The Grind Behind the Magic: The Thousand Steps Behind Every AI Success

Ever wondered what it really takes to build a business with AI? Forget the three-word prompts and magical results you see on social media. The reality involves thousands of iterations, late nights of experimentation, and the persistence to keep going when the technology doesn't cooperate.

In this revealing episode, I pull back the curtain on my AI toolbox - the actual technologies powering Luxara's development. From my early days tinkering with computers in Ukraine to building custom PCs as a hobby, that technical foundation became unexpectedly valuable when diving into sophisticated AI systems.

My journey with ChatGPT alone has generated over a thousand pages of conversations since April, creating a knowledge repository that would have taken years to develop traditionally. I share how I've navigated the limitations of these models, from context windows to token limits, and why I use different AI systems for different tasks: ChatGPT for research, Gemini for creative work, and specialized tools like HiggsField for visual consistency.

The numbers tell the story: 819 website edits since May, thousands of images generated before finding the right visual identity, and countless hours spent stitching together short video clips to create Luxara's first promotional video. These aren't just statistics, they're evidence of the "thousand steps" philosophy that underlies every successful AI implementation.

While MIT recently reported that 95% of AI business implementations fail to show meaningful ROI, I break down why this happens and how I've worked to avoid these pitfalls as a solo founder. The ethical considerations of AI usage (transparency, accountability, and bias awareness) remain central to my approach.

Whether you're a founder exploring AI tools or simply curious about how these technologies work in real-world applications, this episode offers an honest look at both the immense potential and practical challenges of building with artificial intelligence. Because as I've learned firsthand: Luxara would never be possible without AI, but success comes from understanding both its capabilities and its limitations.

Chapters

• 0:11 Welcome to The Immigrant Hustle
• 0:50 My Tech Journey and Computer Building
• 8:02 Opening the AI Toolbox: ChatGPT
• 12:45 Creating Luxara's Visual Identity
• 15:03 Notion, Lovable, and Design Iterations
• 28:02 The 95% AI Implementation Failure Rate
• 34:42 Ethics and Responsible AI Use
• 38:36 Next Episode Preview and Closing

Transcript

Vladlen Stark: 0:11

Welcome back to the immigrant hustle. I'm your host Vl adlen Stark. If this is your first time tuning in to the immigrant hustle, it is an unfiltered CEO's playbook on how to build a business with AI. In this show, I'm pulling back the curtain on my journey. It's about the grind, the challenges and the incredible moments of discovery that come when you're building something from scratch. It's about the hustle. Last week we talked about the discipline and the how behind building Luxara on the side. Today we're diving even deeper into the how and opening up the AI toolbox.

Vladlen Stark: 0:50

As always, a little personal backstory about all this. My fascination with technology started long before the AI boom over the last few years and it goes back all the way to my days in Ukraine. I remember growing up I joined a little coding club at my grandfather's university and I was totally fascinated by what you could do, even back then, with a few lines of code. That curiosity stuck with me, and after we moved to Canada we didn't have a ton of resources to be spending on frivolous things like personal computers, but we did eventually get a pretty decent PC and I remember I was always tinkering with it and to this day I remember at the time Dell was a big supplier of computers and I guess they're still today. But I had this dream to have this Alienware computer with the little Alien logo on it and these fancy processor and GPU in it Totally a branding gimmick, to be quite honest, now that I know better, but it was a total fascination at the time and they looked really damn cool back then. By the time I had my first job, I always was setting a little bit of money aside for my next PC build and really since about early 2010s I have been building PCs from components, tinkering them together and really trying to get the damn thing to boot tinkering them together and really trying to get the damn thing to boot and that became a bit of a hobby of mine that I've always kept very private and now I'm sharing it with the world and for those of you watching, that really nice looking computer behind me is my latest and greatest build from about two years ago, nicknamed Bifrost, which is a nod to the Norse mythology and particularly Thor saga, with 14 fans and liquid cooling for the GPU and the processor and a lot of RGB lights in a rainbow pattern.

