# $570K/Year in a Month: One Guy + Claude Runs a 30-Client Agency

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=1I7-OvFQxBk
Duration: 25 min
Language: English
Words: 4745
Transcript page: https://viewrankai.com/tools/youtube-transcript/1I7-OvFQxBk

---

[0:00] This guy reached $570,000 in annual recurring revenue within a month of launching his idea. You see, he realized that he had a system for getting his site to rank in ChachiPT, Google, and other discovery tools. So, he offered it as a paid service. He signed up dozens of clients and then he let Claude do most of the work. This is the kind of business that can really take off in an AIdriven world. So, I asked my friend Josh Pigford to talk

[0:26] about how it's done. Let's get into it. presented by Zapier, the AI automation company. Josh, this is the tweet that made me just reach out to you on tech so many times to have you come on here. You are up to, is this still accurate? $474,000 in annual recurring revenue. No, it's about 100,000 short now.

[0:46] What do you mean? Where is it? I mean, it's at like 570,000 [laughter] something now. And this is an idea that wasn't public a month ago, correct? So, what was the This is the first tweet. This was on a Friday at almost 5:00 PM on September 25th was what like kicked this off. Um, so it hasn't even been a month yet. And summarize this message. What was the

[1:10] offer that you made in that first tweet? Yeah. So, it's um I've been doing well, I mean, you know, I've been a entrepreneur for two decades and so obviously I've done a lot of SEO stuff with different businesses over the years. So, I had like uh sort of turned all of this SEO uh plus like AI mention tracking stuff um into this whole system that I've been using on my own sites for most of the year. And so I said like, hey, let me I would love to do this for some other companies and just like see

[1:37] if there's anything here. Um so that was a very much like I'm like wrapping up at work for the day about to go downstairs and like eat dinner with the family and so I just like threw this out and said like thousand bucks a month. uh you know this all happens in Slack uh shoot me a DM if you're interested and um and the offer is that when somebody is searching for them on open AI's different chat apps on claude etc that you will help them be the answer that those AI engines offer up right correct that's the that's the basic like value proposition how can you do that what's the approach

[2:17] that you take because this is also new and mysterious curious that nobody has an easy answer, do they? Sure. No. No. Like anybody who's promising rankings is um full of it, but um all this stuff is rooted in sort of good SEO, which is, you know, been around for since the beginning of the internet. Um and so at at the foundation, it's it's just good SEO practices. And then we're tracking all of both SEO and uh what would be called like AEO or GEO um and making changes that address the ways that these different LLMs uh approach finding information and gearing it towards them uh and then tracking what works, what

[3:02] doesn't, and then adjusting. So this isn't like a oneanddone thing. This is like every single day we're making tweaks to to your content and deciding what's um sort of most important to do in a given day. When you say we, how many people are on the team doing this? Me and a and a bunch of uh AI uh pots. What's the AI that you're using to run

[3:27] it? What's the overall approach? Yeah. So, it's clawed at at the foundation. Like that's that's um for for pretty much everything. There's a little bit of like um GPT sprinkled in um around some like image generation stuff when that's required, but that's that's even that's pretty rare. So, it's it's basically clawed opus uh at pretty much every step. Um and okay, how do you keep them separated all these different clients? Is it different

[3:58] clawed chats? Is it something else? Yeah. So it's they are all from a technical perspective very siloed. So there's not this bleed between clients. There's uh they've each got their own um so we operate through GitHub repos. Um so both their own and then we also have our own versions that's uh or sort of instances that we're tracking all this stuff. So that again like there's no bleed between um files or anything like that. And then there's different sessions with with Claude to to do each task. Uh so there's no like context

[4:31] bleed or anything like that. What do you keep in a repo in GitHub for each client to make sure that you're organized with them? Yeah. So that's a lot of its data is the you know it's both um what would be like a work log like here are the things that we did here's all of the research that we've done. So when you produce an article, we might go find a hundred different pieces of content to use as reference for um for producing that content. So we're like we're saving all of that. Uh it's a Google search console for different numbers. It's us doing actual calls to chat GPT um to track uh

[5:07] you know different ways that people could be looking for what you're solving. not just looking for your company, but looking for, you know, the the specific thing that your company does and see if you come up for that and then tracking if you get mentioned or not, like all that kind of stuff. There's just it's a massive amount of data per client that we're tracking all stored in there. And what's your

[5:28] approach? Is it to get them on Reddit? Is it to create YouTube videos for them? Is it just on-site content? And then we'll talk about the tools that you use to keep track of it. Yeah, so it's um it's kind of a mix of everything. We don't do well. So, we do some outreach via email, like, you know, sort of what would be considered um link building, not just like asking for links, but um a lot of it's like correcting information that somebody might have posted about your company so that uh LLMs are um repeating correct data. Um but most of it is on-site um optimization kind of stuff. The the the

