I'm Planning AI Growth and Marketing AI Agents (Road to $100K/month)

hey there I'm Parker Rex and on this channel I go through a nope hey there I'm Parker Rex and I led product and technology for a startup that did well over $50 million a year until we exited after that I did a couple different startups where the tech was great but the idea wasn't so good and they were based in AI I spent pretty much every single day using AI tools since GPT 3.5 turbo came out and the SDK so as long as you could have been building on it I have been building on it and on this channel I go through a couple different pieces of information that can hopefully help you it's going to be coming from someone who's in the space someone who's built a lot of software someone who is using these services to generate real money so let's go through we have a couple different things here so we do Q&A we do AI news so things that I'm either just discovering and want to share with you of what's working and if there is some big update like a new model comes out then I'll talk a little bit about that then we go through the strategy of how I'm building the AI Services business to 100K a month and then I actually go through and build something so you can see how I do it it is raw and uncut unedited so let's jump in I don't think we have any questions but I just want to check so if you guys ever have questions you can just drop them below I went really in depth on one of them yesterday that was about open source strategy so if you're interested in open source uh technology and kind of the the way to do it then you should check that out because I posted about it yesterday uh looks like I got a comment comparison shopping optimize local grocery store shopping lists okay that was just a comment on a post that I made I don't think I have any unanswered questions unless I went into my school which will be paid soon I'm going to switch that this week because I realized by having 33 people in here and only three people that actually engag then it's really just 30 people otherwise you have a bunch of people that aren't engaging they're kind of just leeching so there we're actually a bunch of questions from this guy Aiden in here so I can check those out and try to answer one of them uh actually I answered in there already so I can just skip to the next part so on the new side there's two tools that I wanted to talk about so one is this tool called prompt methus so it's like Prometheus but for prompt and what I found is that a lot of tools exist for optimizing prompts but they don't provide that much value and I had tried tried this before but didn't get enough value out of it so I want to try it again and it seems that this would be the best way to have all of your models in one place and to be able to save them I know that there's a bunch of other ones out there but I've heard really good things about this and it just feels wellmade so the benefit of this would be that if you do have prompts and you want to save them and you want to test them against other models then you would do it in here verel has an AI playground this one may not be as good because it's really just you have the ability to add them but I don't see any easy way of getting into the nitty-gritty you have to open things I don't really like that as much but think use using these tools you can get the most out of the prompts so I will start to test out prompt meeus starting today and it's especially important because I have a workflow down for this content engine that I'm building for myself that I will likely offer for clients later but one of the big Parts in it is let's say you have a summarization node and you want to be able to get really good outputs from it so it sounds like you well doing the iterations in here over and over and over again it's just a little slower and it's a little tougher because you're going up and down and working within this node editor kind of a pain in the butt so I think having something like prompt meeus where you could keep the data set in one place so I could have a bunch of examples of in my case articles that I like I could have a bunch of transcripts of YouTube videos that I like if I was making a script writer but I could really see how this would be the next level of getting the most out of workflows for both businesses and yourself because then you'd just be flipping back and forth between make and PR meeus and when it comes to the promp you want to model them after stuff that works so always look at the top 1% of your field and try to mimic that so what I've done is I started to go through this deck that Gary Vayner Chu put together he just recently wrote a book and I was listening to him last night and say what you will about his cringe levels he's had a fantastic track record on pred the future so if he says something I typically listen I won't consume his content really at all until it's some big platform change or I think that he uh is saying something that is at that level you know the normal day-to-day go go grind stuff so what pointed out is the power of social but the way that it's changing is it's no longer a social graph it's going from a social graph to an interest graph and we'll touch on that in a little bit and how it ties into this content funnel update but protheus is GNA I believe play a big part in getting the most out of these because you can't sound like an AI going into the AI era you cannot sound like an AI has to sound like you so providing it lots of examples is going to be the way to do that and if you don't have examples if you're starting from scratch or if a client doesn't have examples well you need to write examples the way that you write those examples is you would take writing that you like create a data set out of that and then prompt