Vladlen Stark: 2:57

My wife always tells me that I should have been an engineer and not an accountant, because I always tinker and I always MacGyver my way around kinds of home projects, anywhere from electrical to building hexagonal pergola and pouring concrete on my backyard, with about zero masonry or carpentry knowledge. But hey, youtube, baby, and really this instinct to really understand the nuts and bolts of everything has truly been a blessing and a bit of a superpower in the age of AI, because with a lot of these models, they don't just work with a three word prompt, unlike social media is leading you to believe you. You really need to be thoughtful about it. You really need to know how to craft prompts and really how to piece things together. I remember when ChatGPT first came out, I was one of the first people lined up to give it a try and very quickly I ran up against some of the limits at the time and even to this day. I mean there's things like context windows and token limits and a lot of other technical jargon that I'm not going to get into. So there's a couple of key takeaways here. One you need patience and persistence with these models, especially as they develop. They have come an incredibly long way as of September of 2025. But there's certainly still ways to go, especially for some specialized applications. And two you really need to manage the output and your expectations of the output.

Vladlen Stark: 4:35

One of the biggest applications of AI for Luxara has been certainly on the visual asset side, and I mentioned in the last episode and I'll dive into a little bit deeper later on in this one that Higgs field was a particularly big win in that space because it allowed me to develop the consistency of character and really have that first digital Frankenstein moment, if you will, because over the course of trying a ton of different tools for image and video generation, I have gone through and this is no exaggeration thousands of different image generations before I landed on something that I liked and that was consistent. Again, it's not as they show on social media where you type one prompt and you can just take the output and put it on Instagram and call it a day. I mean, I guess you could do that, but I can't promise you that there'll be anything worth doing to really have the Luxara AI avatar explain to people what the business was all about and give people sort of this high level understanding. But it was also trying to showcase this tropical feel that I couldn't do from here, because unless I was down in Costa Rica or somewhere else tropical shooting this video, I would never be able to produce something like that, even with the advancements in video generation models and in image generation models.

Vladlen Stark: 6:13

At the time of the creation of Luxara 101 video, the best I could do was six to eight second clips coming out of VeO3 with what then was this brand new image to video functionality of VeO3. So when I finally put together these seven or eight little clips into one, I think it was just under a minute long video for Luxara that I then edited out about 45 seconds. That was a real geek out moment and even my wife, who's a self-admitted nerd, she's like yeah, that's too much, you're nerding out way too much with this, but it was really cool. It was really cool to see something really come to life after weeks upon weeks of experimenting and, like I said, thousands of different images generated, probably hundreds of clips generated through flow and VeO3, before I stitched this 45 second clip together. So it was a huge win for me in my AI journey and really, if I were to reproduce that, it would take me a very small fraction of time, which I then did twice for the hero videos of the Luxara website, for both the Canadian and Costa Rican front, because I had already learned the inner workings of these systems and how I could meaningfully combine these videos into something that looked pretty cool.

Vladlen Stark: 7:46

All right, so let's really open up the toolbox and talk through the tools. I briefly mentioned them in the last podcast, but today I want to dive a little bit deeper and give you a little bit more context for each one and what I've been using them for and how I've been using them. So, first and foremost, chat GPT unsurprisingly for, and how I've been using them. So, first and foremost, chat GPT unsurprisingly, chat GPT has been my researcher, my strategist and sometimes even my therapist and really, to be honest with you working with it, you quickly realize that, as I mentioned earlier, that it does have some limits and you need to navigate around them. If you wanted to ask it to output a 30-minute podcast text in a single prompt, it just would not work. So you need to make sure you understand how to structure your prompts and how to structure expectations around it. I've certainly run up against limitations, like I said, and it was anything from trying to ask it to generate a Word doc or a PDF with significant amount of content in it to outputting text of a particular podcast or research that I was trying to do. The context windows and the token limits have really expanded since ChatGPT 2 and 3 and 4 and now on to 5, there are still limitations to this day.

Vladlen Stark: 9:13

The more specific you're able to ask a question and the better context and role you're able to give to your LLM, the better the output's going to be. So in my case, a lot of the times I would start with a pretty general question, especially in the early days when I was just learning about high level securities regulations or international tax regulations, and then, when I would get my first answer, I would then specifically prompt about a particular section of a tax code or a particular law when it came to security regulations, and that gave me a lot better output. And this sort of iterative back and forth approach is really, in my opinion, the way to go, and rarely would you have your final complete answer in a single prompt, even if you had a really well engineered prompt. So just have that as your expectation to start with. The other thing I'll say is you know back to my comment about context limits and token limits.