[6:06] kind of Reddit stuff I think is mostly shady, scammy. it's going to get you. I think it's a short-term um play that's going to burn you in the long run. Um so we're we're focused on sort of these like the long sort of slow moving bits that uh sort of pay off in the long run. Interesting. I I've got to say I've actually talked to multiple companies that are very big on saying that Reddit and YouTube is a big source for content for LLMs and they're active on both platforms and they're not doing anything shady. They're using their own accounts, but they are filling in answers. They're

[6:41] asking questions that lead to the kinds of answers that they want. But you're not I think that I think that most people are uh giving a very polished view of what they're doing. I don't think um I I think most of the people who are doing that at a scale level like you know companies that say that they do Reddit marketing for you uh is uh shady as hell. So, I have I have uh I have a very

[7:08] low um amount of respect. What are they doing that's so shady? What's the bad thing that they're doing? They're using um basically like I it's it's the epitome to me of um like the dead internet theory. So, like maybe they're not automating responses, but they're using AI to fake community. And I think I think that's

[7:30] I see. Okay. I think that's that's lowbrow. Okay. So, how do you know what content to write for each client? Yeah, so this is there's a an onboarding process where we both ingest all the stuff that we can find from your own site, from your own help docs. Uh we also ask uh you know a ton of questions just to learn more about both where you're at um where you're where you've been and then sort of where you want to go with the company and like who your competitors are, features, all sorts of stuff. Uh, and then we use that, you know, the tone that you want to take

[8:05] when you're talking to potential customers. Uh, and all of that gets rolled up into, you know, essentially these like research documents, um, that we use for reference when producing content. You know, in the old days, people would have SOPs, uh, what is it? Standard operating procedures for stuff like this. Today, it's skills. Is that what you have? A skill that then guides you to asking questions, a skill that then guides you to creating the content. Is that is that right? Yeah, that's uh so there's there's certainly a lot of skills. There's also a lot of um I mean really it's Python ultimately. I mean there's a ton of code that happens in

[8:40] the background to make this stuff repeatable. Um you know the the pros and cons of an LLM is it's it's sort of non-deterministic meaning like you know you run it 10 times you're going to get 10 different outcomes basically. So we use a lot of code to help make that a little bit more um reliable or repeatable. Um but a lot of that is is sort of these very specific skills for

[9:06] for doing different tasks. Yeah. Like what onboarding I'm assuming is a skill with a set of questions that you need answered. Mhm. So, and there's a lot of this stuff is also uh systems like operational stuff which I I've kind of find fascinating about all this is like okay well if I've got a few dozen clients I'm I'm personally having to repeat a lot of this stuff. Uh so I've got skills for like setting up new clients where it's it you know creates all the notion documents and fills all of them out with the the things that I

[9:38] need from them. um reaching out to them when we've got some new piece of content that we need them to review. It's automating the like, okay, we're going to produce this piece of content, but I need some clarifications on like the way that your app works. So, it'll automatically post something in Slack that for them to respond to and then ingesting all of their responses so that I can keep track of all the, you know, hundred different places that they could respond in Slack or in notion and I have all that in one

[10:05] place so I can manage all of that stuff. So there's there's a lot of just like operational stuff that happens there and all those triggers are happening on what in the terminal that or claude code desktop app or something. How are you keeping track of it all? Yeah. So it's the the sort of main tools that I'm using here are Slack, notion, and then through clawed code via via terminal. Um there's a few things that happen with APIs to make Slack and Notion talk to each other. Um, and that's like these tiny little apps that I have on Cloudflare, uh, hosted there. Um, but for the most

[10:42] part it correct. Uh, and then, you know, at the end of the day, it's all running on like a a Mac mini here in my office. So, wait, the trigger that says we need a new article for this client is a notion doc. No. Well, uh, so the trigger is multiple times a day. I have basically uh a Python script that says kick off all of

[11:06] these different tasks for each client. What what happens though is it's different for every client based on what our system has decided is most impactful. So that varies that could that could be uh you know an outreach email to fix something that somebody wrote about you. Uh it could be a mini tool that we're going we're going to build for you on your site. It could be

[11:29] changing a headline within one article. It could be writing a new article. Like it could be one of dozens of things, but it decides based on data from Google Search Console, etc. And it it decides what's likely most impactful and then does it. So you've got an automation that checks Google Search Console for each client that then based on your set of on your approach says what you should do today for each of these clients. You then trigger a message to them on Slack based on that if you need an article, if you need something or you just write the article

[12:05] yourself and publish to their site. Correct. I see. Okay. This is an interesting way of doing it. Now, you and I talked last about how you were setting Hermes agent up and trying to figure out what LLM to add to it. And I'm surprised that you wouldn't use Hermes agent or Grockbot or Hyper agent or any one of these tools to say each one of my clients has an agent and the agent is managing them according to our process. Why do you prefer to do