on top of that to get the examples that you ultimately want that you can then train on so that is really important that's what I'm going to be be doing every day on this video the last section for building is going to be dedicated towards this content system it'll play into my daily growth and it will take me less time I'm going to spend more time building this so that I can get more leverage on my time and grow an audience quicker to deliver more value second I want to talk about is mcps if you haven't heard of them it's a model context protocol it came out in November from anthropic and it's a gateway or connector between large language models and other tools the really popular ones are in the coding Niche but we're discovering that it can communicate with other niches one being the let's see blender so blender is 3D modeling software and you can see that he's taking industry grade 3D modeling soft sofware and connecting it and then chatting with an external piece of software I think that this really communicates power of mCP because it is totally separate from the coding world it's showing you how strong a connection can be between two tools and if I just want to kind of break this down let's take it let's take this code let's just uh I'm going to start keeping a collection of these in a folder it's on like make space mCP I'm going to CDN to mCP I'm going to do a get clone on that code base I'm G open up the code code base and then let's take a look just so we can understand together how this works because I haven't looked at it this theme is called Midnight Tokyo by the way if you're curious but let's take a look this is just your basic init for python app it looks like you can import this mCP server fast mCP this might be similar to fast API but for mcps we're using websockets to keep the connection open persistent using asynchronous functions we're using some different types we are which is nice because needs to if you're unfamiliar a type is essentially a definition for data it's going to be logging information out with info so the time the level name this is really important when you're doing any sort of logging is what level of logging it's going to have so you're classifying it it might be a warning it might be info but having all that logged and categorized and time stamped it's just a way to do it especially in the AI era we have a big class we have a function for connecting and making a connection it looks like blender has an add-on for a socket server so it keeps it open so that's how you get that real time connectivity that's connecting a disconnecting we have one for receiving the full response so it looks like we send something over socket to blender this is saying this is how we're going to receive this and then we pass it a socket in the buffer size so the size of information coming across if it's too large we chunk it we get an empty chunk connection might be closed if there's no chunks if you haven't received anything yet this is an error okay so it's logging out it's covering our error if okay so we're joining the chunks back together and the reason you chunk it is because if you have too much information going across the wire at once LMS won't handle that well at least in the context of writing so I'm assuming it's the same probably a lot of information going back and forth because if you look at blender there's significant amount of info that needs to get into there get to here we've either timed out or broke the loop so it's just looping back and forth can send commands across the wire has timeouts in place else has a server lifespan coverage it creates the mccp server with life this is a really nice file and typically this is a best practice you won't actually read the code yourself you can go and ask the AI for a summary of course but for something as small as this project I find it more beneficial to actually understand it when you have a a little bit of depth on the project then making additions changes feature additions whatever and what while using in AI will be a lot more effective uh getting the scene information so all of these are tied to the CP tool that we've started so everything at the top is server creation and error handling and then these are actually the tools you can pass all these different pams really cool wow so yeah this is a fantastic example of what's possible with mCP if you could then use this as an example where I wanted to create an mCP for another tool what I would do is I would take this I would take the documentation that is associated with this blender file and I would use that as an example for creating future mcps then you have an mCP agent it's just the same thing that we spoke about with creating the content one you want examples you want best practices so because this one has 6,400 stars and it had a fantastic demo demos are tricky you can fake stuff but I would feel comfortable using this as a base to work on let's talk about the content funnel update and some strategy on that so if you're unfamiliar a few videos ago I was ideating on how could I get all of the content that I'm creating on this channel and on my main Channel turned into every platform so all of these those are dated but you get it and it's following a similar format where it's a bit of an update on the Gary ve thing but it's the long form to short form across all platforms Iman gaji bit of a guru type but very successful $100 million speaks for itself so follow what he's doing he's a top one percenter and Luke Belmar as well are doing a similar strategy where they have their pillar piece of content they transcribe it I'm assuming this is the workflow I know they have hundreds of employees I can't afford that maybe later so I'm trying to figure out how do I leverage AI using this unfair Advantage I