Vladlen Stark: 10:19

So early on, and especially more recently, as I've been spending every evening on my PC really hammering these models, I have maxed out several chats and I don't know if you knew that you could do that, but you can, and you get this error message at the end saying that you've reached the limit of a given chat and you need to start a new chat for you to continue prompting the system. So what does that even mean? If you look at something like a chat GPT-4, at a base level it has a context window of about 8,000 tokens, which is about 6,000 words. Or if you translate it into pages 20 to 25 pages If you're using a different variant of ChaiGPT-4, then you can maybe move up to something like a 32,000 token context window, which is about 24,000 words, and that goes all the way up in the ChatGPT-4 model. I think at an ultra level or a turbo level, it was something like 128,000 tokens. So if you were to translate all of that into pages of text output, or in other words written by me as a prompt or output by ChatGPT as an answer, you're looking somewhere between 300 and 400 pages of text. That would be the exchange, and I've done that several times.

Vladlen Stark: 12:02

So I'm probably sitting somewhere north of a thousand pages of conversations with Chat GPT alone since April, which, if I say it like that, sounds a little bit crazy. But back to my comment in the last episode. It is a lot of grind and it is a lot of these small little wins that stack up over time. So I didn't get to a thousand pages on day one. It took me three months of nightly sessions with chad gbt to get to those kinds of numbers. But now I have a thousand plus pages of research and knowledge that I've been able to accumulate. That would have taken me years, if ever, in the past without these systems in place.

Vladlen Stark: 12:54

Yeah, so the last thing I'll mention about Chat GPT is just the repeat from what I said in the last episode, in that I found it to be quite good in its pro mode or its agent mode when it comes to technical research, really need to give it the role, the context, the question, the format for the output and any other specific notes that you're trying to get at as a little recipe for success, and I've had really incredible results come out of it that I then take either to Claude or Gemini to then refine it into a more creative or easy to understand copy. Because once you start going down the path of very professional research grade language in the chat, even asking it to change its tone and narrative style doesn't go all that well in my experience right now. So taking it to a different model and asking it to paraphrase or simplify it, or even starting a new chat inside of ChatGPT and asking it the same thing, that's really been kind of my go-to to make something a little bit more understandable for a conversation to be had later. And it really allowed me to prepare for meetings with lawyers or accountants or bankers a lot more, because I'll be the first to admit that areas of you know, say, security law or international tax or even commercial lending, have been outside of my area of focus in my career and I really needed to do a little bit of homework when I was heading into a lot of these meetings. So it's been totally invaluable getting a lot of these well-researched and well-referenced notes in a very professional, very jargon, heavy type research that comes out of chat gpt and then having it simplified by other models.

Vladlen Stark: 15:03

So the next tool is Notion, and it's the one that I said was my most underutilized in last episode and I've really spent one of the weekend days trying to change that. That was my little learning out of the last week's episode was to really challenge myself to utilize it more, which is what I've been doing last week and again, I'm still early on in the process but it has been incredible in the content scheduling and in planning, and what I learned just recently was that, because it has its own AI internal chatbot, I was able to download a template that I liked for content planning that would have the sort of an idea board and it would toggle ideas between the idea stage to writing stage, to recording stage, editing stage and so on and so forth. Then it would have a type of the content that I was trying to make whether it was short form or article or a post and then finally would have a calendar that would actually just let me schedule these various different social media posts or articles, whatever I was doing over time, and I was able to take my creative copy and my schedule that I previously came up with inside of Gemini and then just paste it into chat inside of Notion and ask Notion to create a content calendar, to create series of posts by week, starting on such and such date, and within seconds it populated everything and I have this massive column of the idea pile that I will, one by one, start moving into the writing pile, into the recording pile and then ultimately transition all the way to completion, and it scheduled everything on the calendar. I have my next 12 weeks planned out and it took me 15 seconds to do that. Mind you, I already had a plan long before this from a ton of other work, but just making it into something tangible, something visual, was incredibly fast. Same goes for the business plan and the financials. What I was able to do I was able to put the business plan into Notion by section, and then anybody on my team was able to get into Notion to take a look at it. They can very quickly search for it or they can just prompt the chatbot to find a particular section of the business plan or of the financials, and it's always up to date. It links back to my Google Drive where everything is stored on the Luxara, because it's a Google Suite company, and it's really been awesome. So, again, I'll keep working on Notion and that's something that I continue to challenge myself on and continue to learn on and lean on a few of my friends who I know have used it more extensively than I have. So stay tuned for more updates about Notion, but it's awesome.