[12:30] it this way? Uh I've essentially built my own version of those. So, you know, it's it's sort of um my problem I guess with like Hermes or uh you know, Open Claw or whatever is there's a lot of bloat that I just don't need. And so I, you know, at the end of the day, I would be fixing problems with those versus just building what I need. So I've just built this

[12:55] system to run this specific business. Um, and and so I can be I can move a lot faster than having to kind of like work within the confines of someone else's system. And when you started it off, how did what what existed before you built your ver your version of Hermes agent? Uh, nothing. I mean this was it did start as like a single SEO skill that I like that was that was at the root of it. Um and then I just sort of it just kept expanding um past that to get a

[13:22] lot. So in the beginning it was a skill handling all your your approaches for yourself. Great. Then you added another client still using the skill. Then you started to find little things that needed to happen and you started building on top of it. And then at some point it sounds like the uh lack of predictability of of of an AI model made you decide you needed to start writing code. What are some of the things that

[13:44] you said I cannot leave this to chance? I know that AI is a probabilistic machine and it's good but probably it's going to make mistakes. What are some of the things that you turn to code because of that? Yeah. So there's a lot of um checking for the validity of statements made. So making sure like we've you know for every client we've essentially got a like a truth document. So here are facts about uh the company and what they offer. So using code to make sure that numbers stay accurate or correct across you know a thousand articles. Um it's making sure that um we don't use too

[14:23] many links in a in a given article. So counting the number of links um it's it's bug fixing because I you know we're actually coding here on like if we build this like a calculator for something for some client uh we need to actually run code on the machine to to basically run tests for it and build all the code make sure it works. So um there's a lot of moving

[14:46] pieces there. Can I see the site itself? It's pretty bare bones. When I was looking at it with my agent my agent said there's not much here. Even the to get started took took the agent to your email. Yeah, it just sends me a it's a little form here that just sends me an email. Yeah,

[15:04] that's it. That's it. By the way, on I think Uhhuh. Go ahead. Uh I I think like long term um it would make sense for me to like build this thing out to to have this sort of major um you know SEO AI presence. Uh, at the moment I'm too busy making money though, so I [laughter] don't I don't necessarily need to to do a really

[15:30] elaborate site at the moment. You know, the other tab that you have open is one of the tools that you use. Can you click over? Tell me how you use it. What is this? Yeah. Um, so data for SEO, this is anybody who's like dabbled in SEO, uh, certainly from an programmatic perspective is familiar with them. Um, it it looks very dated. It probably has been the same design for who knows how long, but it has a massive that it's exactly what it is, right? So, they're they're rolling in it. So, um, it's basically an alternative to something, you know, there's lots of different SEO tools that people pay hundreds or

[16:08] thousands of dollars for to get data access to Google rankings and things like that. And this just gives you very inexpensive uh access to to those tools. How much is it cons like roughly for all the client? How many clients do you have and how much are you paying for this data? So, I think we're 30 clients right now, and I spend maybe $100 a week on all the data versus if you're using Hrefs or

[16:38] I would be spending many thousands. Wow. Yeah. This is incredible. I mean, really, within a month, you're at over half a million dollars in annual recurring revenue. You could potentially end up with a multi-million dollar business here that's run by just you and Claude. and a set of skills and tools that you're building yourself.

[16:59] That's the hope. What's the big takeaway for this for somebody who's listening into this? So, I I think that I've um where I've found myself going in sort of the entrepreneurial side of things, you know, as someone who's just um builds too many things and and like, you know, I've got a hundred random projects going at any given time. um in a in in a world where like anybody can build software quickly now um using AI for these sort of to to make humans more efficient versus replacing um like I think what's valuable about first index and what I'm offering is like it's a thing that people want to have done

[17:47] for them and they don't want to think about it. So it's this like done for you kind of agency services kind of component. So focusing on automating that kind of stuff like stuff that still needs a human in the loop um and makes me more efficient and where otherwise it would have taken a team of 10 to do this uh is sort of some lowhanging fruit at the moment. Um, so I also do this with uh another business I have called Knockoff, which is finding, you know, people who are copycat, kind of stealing people's images, uh, e-commerce sites, um, using their brand to try to fish people, like that kind of

[18:30] stuff. There are these take down processes that you can do across Amazon, Walmart, whatever. Uh, but they're very sort of time inensive to to to do this. But I can use AI to heavily automate the process um for companies. So then companies pay 800 bucks a month for that. That's the model behind knockoff. Knockoff is this browser plugin that lets you know if you're on Amazon um if what you're looking at is a junky knockoff that's not going to work within a few weeks versus the original. Um and I know you were looking for a business model once you created this tool. That's the model you came up with. The question