have which is being very technical and get a similar effect so from a high level you record the long form you have an AI transcription that's created that transcription is stored somewhere alongside the video edit it is then sent to a thirdparty service that makes the audio sound better if this audio sounds good it's because of this third party service that's watching for it that third party service does two things one it processes it and creates a couple different outputs so it creates an mp4 that sounds better it creates a captions and subtitles and markers and summary files all separate files so those would all be stored somewhere then it also posted to YouTube then the idea would be that this is being stored in a project management software or even a Google sheet where we can keep the status of where it's at in the funnel or in the pipeline so you'd be watching for some sort of status maybe that it's been processed you would add that information with links to where the things are in that template you'd have all the fonts the colors the creative brief what to do what not to do this part likely would be down the line this would be for more so an editor so I put green check marks if it's fully automated and then I'll put Blue if it's human in the loop the human in the loop would then go in and do this now I need to think about I haven't found a third party that can do short form clipping that has an API so I need to figure that out but at the end then I get an email to review it before it goes and then I'd update it to ready to post and then I'd manually post it then we go from video to text so we've covered short form and the creation of that this would be for text and images and voice so that covers X LinkedIn Facebook blog post potential audio snippet at the top of the blog post so then you'd use the same watch on the G Drive filter it by the extension you run a chat module which would cover the creation of said blog tutorial you'd store that update the status have it be ready to post you could actually do the post of the superbase module and then update the status back into either clickup or Google sheet then the chat would be watching for blog posted so after the long form is there then you'd create the LinkedIn and Facebook from there you actually need images for them so it kind of ends with a kick off to flux this would probably be an HTP request with a really good prompt that you can find on my website par x.com for some sort of perplexity like image and then you combine the image with the text to put it into canva for bulk create so canva has a bulk create feature where you can put in a spreadsheet so image image text and then you would store that back with all the images you take the transcript or sorry the video and audio file and have a prompt which I think was before but you take some sort of summary and then feed that into 11 labs and you end up with a blog post that has the actual text the actual video that's posted that's referencing the daily upload or whatever it was in that video and then you have a summary in audio format using 11 Labs that's trained on my voice so that's the whole system and we're picking off in pieces of time so what I've done today is I'm trying to figure out just how do you go from A to B how do you go from a blog post or a piece of news that I ultimately will talk about to a nice looking AI generated Instagram host so in here what I have is it's uh G form to fire call to Instagram text and Caption This is actually RSS because I figured originally I would take the links that are in this video I'm talking about right now and I would post them in a form and then that would kick it off however if I can automate that first part where I don't have to go out and find the news then that's even better one part that will be tricky is getting the RSS feed to show the things that would matter to me so I think I'd still need this human in the loop where I could get PED in the morning with this so that's probably what I'd want to do is change it so that I can still be the the stop gap between what's what is good and what is not so potentially or add a module for messaging me the latest news this could be an email it could be whatever but I think that that would be better so regardless right now as it stands if I type in five items and I run it what it's going to do is it's going to this bundle of RSS feeds which is going through three separate feeds which are keywords chat GPT prompting and AI news and it's taking five of those and it's returning the name of the article the description the image and the URL next we use a third party service that's called fir crawl and what fir crawl does is as you can see it's kicking off these scrapes is it goes and scrapes the page but it does it in a format that is very llm friendly so you'll get down to what you actually want if I can find the example yeah they crawl they gather clean data they handle Dynamic content they have Smart weight built in so that you're not going to hit any sort of API walls can parse it does all this stuff that you just don't want to have to worry about and then that way the output that it brings back is going to be optimized for llm readability and so we did I think five and after that we did an iterator an iterator lets you iterate or go through the list of the items that are return returned so for this case RSS had five and then iterator is saying after we've gotten this from the previous step go do the following steps individually so if I didn't have this then it would just look at the first one or maybe it would look at all five and try and do a summary that's not what we want we have five separate outputs that are being done by this article summarizer so you can see that cool we sent five separate fire calls because we iterated