Vladlen Stark: 18:15

Now the next one is lovable Lovable. In my last episode I called the number one tool that really allowed me to iterate a lot of ideas and even though it's not a tool designed for that, it's a tool designed for creating apps and websites. The work of creating the app and website for me was what allowed me to have this iterative back and forth evolution of ideas and really different experience of looking at the same problem from a different angle, as a customer or as a lender or as whatever else. So it's been super powerful in that. And to reflect back to my theme from the last episode, which was you know that building a business is a journey of a thousand steps and it's this idea of relentlessly stacking you know little wins, one on top of another over time. One of the cool things about lovable is that it actually gives you a statistic of how many edits you've done on a particular project and over what time frame, and I just looked at it and since May, I have done 819 edits to date, and that just goes to tell you that that let's round it up to a thousand times.

Vladlen Stark: 19:42

Something has changed on the website, something evolved in my thinking. Something was however small it was, it was updated right. So we're nowhere near having a Canada-ready website at this stage and I'm already at 819 edits. So I'll probably be well over 1000 before I'm ready to publish the Canadian website and if I have a team around me, then we'll probably hit two and three and 4000 edits. And it's something that I never really appreciated, having had different firms and consultants deliver on things like website design in the past. But now, having these statistics and seeing these iterations and really working through these iterations in real time, that has been a really powerful experience for me, I think. Even though it's very time consuming as a founder to do that, it is a really cool experience. So I would encourage those of you who are starting your own hustle to really do that as well, at least early on, until you have enough cash flow to support a hire or a service that could help you out with your website design and the iterative process around your systems.

Vladlen Stark: 21:03

So the next tool is Higgsfield. I already mentioned earlier in the episode that Higgsfield was a real saver when I was ready to give up after generating thousands upon thousands of images trying to come up with consistency in appearance and consistency in brand messaging as I was generating creative assets for the Luxara video and for Luxara website. So Higgsfield has been excellent and it really allowed me this consistency of character and and even though at this stage I haven't found anything that worked on a custom avatar, a lot of them offer you these sort of pre-packaged avatars and even those don't work perfectly well at this stage. So I'm really excited to see where the technology goes and how quickly we get to the point where I don't even need to generate these images of a starting frame and an ending frame and then prompting it to do this and that and the next thing, and transition in such a way and have lighting in such a way where we just get to the point where I can say, hey, luxara is walking by the pool talking to the investors and she's saying this, and here's our Q1 update and everything else is handled. I think that time will come. I don't know when it will be. It certainly has come a really long way, even since April or May when I started really tinkering with these models. So I'm pretty confident that in the next sort of six to 12 months we're going to get even further when it comes to these AI avatars that are a lot more powerful than they are today. So back to LLMs.

Vladlen Stark: 23:11

Gemini has been a central tool in the toolkit and you'll ask well, why are you using two different LLMs if you're already using ChatGPT to such an enormous extent? I mentioned before that I prefer Gemini for creative tasks. I like the style a lot better and it tends to simplify things and write things in a more consistent way, at least for the style preference that I have. So I basically just take the raw research and the high level ideas from Chat GPT, move it on to Gemini and then have Gemini do the actual creative narrative or creative script or whatever other copy that I'm working on.

Vladlen Stark: 23:55

And really, initially I was a little bit skeptical about Gemini, but then, when I made the decision to put Luxara on the Google Suite and I knew that I was going to be using VeO3, which I'll talk about in a minute for video generation, I thought, well, might as well start using Gemini. And I'm very happy that I did because, again, with my subscription for things like Google Workspace and my email hosting, I do have that paid tier of Gemini, which has been really powerful, and because I had to subscribe to the ultra level for the flow and vo3 capabilities, I've been able to utilize it a lot with much larger context windows, with much larger token allowances. So it's been really nice and, again, I'm very happy that I'm using both and I'm actually at a point where I'm considering Claude as well, purely because I've had really good feedback from some people who are working around website design and website optimization that Claude has been really good in SEO optimization and LLM SEO, which then puts your content into the research basically and into the knowledge base of the LLM model. So if you ask who is the best company out there for co-ownership of luxury real estate, chat gpt or Gemini would actually output Luxor as one of the answers. That's something that I'm looking at quite closely right now and I might end up going the clot route to get me there, but I'll try first with Chat gpt and Gemini and see how it works out. Try first with chat gpt and gemini and see how it works out.

Vladlen Stark: 25:45

So back to vo3 I mentioned that I was using it for video generation. I just used it this past week when putting a hero video together for the canadian portal of luxara which I was super happy about because now VeO3, as I said earlier is supporting not only the text to video but also image to video I was able to actually generate some pretty cool images at starting points and then have it generate these little video clips that I now put as a hero video for the Luxor Canada website. So it's been pretty cool, pretty cool as far as learnings go. I would encourage anyone who's using veo3 extensively one to look for it as part of other subscriptions that allow you some generations, so, for example, something like a leonardoai or higgs field. That would have veo3 as part of the stack and you would have a broader access to more tools than just veo3, because on its own it is quite expensive.

Vladlen Stark: 26:46

I think there's tools out there that can rival its video capabilities and its sound capabilities. A handful of months ago it totally broke the internet by being the first engine out there to publish these really crisp, high quality, consistent videos with sound. Now companies like ByteDance with their SeedDance models they've come out and they've sort of matched those capabilities and those models are a lot cheaper, if not free, and so, again, I would encourage you to just shop around. But if you're not able to find something that you like, reverting back to VeO3 is certainly a good option, because it is incredibly powerful, and now that Google has released its nano banana image generation, that's just another layer of why the Google tools are so good and they work really well together. Tools are so good and they work really well together. And then something like Google Flow, which is what I've been using to sort of stitch these clips together Another tool that I think was actually designed for filmmakers, to be quite honest, but I've been able to use it to stitch these longer videos together from the short VeO3 clips, so that's been a cool experience. It's something to check out again if you're looking for video generation.

Vladlen Stark: 28:01

Okay, so let's zoom out for a second and just talk about AI at a high level. There's been an article recently published by MIT that 95% of AI implementations in business have failed to show any meaningful ROI. Now, that headline sounds a little bit crazy, but if you really think about it, what has happened in the last 12, 24 months was that everybody got this mandate to start utilizing AI, setting off this unprecedented AI race, not only at a level of a provider so meta versus Google versus open AI versus ByteDance versus whoever but also at a corporate level, where competitors within the same industry for example, in oil and gas would have an arms race of who was utilizing AI more and how they were utilizing it. The big challenge was, though, that there isn't enough talent out there who really understand AI at such a level that allows them to deploy these enterprise-wide projects effectively. And two, I think there's still a pretty big knowledge gap and understanding gap at the executive level and at the board level when it comes to the use of AI. So when a board gives a directive to the management to implement an AI system, a lot of these directives have been too high level and not tactical enough to really deliver on the ROI, so that's been a huge challenge.

Vladlen Stark: 29:47

So what could you do to combat that? First, you need to have a clear strategy. Like I said, a lot of companies rushed to deploy AI projects without having a strategy of how they were going to actually utilize it in their specific business, in their specific use case and, most importantly, how the people were going to fit into that equation. That equation Because there's been a lot of buzz over oh, we're going to implement this AI and we're going to lay off 15% of our workforce, because AI is going to do X, y and Z and where that has had some success in super high tech companies like, say, meta and Microsoft, they're the ones building this stuff and they're the ones with all the domain expertise in this stuff. It's not the same when you're trying to lay off people who are technical in, say in my industry, in oil and gas, who are technical in a particular task, in exploration or maintenance or something like that, because those are different skill sets and to truly train AI to replace a role like that within your organization, it takes not only tremendous amount of resources and investment into the AI model. It also takes a lot of time to properly train it and not a lot of companies out there have the budgets and, more importantly, the people to fully train their own LLM or their own AI model, even if it's not an LLM, if it's some other AI application. So that has been a huge hurdle for people.

Vladlen Stark: 31:37

The second part is the lack of integration. Where it has been quite successful is in things like internal chatbots that just retrieve certain things from a database. I know there's been some applications in HR, for example, where you could prompt the bot to give you a policy on your sick days or, you know, retrieve a balance of my vacation days, or you know anything like that, as well as things like onboarding, for example, where it could pull some pretty static manuals on onboarding and then it would provide you that information. But hallucinations has been a huge problem and even with techniques like RAG or retrieval of metageneration, a lot of the times these models would fail and hallucinate regardless, and there's been some really recent research into that and I think there's going to be some pretty quick wins into overcoming the hallucinations in the coming months. But as of right now, that's been one of the big hurdles is integrating it with your existing systems and with your existing data.