[19:07] I have on that is can't somebody just say, you know, I I see that there's a process here. I'm going to give it to Claude or give it to one of these agent tools and tell it go figure this out for me. You would think that, but no. So, most of these have um there's a lot of legal documents involved. Um there's lots of most of these tools have their um diabolically opposed to uh uh automated processes like you know an automated browser going in and trying to fill this stuff out for you like you don't get blocked instantly. Um and you're making legal statements. Uh so

[19:48] having AI basically make a legal declaration on your behalf um is pretty risky. Um, and it's also ripe for missing stuff. So, or doing it incorrectly. How if if it's a oneman operation, how do you how do you get to make the declarations for people? So, they sign I basically am a registered agent. So, uh they they sign an agreement for me to operate on their

[20:13] behalf. But isn't it your agent that's doing it? So, then ultimately are we still doing it? It No. So, it's me um filling out the form, but I've got what I have AI do is collect. You have to put together in the same way that you have if like a if you have a chargeback and you're trying to prove that the chargeback is illegitimate for your for your company, you have to put together a package basically proving that this was a

[20:37] legitimate customer. We're using AI to basically create the entire takedown package. uh assembling the images, writing all the the legal jargon to do this and then using and then I go in and click the buttons to submit all the data. I see. So your vision is AI companies that require a human in the loop just make that human much more efficient. And what you do is you're looking around and saying what do I as a business owner do already? And maybe I already have these skills built up for myself. How do I sell them? And you make an offer on Twitter. let me see your site again just

[21:12] so we have that up on the screen while we do it or what do I see that's necessary for others and I will be the human in the loop that has all this automation designed to make my work super easy and all I have to do is the human part that's the model that we need to follow from this correct and I think like where people get caught up on is they think that AI is the selling point and in some cases it is but you know in this case with first index and with knockoff the people who are paying kind of don't give a rip

[21:40] that AI's is involved. It's because they just want the problem solved. There there's no for for both of those. There's not really any UI. There's not some like fancy app for them to be involved in. They set it up and then like they might get a report um that gives them some information, an update, but like they

[22:00] don't want to interact with my tools. Like they want the they want the thing done and their business to keep growing and they can focus on what they're good at. Well, Josh, why can't First Index, for example, just be a fully software as a service business where people sign up and then the whole thing is automated? What do they we what do they need you for? because they still the there's the there's judgment calls which I do think can eventually be automated but I like again that goes back to the software side of things where uh I people don't want to be involved like a developer

[22:32] doesn't want to have to go and like build a mini tool or whatever or they they want to just continue about building the product and then all the marketing stuff gets done for them essentially. All right. And finally, what other software are you using? What are you using to keep track of your client's results in LLM answers and and other

[22:52] tools? That's that's that's saved in notion because you have Claude go and ask chat GPT a question and then save the answer in notion to see where where they're showing essentially. Yeah. So like it's notion is just the database um plus some like files and and GitHub for sort of saving it and you personally or your agent is going in and asking the questions on behalf of your client's users and then storing where the answers are and then watching

[23:19] if it rises up. Correct. But then don't you need to like have it in different IP addresses or different users or anything like that? Isn't that why people use those tools, those third party tools? Yeah. Yeah. So, there's a little bit of like secret sauce there on like pulling that off. But like at the end of the day, yes, there's you're correct on what you're saying, but I am also pulling it off. So, but you're building it your your own version of it. I cannot believe you're

[23:43] using Notion as a database for for AO. That's unreal. Yeah, that makes me feel so much better about my notion database usage. I just store everything in there. there's so much stuff that you know there people have a tendency to want to over complicate things and uh at the end of the day like building something that solves a problem and it kind of doesn't matter what the tools are es certainly at this stage um you know it'll get cleaner as we go but right now it's it's fine being a little bit messy really impressive all right I love that you have the dot comfirstindex.com by the way when you talked about how AI is

[24:18] a prediction machine and at some point you need to shift to code I don't write my own code for things like this, but I come across these problems all the time. My solution is to use my sponsor. And if you're listening to this, you should see my video that I did with Wade, the founder of Zapier, where he said, "Listen, I recognize that this is an issue. Things end up going to LLMs that really should be set in code. Most people don't want to write their own code. That's what the nextgen zaps from

[24:42] Zapier will do." He has examples of it. I've got a link to nextg zaps from Zapier if you want Zapier to handle the code for you. In fact, Josh, I have a whole Grockbot whose whole mission is every day look at what all the other bots are doing and then see if any of it should be turned into code and then put it on Grockbots, put it on Zapier so

[25:00] that I don't pay money to Grock anymore. I just published how much I spend every every month on software and people said, "Why do you have all these lines for for cursor?" It's because my agents keep running out of tokens and I have to keep, you know, so my solution is to move more of it into code. And if you want that, that uh video is right here up on the screen. And I'll see you in

[25:18] that video next.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt