Transcription
So I've built my first AI agent startup back in 2023 when nobody was talking about AI agents or LangChain. I've also created my AI agency at the same time as well, and I've been running that for around 2 years. Through that time, I managed to speak with thousands of users, clients, and experts in the space. And through that and my own experience, I managed to recognize real patterns and big pivots that are about to happen in this space. And I might piss some people off here, but I just want to tell you the uncomfortable truth.
If you want to make a dent in the AI space and all you're doing is binge-watching all these N8N masterclasses, which essentially allows you to become a master of N8N, which is just becoming a master of clicking stuff and typing stuff on a computer, I have to tell you that you're setting yourself up for failure. And why is this? Well, I'm going to put it from an AI agency point of view. So the way we are approaching, and you know, people are teaching the way we should build and sell solutions for our AI agencies is just getting outdated as we speak. And so in this video, I want to show you the new approach that I have taken and I am currently taking to approach the AI space and how to sell AI tools and solutions.
Why am I sharing this? Well, somebody's going to share it at some point, and I think it's been slightly gatekept for some time. The space is open to everyone. The AI space is really, really big, and I know there's potential for all of you to build amazing solutions. So in this video, we are going to cover the exact strategy that I am currently approaching at the moment and, you know, based on that, the plans that I have for the rest of the year and next year. So let's get straight to it.
Just a little recap on this. So this is LangChain. My first video was one and a half years ago. I believe it was a RAG chat, but nobody was talking about that at that point. That's why I took the approach and, you know, built my software startup because, you know, that was completely new, and I knew it was going to change the world. And well, here we are. So I've done a thousand things, but this is where we're at. I started early in the visionaries area here, or the early adopters phase. I was at the start of this. So now we are essentially crossing the chasm over here. I started here. Now we are somewhere around here, or even in the middle as well. That's where we are right now. And when I was in the early adopters phase, the following that I'm showing you here was pretty hard. Okay? You had to scrape API docs if you wanted to build any AI solutions, even no-code solutions as well. It applies to both. So scraping API documentation to find, you know, how to call the right API that wasn't available in an automated way, similar to how we have it now on any N8N and applying RAG, Graph RAG. There's no good documentation for that. Fine-tuning, even worse. Troubleshooting, right? Anything using LLMs, using ChatGPT. Okay, ChatGPT was very, very dumb back then. And knowing about state-of-the-art models and techniques, which back then was more useful than it is now. Again, it was hard to keep up with all this stuff. But this meant that it was valuable to do. If something is hard to do, it's usually valuable, okay? That's how you usually measure this stuff.
But the current state of things is the following. So tools are getting easier. Tutorials are everywhere. Not like me. When I got started, there was barely anything. Interns can now do a lot of work. What used to take me five months, four months to perfect. As I said, ChatGPT Cloud can generate nearly finished anytime workflows because they have connections to N8N and documentations. So essentially, these platforms that you see over here in the top part, they have solved the first layer problems, and also ChatGPT and Cloud. Okay. So we also have templates. Okay. And most essentially, now with software, we've reached the software layer where AI agents are making it super easy to create prototypes and MVPs. I just want to show you what you see on the bottom here. If you don't know what that is, it's a Figma prototype. It's a Figma wireframe. So back then, when I was building my startup and I was building other apps, this was the reliable way to do it. There was no Lovable. There was no Bolt, or I think that's well, at least I didn't know about it, or nobody knew. So this was the most reliable way to do it. And then you would tell your developers to build it for you, or you would build it yourself, but you needed some, you know, design guidance. But now you can skip all that. You don't really need this. You can just go straight on to Lovable, chat about what you want, and then it's going to help you with that design that then you can push into code, or you can just launch it as a software with Lovable, right? We're getting to that point already. I think Lovable has the Lovable Cloud feature now. So you can connect to the base easily. So it's becoming extremely easy now. You can't just say I build AI solutions for you, because everybody can.
What is the moat then? What uniqueness can you have in order to be able to sell AI solutions and AI tools reliably? The answer is well, specialized industry knowledge, okay? As tools are becoming easier to use, then everybody's going to be able to use these tools. N8N, okay? It's not as hard as people think. But now, what's important is what do you build with them? Okay? So even though you're watching, okay, here's this new N8N workflow, this, that, this, if you don't apply it to solve a very specific pain point that hurts, then that information becomes irrelevant. It's not useful. So you need to understand a problem very well that hurts, and then use these tools to build on top of that. Okay? And I'm going to explain you later how to do that in the most effective manner. And that's how you build a moat around your AI solutions and tools. And obviously, after that, once you build it, it's important to know how to distribute and sell. So I can discuss that in a later video if you want. Whatever you want, just let me know down in the comments.