Vladlen Stark: 32:49

The next issue that I've seen in many different cases is really just the change management. It's not a new issue for AI. That's always been the case with every new technology that's come out. People have been failing for decades of the technology life cycles, especially if we look back to things like big ERPs coming out a few decades ago, then the cloud architecture, now with AI. It's just an endless cycle of having to manage people's transition through a particular technology use.

Vladlen Stark: 33:24

I was quite fortunate actually to be on a panel with somebody from Microsoft a few months ago and their take was that they believe that over time we would develop a number of these hyper specialized LLMs that are plug and play for particular industries and that really sort of mimics what happened with a lot of the ERPs that have really evolved into these industry specific or even sub industry specific ERPs that are highly customizable, highly valuable for one particular use case but not so much transferable across the board, compared to something like, say, sap. That is this huge behemoth that could work on any industry, any company, or Microsoft Dynamics or Oracle. I don't want to get political here, but bottom line is that we would get to these hyper specialized systems that would be more plug and play. We're certainly not there right now and hopefully we do get there at some stage, because I think that's the real win for a lot of smaller companies and solopreneurs that don't have the resources to train an entire model to do something special in your niche, which is one of the biggest challenges that I've had with this.

Vladlen Stark: 34:42

One last comment about productivity and ethics, and that's really around. You know these statistics that get thrown around that AI can improve productivity by something like 25%, let's say. For a team, that could mean a couple hours a day. Multiply that by a week huge win, huge productivity boost. For somebody like myself who only has a few hours a day to begin with, ai productivity is really the only way to accomplish what I'm doing. But really you always have to consider the ethical side of things when you're working with this, especially when you're working as a solo founder like myself, because the puck always stops with you. So you could look at best practices around ethics in AI and you could look at something like the pillars from Microsoft or the AI Act in Europe that outlines responsible use of AI.

Vladlen Stark: 35:44

But, bottom line, there's a couple of concepts that are really just present across all of these different frameworks, and a few in particular that I want to mention here. So one is transparency, and that's around use of AI. So if you're looking at something that is generated by AIs, if you're chatting to an AI bot, that should be disclosed and given to your users so that they know that they're interacting with AI. Number two is accountability. That's another huge thing. Accountability that's another huge thing. If the output is wrong and it gives you wrong advice or it gives your customer wrong suggestion or wrong booking or wrong answer to their query, you have to take the full accountability. You know you can't just go blaming the system.

Vladlen Stark: 36:35

And the last point I'll make here is around bias. Bias has been a huge issue with machine learning and now with AI models, and it's simply based on the training data that LLM or ML is trained on. And, as I mentioned earlier about these hyper-specialized systems, I don't have the resources to really train a model on Luxara in particular. I have to rely on more general queries and more specific prompts to get me the right answer. So I'm relying on the publicly available LLMs that are trained on sort of the breadth of data and I de-risk myself by relying on open AI and Google to make sure that those models are the least biased versions they possibly could be. But at the end of the day, what that means practically for me is that I always have to check the output and I always cross verify anything that is of significance between multiple different models and, if I'm totally honest, sometimes I just Google things and go outside the LLM space just for that final check and more often than not lately, as I get more and more technical about some specific issues, and I always consult a lawyer or an accountant or an expert in a particular thing.

Vladlen Stark: 38:04

Don't get me wrong I believe AI is an equalizer and enabler of opportunity like we've never seen before. I'll say it one more time Luxara would never be possible without the use of AI. Just make sure that you navigate AI appropriately, you keep in mind the ethics, you keep in mind the limitations and you really sit back and question everything that comes out of AI and really try to understand it beyond just blindly trusting the model. Thank you for tuning in again.

Vladlen Stark: 38:39

Next week we're going to dive into the people and even though I am a solo founder and I'm the only employee of Luxor Canada, there are a lot of people behind the scenes. There are mentors, there are advisors, there are your friends and your family who support you along the way. Really become your support network, your champions and, ultimately, somebody you can share your wins with. When this thing is all said and done and you sold your company to Meta for $500 million, the best way to support the hustle is to share it with someone else. You know somebody who might be in the trenches building their dream right now. If this episode or any of the previous episodes have given you any ideas or provided you with any motivation, I ask you once again and always, that you pay that forward, whether that's by sharing this with someone else who might need to hear it, or sharing your own story, or giving somebody a helping hand. Let's build this community together. I'll see you next week.