages especially after Google updates and so on so I'm even thinking for like outbound sales or sales calls like being able to generate insights about them and their competitors like it's super valuable like that's stuff that people would spend you know hours creating manual reports on so now you can connect HFS GSC Google Analytics and the it can also yeah it can also open the page basically and see what the content is and analyze all of that this might be a case where you actually don't want to use fire crawl because you might actually want to see the HTML if you're analyzing technical SEO stuff here as well yeah so there's different ways to and again it's probably a prompt chain i think like it's a bit too much for like one LLM call so I would probably break it down on like okay the first one collects the data the second one analyzes and the third one was the report for example so that it focuses more on each task but it's totally doable I think or we're very close to it at least yeah you could even then have it create like a you know one of these Loom video scripts that people send out when they're doing outreach with like a yeah hey gen AI avatar and pull all that together and like just fully automate your automate updating because you could plug that into an AI writing agent that also does the research does all of that etc and then you can potentially update the page depends on like how it's formatted and uploaded on your CMS but I think for like simple pages it might be doable actually we're quite close to like automating SEO at this point that's going to be one of my challenges soon i think Tim kind of challenged me on this at a hfs so we'll see but it's powerful all of this to say to fully automate SEO like press a button and do SEO you think that's possible the dream that we've had since SC Nuke you know so how much does that cost i mean to hire an SEO agency and I'm not talking about to work on a local business for like a normalsiz website you're talking $2 to $5,000 a month easily seo analysts now in the US earn on average $85,000 a year and on Upwork uh junior about $50 an hour senior $100 to $200 an hour now of course there's more to SEO than this you got content creating many other parts but you know this is going to save what 15 hours a week of an SEO manager an SEO analyst job by building this out properly i think it's going to be again I think it's nice for the overall site but I really like the URL by URL thing like I can imagine like taking the site map and then looping over each URL and having like three four five prompt chains and then essentially you just end up with like a big list of tasks for the SEO team or even like whatever cannot be done you make a task on notion and people just go and do it and they don't have to think too much they just verify and they either discard the task or actually do it it's going to change things a lot and for everyone who has a few years ago been forced to switch to Google Analytics and still has no idea how to use it well you don't need to you can just ask this and it will find all the data you need and explain it to you in simple terms so there you go the MCP for analytics is still a bit basic it still needs a little bit of progress but it's going fast so I would say 3 months from now it's going to be quite good actually and this is a very new technology right you know this is just integrated into typing mine like a week ago or something mcp was like end of last year it was released but nobody used it back then it was just entropic that kind of invented this but now it's like every big company is making an MCP server so like you can officially connect to them the same way we saw it for API and that that's it that's what AI is this year it's like if you're still prompting on charge you're already behind you need to learn how to do agents because that becomes as you saw a lot more useful and practical and not only that but like we're doing this within typing mine today but you can do this on NA10 and other tools that run in the background and you don't have to be the one prompting every time like this is a great place to like test stuff or quickly ask for questions etc but the point is that I said like imagine looping over thousands of URLs on a website and each running these prompts one by one feeding into a database and preparing the tasks that need to be done and or doing some of them that's where we're going with all of this because that technology doesn't need an interface it can just be done in the back end and that's the power so I guess I'm going to say if we get more than 10,000 views on this podcast which is unlikely I'll make TNA 10 version maybe otherwise so just share it basically that's the point and it's just crazy in my mind like how many opportunities there are for different processes different businesses that you can build out of this like right now and that's just doing the basic stuff that's not connecting it all together if you have to pick of all of the things you've talked about today if you were starting a business right now which one would you build for business I think the SEO stuff i'm sorry but many SEO people are a bit at tech and to be fair it's one of those weird subjects you have to be a bit good at data you have to be a bit good at uh tech technical stuff you know with the website but you also have to be good at marketing and brainstorming the creative stuff and not many people have all of those things and also we have credibility in SEO so probably I would pick that one because like to push back on that if you had no podcast reputation email list to get started with like it's very hard to break into SEO because there's so many it's the biggest market I think of all of this but there's so many people doing it that's quite challenging and I think because I like stuff that fits in nice boxes and it's easy to track and measure the email generation the email management for newsletters would be what I would choose because it's high value you can sell it 5 10k plus you can get results for them like sales which are can be measured generally when people do this stuff they keep doing it long term so and it's not hard I mean like again like I don't want to give all my tricks but I'm going to give one more okay just for those people who stay to the very end who stay till the end like honestly like and encourage you to stay to the end on featur but like all right for example you know there's a lot of platforms that summarize books right the one that you're using blinkist yeah yeah so like think about blinkist right they make this is this is blinkist is should be fully done by AI now right i mean there are like Headway also is another one that is now done does like AI voiceovers for the summaries and so on but the point is that there's lots of very good copyrightiting books for example in Blink that are summarized in Blinkist in like a 15 pager PDF right these are incredible training for your newsletter copyright etc you don't even need to be that good at that stuff you just need to find the right resource to fit to your model that's going to teach it how to write and so I find the book summarizing platforms to be excellent places to collect okay training data for these kind of agents mixed together with the tools etc and so that is going to make up for the knowledge gap you might have for the picking i.e newsletter writing for example for copyrightiting but also for SEO and stuff i thought you were going to have an idea to create a blinkist but that did YouTube videos podcast all media all media formats imagine if there was a Blinkist for the AI industry a blinkist for the SEO industry that just followed everyone subscribed to everyone and just pulled all the insights this week and you just logged in and there was one page what do I need to know that happened this week yeah I mean we could do that potentially there would be a lot of backslash because when you scale scraping then you face legal issues um but other than that yeah that could work that's the problem but then again is Google going to sue going to sue you after they scrub the entire internet to make their their exactly like who's going to sue you exactly that's not guilty themselves you know that's the question you need to ask but potentially yeah so that's AI agents for you i wanted to show some of the tech i wanted to show you some of the results and again lots of opportunities for site hustles here yeah it's going into the service industry but a lot of the actual service delivery can be automated or largely automated and you kind of put the finishing touches on it and that's very good money per hour basically all right guys let's try and get this podcast episode to 10,000 views on YouTube so version share it with someone who you think would benefit from this leave a comment to pump that engagement which of these business models do you think you would do do you have any other ideas for how you could chain some of this stuff with what you're doing make sure you subscribed as well so you don't miss the next episode we'll be back in two weeks see you then [Music]