Essentially, that and how do you attain specialized knowledge? Well, look at yourself first. You might know stuff that other people don't know about. So the best way to get started is, well, do you know or have worked in a specific industry? I'm sure most of you have. And if you haven't, go talk to people. Go talk to people that worked in an industry that, well, preferably you like. The second thing that you should ask yourself is, do you understand some of the biggest pain points any employees of yours or any businesses that you know in that specific industry have? The last one is, do you understand an exact process or framework on how to solve the specific pain point? You might not know the last two bullet points, but you can definitely answer the first pain point. These questions are really important for you to ask because they are going to scale up to what I'm going to explain next. And I am going to go through two examples. Okay.
The first one is when I was building my first AI agent startup, my software, DocSphere. So what I did is that I spotted a pain point. In this case, it was my own pain point. As I said, you could go to other people and spot pain points there. But the main thing is that I spotted my own pain point, which was analyzing Excel data and creating graphs with it. That was my own pain point. It was long. It was unbearable. And then I also had to export it onto Power BI. And the whole process was just a long mess. And it still is. People are still using it and still doing this manual process in large enterprises. But what I did then is I got that pain point. I started talking to people. I said, "Okay, are you having this problem as well?" and tried to dig deeper if that's the exact pain point or there's a deeper pain point that is much worse. Okay? You want to essentially touch on that pain point that really, really hurts and that people would pay for. So I found that, and it's just creating reports from Excel. It's literally the same process that I was going for. So I was lucky that the process that I found painful, also other people found as painful. And then this is very important. You want to try to ask people to pay for your solution. You would say, "If I build this, will you pay for it?" And how much? The how much is not that relevant at this point, but you want to make sure they pay you something. That means they're finding value in solving that pain point. It's very, very important. So before getting paid, actually, I built a pilot, very, very scrappy. So I built it on Python and something called Streamlit, which allows you to build front-end. So this was before Lovable and all that. Okay. So I built a quick script on Python that would search on Excel and also do visualizations. So I did this using Python Streamlit. I was asking for a lot of people. I found one that paid me a couple bucks for it. And with that user, what I did is I tested and refined that script. So there's a lot of manual things that I had to do along the way, right? Not everything was from the script. I had to manually click stuff, import because I didn't know fully how it worked. But the point is that I wanted to fix the pain point first before automating anything. But once I did for that first paying user, I built this crappy solution for them. I then started to look for other people that would pay for me. That's the stage where you start refining and automating more. And then once you have a proven end-to-end solution that you're like, "Okay, well, this works. A lot of people are using it at the moment. Okay, let me scale that. Let me allow more people to use it without having to be involved in the process," and that's when you turn that into software. And this is what I built. So DocSphere over here. And yeah, it was about that. So that's when I started doing the distribution and getting on sales calls and so on and so on. But even before this, I was just calling people, getting on Google Meets, and just reaching out to friends of friends or friends of employees. Very, very scrappy. And that's the whole point. You want to get from point A to B or from 0 to 1 as quick as possible, no matter how scrappy it is. So that's what you've got to learn.
This example is probably more tailored to most of you that are starting AI agencies, and I guess that's why you're watching my videos. So I did this as well with Architect over here. Funny thing is that I started as a very generic agency like most YouTubers and most people started back, back in the day. So I could allow myself to be that generic agency before because nobody knew about the tools involved and what APIs to call and how to call them. Okay? So you could say it was proprietary information back then. Okay? But then, as competition started growing, I had realized, "Okay, well, I have to find a very specific niche." And that's why all these vertical software startups, right, the AI agents that are focusing on one thing, that's why they are dominating the space, right? There's one for video editing, there's one for image generation, they're doing really well because they're focusing on one single problem. And because they're doing that, they are advancing so quick than all these generic horizontal tools. Okay? But basically, now, especially now, what you want to try to do is build a specific service, get a very specific offer. You're going to solve X for Y. You're going to fulfill that specific service. You're going to try to get a client, whether it's a pilot or, you know, you tell them, you tell them that you have the solution and you just build it. Okay? Very scrappy way. You're going to use that first client, even if it's free. Okay? Or, I mean, ideally, you tell them to pay, but very low amount. Make it an offer they can't refuse because you're going to use that first client as a piggy bank, as a test pig. Sorry if any of my first clients are watching this, but you've got to start like that. You're going to test, refine the process, and then once you have something that you've proven to the first client that provides value to them and is solving a painful solution, then you're going to reach to more clients. You're going to refine, you're going to automate, right? You want to add more N8N workflows, you want to add a CRM, completely fine. Then you add an interface. You can do this with Airtable, Lovable interface from the last video, which I think many of you did not like that much. And then you're going to turn that into software. Start using Lovable stuff like that. Okay? So that's the natural progression that you should follow in order to build a tool that's going to stand the test of time. Well, at least in the AI space, or you have the highest chance of standing, rather than just staying as a general AI agency. Okay?
So this is the process, and I'm going to be specific about this. Okay? I'm going to give you an actual example. So I can't say names because this is a, this was a big company. So they are a Saudi Arabian logistics company, and they came to me and they said, "Oh, I have this, this problem, whatever." Okay? I barely spoke. I barely told them, "This is what I have." This is, I just listened. I listened to what they had to say, and through that, I tried to dig into the actual problem. So they were telling me a bunch of stuff, and they have problems all over the place, but as an agency, you've got to, you know, get the problem that's the biggest pain point and start with that. Okay? Start with one problem that's giving the biggest pain points. So their problem was that they were having a hard time following up with customer support and tracking packages for clients in WhatsApp. So they were dealing, they were coordinating end-users and suppliers at the same time, and just handling this side was very, very tedious. So what I did, I started with an N8N flow, very simple N8N flow on WhatsApp and a Google Sheet. Okay? I called it CRM just because I wanted to be nice to Google Sheets, but that's what I did. I laid them with Google Sheet. That's it. Scrappy, unscalable. That's what you want. And through that, we got information. By doing it unscalably and scrappy, we got information. I actually understood every single API that I had to call. I ran into a million errors. I knew how to solve them. I knew through connecting to Google Sheets what data is going where, how do I pull that data, where do I push it? Okay? So it gave me a lot of information that it's hard to get if, let's say, you're building something on Lovable and you don't understand the code, right? So this is a great way to understand the process, especially as a non-coder. And then I got paid. Then I got paid. And this proves there's value in the process. If someone pays you, that means that the pain is big enough for them to pay you. So that's a good, you're validated that this works, okay? That's probably the biggest step in this journey is that they pay you.
Now, you want to prove the process with other clients. Okay? You don't need to automate the scrappy parts yet. You can do a bit if you want, but don't go overboard. You just want to keep testing with other clients. Naturally, you're going to start automating your process out of necessity because so many people are going to come in that you're going to figure out. So, once that happens and you see that there's actually room now for the process to be automated, you know, you see that the process becomes repeatable. You're like, "Okay, exactly. Need to know what to do here, here, here, here, here." Then you can start automating as much as possible. And this just helps to scale, right? This just helps to scale your impact with multiple businesses, multiple clients with you being involved as less as possible. And once you get to that point, then the next step to scale is turn it into software. So you've essentially seen this in both examples. You've seen how I'm going from a service, how I'm automating that service as much as possible in order to understand what APIs to call, what data to be transformed and manipulated. Now, once I know that, then I can build the software as a service. I know exactly how it should be orchestrated, or either you do it, or you have from automating it on an end, which is not as hard. Now you have the specific knowledge on what APIs to call and all that to give it to a developer. Now you can map out the process. A developer can then code it. They are meant for that.
And I wanted to drop this here because I think many of you need this advice. Many of you have asked me this. How to sell AI solutions, tools as an AI agency in the most effective way. There's a lot of playbooks, very long videos about this. No, you don't. It's very simple. So the key point is that they don't buy automations, they buy business outcomes. Okay? So if you are in a call with them and you explain how complex this automation is, how it has 40 nodes and it's scraping in a loop, 60 different, they don't care. Okay? They don't care. I'm giving you money. Are you making me more money than I'm giving you in terms of, you know, revenue? You're bringing me more leads, or are you saving me time? Because you can actually translate time saved into money. So those are the two variables that people care about. That's all it is. If you can convince them that either one or the other or both is true, then it's an offer that's hard to refuse. So, they want solutions to business outcomes or pain points. Okay? Has to be painful enough. Otherwise, they're not going to pay, especially if you're doing outreach and, and calling on them on the phone and all that. Okay? If you want that to work, first of all, it has to be a big pain point. But yeah, here are some good, good examples. So, we removed the need for you to message people manually on WhatsApp. Manually following up with people on WhatsApp is very, very painful. Okay? So we're going to remove the need for you to do that, and an AI agent is going to handle that for you. You don't even need to say agent. Don't even need to mention AI, especially now because it's kind of a tacky thing to say. We will bring you three times more leads without lifting a single finger. You have to do nothing. All you have to do is do a consultation with us, or, you know, do the setup, or whatever. And after a month or two, it's just going to the leads are going to come in automatically because we have a system that's going to do it for you.
And when you're having a chat with a client and maybe they don't know what they want, I have this issue. Okay? You need to be very clear on how you want to help them because depending on what you do, it can become a black hole. Okay? This is, this is a different topic. This is more of, you know, they're coming to you with a problem and they have an issue, and you want to be their AI transformation partner, or, you know, they need clarity from you. So you need to decide. We can't automate this. Be honest. You can slightly, you know, give yourself more confidence, but there's some places where it's impossible to automate. Then you show them, "This is an overview of the approach. If you can automate it, this is how much this new automated process will save you." And you can save them time or bring them more money. Okay? So again, those are the two big variables. For example, if we bring you 20 leads a month extra, the lead value is $3,000. Then I multiply that by 20, which is the amount of leads, and I'm bringing you $60,000. And I'm only asking you for $3,000 a month. You, as a business owner, would you be happy with that? So subtract 3,000, you're, I'm giving you $57,000, right? I mean, the cost is great for me. I'm getting a decent amount of money, but if I'm giving you $57,000, then it's a win-win situation, right? So, it's about phrasing it in a way that clarifies to the client that he's also winning. He's actually winning more than you. So obviously, there's a lot of assumptions to take into place. Maybe all the leads that don't go through. Maybe there's an appointment setter that ups the call, or whatever, and maybe they drop. But that's not your problem. Okay? You're giving, that's why it's good to make assumptions. It's good to put estimates. So this is the estimate amount that we think we might bring you, or just say "estimate amount." That works better. We save your staff 10 hours a week. Okay? This is the time one that I was explaining to you about. So let's say we're building an AI editing video editing tool, a very custom one. Okay? That's not out in the market. The cost per video is $60. If we do this for one week for you, it's going to be $60 * 7, $420. And by the way, I just realized 420 now. It was a complete coincidence. So, we are going to charge you $2,000. Okay? This is one special word as well. It's going to be an investment. You're going to invest in us $2,000. And I don't know how many months, I didn't calculate it, but in 3 months onwards, from the 3 months timeline onwards, you're going to be generating money in terms of time saved. Okay? So this is also a good play if they're willing to wait 3 months until they start getting their time back. So these are two big approaches, or maybe the two main approaches in which you want to explain to your client what are the advantages of them onboarding with you.
Okay? And before you call me crazy for making this up, well, I haven't made this up. The service, automated service, or package service, and then SAS is nothing new. It's just that I added the AI service part, but it works the same as most of the giants, most, most of the successful giants that you know have done it. So Mailchimp, don't know if you know Mailchimp, but it's an email marketing platform, okay? They automate email marketing for you. So they started as a design marketing agency. So they offered email campaign management as a package service. Then they automated that in the background, and through that, they automated pretty much everything at some point. And then they turned it into a software product, software as a service. And then we have Stripe, similar. So the Collison brothers, I mean, you know, you know Stripe. I think all of you use Stripe or have used it. They helped startups integrate payments directly into their apps in a very, very manual way. And they just refined it by going through client and client and client into repeatable API integrations. Then they created the API out of it, and look at them now. So yeah, pretty, pretty big.