through that list and then now I need to do an additional iterator which I don't know if I did yeah it is mapping through the markdown right so it's doing five separate iterations and then passing it to this article summarizer so I'll get five different article summaries and capture texts and then after that it will store five separate rows on our database so if look at this and I scroll up boom we've got five different pieces of information it looks like we got some repetition yeah so what it's doing here is it's actually taking I had just changed the prompt and this is good it's it's real I'm not going to edit this I had changed the prompt a little bit to give it examples and it looks like I may have made a mistake on selecting user instead of assistant so let's change that real quick we'll go into the article summarizer this will load this up you can see why prompt methus would be more helpful for something like this because I could really handcraft exactly what I want in a platform that's optimized for it so what I'm actually going to do is I'm going to switch text content I'm going to do it to developer system and then I'm going to put in the examples I'm going to put in the examples as a developer SLS system then I'm going to put in the examples as a developer system cool let's hit save and now let's go back and we're going to delete this stuff i' originally put this in here but it was overwriting it so forget it we're going to delete this table as well and let's put in summary slash caption textol and let's make this look a little bit decent great now when we run it it should give us what we want let's do two so it's quicker iterator is going to there you can see they're color coded that's really helpful I'm not going to open this up because it has an API key in it let's do it it looks like we got two bundles back this year is the year that AI gets better than humans says Kevin well from open AI he said that yesterday cool and for examples what I did was I jumped on to Instagram and I just typed in the keywords that I want to cover then I looked at what was working and I just posted that as the examples that's something that you should do and something that I've recently learned is if organic posts are doing really well remix those and put a Twist on them don't reinvent the wheel the Market's telling you exactly what works so now let's go back seems to still want to do this first one so wondering why the caption is feel like we have something going on with our prompt but the second one geek Innovations unveils a new Gadget designed to simplify daily tasks okay it's just not that good of a an output so let's go back into here and let's make sure that the examples are not used in the actual output so I'll come in here you're responsible for this yada yada yada we pass the total number of bundles which I believe no we actually want value yeah so let's here's the Articles and then value okay that should fix that and then use a company name people's names when possible don't use emojis don't use exclamation marks be professional UNC concise examples of image text let's just dump all this in here oh3 mini high I can handle it well I'll delete those example image text to use or inspiration not actual output just to be really clear and then we'll go down yeah that should do it okay now let's try one we're going to go back we're going to delete this kind of doing the build and the strategy at the same time so this should look for a header so we need to change that too go back in here Google Sheets contains header yes wonder why it's adding it below te headers okay now we need to put in one and let's run it it's calling fir crawl won't really need iterator it's just one what I think will happen is I'll have a bunch of these individual scenarios so we're starting from the top right and then what I can do is just reference the same sheet in a different scenario I don't think I think that'd be more manageable okay so it found chat GPT as a therapist balancing right benefits with clear limits and real risks Chach is being looked at as a tool that might help with therapy some say that it can offer friendly advice when people need to talk I'm curious how this would work I still would want to make it a little bit more click baity I feel like there's a balance where it needs to be obviously engaging but it can't be over the top so let's do a little bit more tweaking in here okay is the image text no more than 20 words make it like batty and engaging you're uh extract the most interesting pieces of information this is super important to get right and must capture the reader attention get them to stop scrolling see how that does then while that's going what we can do is let's let's grab the output ands take that delete these we're going to post that into an image generator and just see what it gives us uh let's go back here let's run it let's go to gamma I discovered the AI image gen stuff by purchasing a tool that's totally unrelated but it taught me what to do so that's cool so what I'll do is I'm going to use this I'm going to grab our prompt or perplexity style image go to the AI tools resources scroll down to top 1% this up and let's grab this prompt then we'll go back into well I just covered the previous clipboard but it's fine let's go into here going to paste this and then under user input we're going to take whatever the output is so oh it got the same one interesting I guess there were no new articles in the last five seconds or minute so let's post this in here and I'm going to do this and let's just see what happens I'm curious no style let generate based on the image text so probably need to separate those I mean that is kind of cool looking probably need to modify it but if I took this let's take this image and just see what an example post would look like and again I'm doing this stuff manually before I go and automate that's something that I recommend so I had scaffolded some stuff yesterday what a cool post could look like so let's just take the image we got and let's paste this in here make it white now we need a different background so oops let's grab the background hide that for now and we need that image I'm gonna have a similar style for all of them I don't know if this is the style in fact I don't yeah I don't know if that's that's speaking to me but this is where just tweaking and tweaking and tweaking comes in so also don't think that this would be detrimental towards a page that's G to have no following at the beginning so if I just looked at it like okay what would this look like if I did post then I would see sample I don't know I think the next step I would do is I would put this on a phone so I'd get the figma mirror app open on my iPhone and I just look at it but yeah next steps for this would basically be adding some sort of suffix here at the end it be like go subscribe check out my email newsletter or whatever some sort of call to action and then just working on this prompt a bit but that's definitely a good V1 and I think that it would make a lot of sense I said I was going to finish this MVP let's just try to see what Cam's capable of before we jump I don't know how many minutes I'm at I'm gonna cap these at 50 minutes we're okay we're at 35 so how does the C oops I don't want to run that stop it oh I didn't know space bar does that interesting let's see how canva Works in here so let's grabb a canva and can we do bulk create create a new design dat a new design dat a folder create a comment make an API call the image upload the image canva flux flux ified turns it okay cool so we could use this app to use flux so let's use it in a new design it's going to be for Instagram posts I don't know which I want the tall ones so Instagram post Instagram ad Instagram real Instagram post Square okay so is this one yeah so can we use flux to go into this flux Chanel flux Pro I used to buy image credits they probably have yeah so flux ify is doing a stu that we just need to do ourselves so Aug blend Labs is a company that realize oh my gosh everyone's going to want flux inside of canva but we don't want to do that it's going to cut into the or it's going to increase the cost so what I'd likely need to do in this case is let's just sketch this out let's pull up the existing flow and we need to use flux because it's it just looks good right the whole point of this is we're not going to put something out there that degrades the brand from the the start like it needs to look good because then it'll work work so if I look at this we have RSS goes to iterator iterator goes to fir crawl fir crawl goes to iterator iterator goes to chat GPT we might find that we want to use anthropic so I'm going to put um red for changes and thropic they're better at writing people will say they're not but they're wrong after chat gbt we'll go to G sheets well so a couple things I know off the top of my head are that we're going to need separate image Tex we need to basically pull out the information and separate it because as it stands today we have one that's going and it just lands in here and it's messy what we do want is to actually have image text in here we want to obviously have it like be separated so let's grab that cool so we want that that to happen we have a header the header is some reason being in pain well so I know that we want that I know that we want to [Applause] else I want to make this prompt better so we need we're GNA get the information back we need to iterate through it I think the best way to do this would be we'd have two anthropics some sort of router I'm guessing is what's it's going to be called and then this would go off for image text then this go off for image option which I almost want to just have it be the summary because the summary as an output will be used for blog article later so each one of these outputs is your asset now that means that we need to have one that makes the kind of click baity um actual caption but then both of those will go back into the G sheet and we'd have ideally we'd have Let's see we have the source because we want to use that later in the blog so we can reference it we can basically say like this is our take on the AI news we're not just copy paing it this is the remixed version of it in our tone so we need the URL and it also helps with the domain authority of the website because if it's linking out to other places then you're going to be become a more reputable Authority in the eyes of every SE engine not just Google so you need the URL you need the image text you need the summary need what else could be interesting is we need to write an image prompt at some point but yeah so that would just store it so then after we've stored it what do we need to do we've on through we're at the start we need to also tweak the RSS filtering maybe it's really like we're classifying the stuff that we want to do and talk about before it even hits it an iterator so what may happen here is we could have a chat GPT CPT so instead of this tweaking of it like that's what we want but I don't think we're going to do good by just doing keywords so I think if we return all of them and then if we had another if we had a chat GPT instance where it could classify the style of story like the topic where if you type in chat BT and you get all these results I basically want to have like a blacklist think you can do that in RSS actually but celebrities I don't care if oh what's you know the one of them that came back was like what's demy levado think about AI like I don't care at all about that uh I don't want to have celebrity stuff I i' really could just start there actually on the RSS and then I'd still want to go to this chat GPT because then I can say you know I'm going to type this so that it's going to fit in the space but you are an expert I want to say like staff editor and chief something you are responsible for highlight uh selecting which articles get published a popular news publication you will receive a list of Articles and you will analyze them for the you'll Rel receive a list of 15 articles and you will analyze them for following or on analyze and select the top five based on impact to global economy impact to the job Force popularity I'm giving it a ranker saying you're going to get call it 15 of these articles sift them down to the best five so if this is running every half hour so 30 every 30 minutes you get 15 stories then every 30 minutes you're G to sift that down to the best call it even like two maybe three say the best three and then this Chachi p is going to to make sure that our inputs are better there's a common ISM called garbage in garbage out that's what we're doing we're making sure that at the source we're getting the best articles that's going to be done by blacklisting and then having this sort of like classifier so then after chbt does that then we'll go back into the iterator it'll sift through the three it'll do the fire crawl there really no modification on there just turns it into llm friendly markdown and right through that again we're going to do a router I think is what it would be because then it can maybe it's a iterator and a router I think so and then this would be filtered on maybe it's like data dot image text some sort of filter right so we'd have a filter and if it's the image text then it goes there if it's the summary or no I need some sort of information what do we get back from fir craw need it. markdown let's run this again hit save there then we're going to actually change real quick we're going to change this to filter we're going to have a black list add Blacklist keywords yeah so that's the thing I can't just put celebrity in here so yeah so will need the classifier and then I will need to rotate my AP key so that you guys don't take my fire craw account so forget this Blacklist thing we will just do that on an API call to chbt so can modify this articles to avoid since uh any that speaks about celebrities cool and it'll go on we don't need this anymore so then we'll get the three stories three best stories every 30 minutes so what uh 48 * 3 we get 144 posts a day whoa that's crazy 144 posts a day it's 52,000 pieces of content oh my gosh that's a lot well if it works it works right we can get the outputs and see could be a pretty expensive thing to run but we'll test it iterate we get a good image text a good summary now I need to figure out how to use the flux generator can't use gamma see flux how to generate images using flux over API automatically do this with this API call hey everyone my name is Janis and in today's video I'm going to show you how you can generate we're GNA make a request so he's using F so we are on the back end now ai so we're gonna we're going to use to generate the images using flux AI image generator so we need first if you don't have an account with fall AI so you need to click sign up and you can do it with your GitHub account where do we get the URL head back to make and as you you can okay generative media platform for developers would this be the best one or would it be replicate because this replicate probably be better yeah so I'd use either replicate or FAL so let's see need to figure out the ca of those replicate probably do this one or FAL and you take before that we need to give it a good prompt right so we know that this replicate thing will be for the image gen oh before we get there we need to have a really good prompt so then move this over how do we get the prompt well of course we're going to use meta prompt so after the G sheet we will hit I have a an assistant that does this so instead of a chat call this will be open AI assistant and what I'll do is I'll call this is for the meta prompt for image yeah cool what that does is it's going to say hey we have all this stuff and can you give us a really badass perplexity style prompt that we can give to an image generator so then we'd have the prompt we probably want to store The Prompt so we don't have to run it a million times and week so then I'd Store The Prompt back to G sheet so we're back into G sheets I just like to always keep an eye on what's going on it makes a lot easier to build so in G sheets we're going to add the image prompt The Meta prompt yeah okay we'd get the meta prompt image prompt after image prompt we'd see it and then we do HTTP call so scooch this down yep it's going to use one of those but the actual module in here will be HTTP yep I like this a lot cool then finally we need to store it and we're run up on 50 minutes so I got to run because I have one hour to do these every day and it takes me about 10 minutes at it so upload um finally we need to store that image where will we store the image we can actually just stick it into the actual G sheet itself as an image in cell um yeah so store image on Cell cool that seems like a good plan we're going to continue building it tomorrow if you enjoyed this video make sure you like the video that's the way that you pay me check out the school yeah it's it's actually not free anymore so check out the school anyways and like the video subscribe check out the main Channel as well do all that awesome YouTube stuff and I'll see you tomorrow see you and we're doing a thumbnail this video is going to be called planning a Content system with AI that doesn't suck is it possible that's pretty engaging right in a smile what the all right see you