So I just want to recap how this works so you can take it home in a bite-sized nugget. So what you want to do is find a specific problem. Find a specific problem. It doesn't have to be a painful problem right now, but find a problem that you at least like. It's going to matter later when you build something big out of it. You're going to validate and understand the problem for other people that have this problem and are in pain. It's very important that they are in pain. Okay? Otherwise, they're not going to pay. If not in pain, they're not going to pay. And if they're not in pain, you could just go back and find another problem. Usually, for most scenarios, you're definitely, the world is really big. You can find someone that is in pain. I mean, it has to be a good enough problem as well. So you have to be somewhat reasonable with your problems, but yeah, you could just look in yourself and find something like that.
But then you're going to create a scrappy solution. So, once they paid you, right, you're going on the agency route, they paid you, then you're going to create a very scrappy solution for the client or user. I suggest client because you can charge more. Then you can automate, optimize, and turn it into software. First of all, get N8N. You can just go and use N8N straight away here. Okay? That's part of the scrappy solution. I'll explain now how you can do that. I'll explain tools in a second. And yeah, then optimize, turn into software. You don't have to do, you don't have to build the software yourself. What you can do is find a CTO, find a partner that knows how to code, and then you will just verify that for them. I mean, they will help you build that for you. You are the one that has the specialized knowledge. So you're going to map out for him what is he or she going to build, and you can just work together or hire a developer, whatever you want. If you want to build something big, it's better to get a partner, a co-founder. But anyways, these are the tools. Especially now with the AI agency setup, now you have access to N8N. I covered N8N. You become a master in N8N. One of the courses in my private community, so that is covered. CRM as well, so put it on Airtable. Ideally, do it on Airtable. Lovable, I know it's a bit expensive. There's an alternative called SeaTable, like you're surfing in the sea, and then table. So that's an open, that's a really, really cheap solution that is open source as well. I know you guys like open source. Then from scrappy, we're going into a prototype. So now we're using Lovable, maybe plus N8N, attach some webhooks. I have a video about how to build a software with Lovable and N8N if you want to see it, and connect Superbase. Okay? Or you can do it even scrappier. It's Google Sheets, but I recommend you do Superbase because Lovable has now the cloud version, I think. So it's much easier to do that. From there, you're going to turn it into an MVP. So something that you can showcase to people, and you would need Cursor or an IDE for that. The term MVP, prototype, can be exchanged a bit more. I just wanted to add them both to differentiate between Lovable and, you know, moving on to something more serious, and Cursor or an IDE. But this is where you would need a bit of technical experience. Maybe you want to have someone, some other one, some other person step in. But what I have done, I'm not a technical person. I'm just technical enough to be able to learn along the way. And I've built really good prototypes with Cursor. So I recommend you do that. You can ask questions along the way, and, you know, you can put on thinking mode, not on execution mode, and you can ask, "Well, what does this do? I want to connect an API." You can figure out on the way as well, as long as you have the overall set of actions that you want to automate, then that's completely fine. But give it a shot. Give Cursor a shot. CloudCode with the launch, yeah, there's a 5-hour window that is not very nice. So Cursor, I recommend. And then once you have that, then that's when you can hire a professional developer, hire a co-founder to do this for you. They're going to be surprised by, you know, what you've built by this point. If you have gone through the Cursor path or Lovable path, and they're probably going to want to join you if the mission is good enough and there's obviously benefit for them to do so.
So there you go. Take it, use it, whatever you want. I'm already doing this, so I'm winning money with this. So take it and use it. Honestly, there's space for everybody. And yeah, let me know what you built in the school community. I'm really excited to see what you guys build and how you follow this approach. And yeah, that's it. I just want to leave it here. This is probably the, the one of the most important slides here. It just summarizes everything. So service, prototype service, or AI-powered service, and then SAS. So hopefully you learned a lot in this video. Okay, this is the path that I'm following right now, that a lot of YouTubers that have agencies are following, or if you're not as a YouTuber or teaching this, you should follow this. The most scalable way, the easiest way to make money with the AI space. And the window's closing, so make sure you act soon, okay?
And if you want to see any other topics that I mentioned on this video, distribution, how to sell, you know, using LinkedIn, you know, all these distribution methods to be able to push your app. Let me know in the comments below. If you like this video, please give me a thumbs up and a like. I would really appreciate it. It helps the channel a lot. And if you didn't like it, please let me know down in the comments below. Okay, I'm very hungry. I'm going to go now for dinner. But yeah, any questions again, just drop it again in the comments below. Thank you so much, and I'll see you in the next one.