From NFTs to Seven Figures to AI Agency

Sean got into cryptocurrency and NFTs around November 2021. He got in touch with a really big influencer in the space, worked his way under his wing, and helped launch a project that did over seven figures in revenue in March 2022.

Summer of 2022 he was still focused on NFT stuff. Later in 2022 he started his first agency (more a glorified freelance operation). He started working with different projects, built up connections, leveraged those to market projects and help them launch.

Going into 2023, one of their most successful months was around March or April doing about $40,000.

"But that whole space honestly just burnt me out big time. I didn't feel like I was launching anything that actually had any value. I just didn't feel like I was building any actual skills," Sean explained.

One person he met in that space was doing AI development for a Series D biotech company that raised over $300 million. She started teaching him about AI.

He started creating content around AI, mainly AI news. His first viral video did 3 million views, got him 30,000 to 40,000 followers from that video alone.

They started working one-on-one with businesses, slowly moving into B2B AI implementation space. Started simple: Zapier automations, basic stuff. But over a few months, the progress and type of stuff he was building went exponential in difficulty.

Late 2023, he split with his old business partner. Partnered up with two other guys, built their agency up to $30,000 to $40,000 a month. Eventually he wanted to go solo. They were moving into info coaching, he liked the B2B, so he split and started his own thing.

Why AI UGC Creators Will Replace Humans

For a lot of companies now, especially going into 2025, it's not going to matter anyways. We're going to see a lot of AI UGC.

Sean put out a video in 2023 (before he initially went viral) talking about this Santa AI account on TikTok that amassed 150,000 followers. They were selling merch and everything, capitalizing off this AI Santa Avatar.

"I remember the video I was like: are we going to see AI influencers eventually? And getting into 2025, we've tested with AI avatars. There's a lot of different software coming out especially that is focused on UGC," Sean explained.

There's one called Lipdub AI. You can take a 10 second UGC video and the lip syncing is so good that you can't even tell it's dubbed over. You can make a video about any product basically.

There's so many of these AI generated influencers and UGC creators. This isn't a dig at any market, but in Sean's experience, UGC creators can be hard to deal with. Really, really difficult people to deal with.

A lot of companies (whether it's for better or worse or however you look at it) are going to look at that and be like: all right cool, we don't have to pay somebody and wonder whether they're going to deliver. We can test six, seven different hooks with this one creator and this one product and this one script. We can test that basically. It's going to cost us one tenth of what we would pay an actual creator. We're going to get the result instantly. We don't have to wait. We don't have to pay extra for usage rights. We don't have to wait for all the stuff to come back from that creator.

"The actual logistics and the ease of use, especially as the technology progresses, companies are just going to opt for these AI influencers, avatars to promote their products. We're already starting to see it," Sean said.

How to Actually Create AI Avatar Videos

I asked Sean to explain exactly what making an AI video on Instagram looks like. How do you make it? Tell me all about it.

Really you're going to need two software. Sean's simplifying it and won't talk about automating the process, but at a high level:

First software: HeyGen. Second software: 11 Labs.

With HeyGen, you only need four or five minutes of raw video to actually create an avatar. A lot of people botch this process because they don't put a lot of effort into creating the avatar or they do it wrong. They'll look away or do something stupid during the video and then it comes out looking weird. It might take you a few tries to actually get the avatar down.

If your first one comes out and you create a video with it and you're like "bro I look stupid," don't beat yourself up. It takes a few tries.

Once you do have one that you really like, HeyGen as good as it is in the actual video aspect really lacks in the voice department. You can 100% tell it's AI. It's so bad.

That's where 11 Labs comes in. 11 Labs has a $5.99 or $11.99 a month plan. They have something called professional voice cloning.

You have instant voice clone (where you get it back in a few minutes and it's okay but not good). The professional voice cloning, the minimum you can upload in terms of audio is 10 minutes (recommended is 30 minutes, and you can actually upload up to 3 hours of audio). The quality is you cannot tell that it's AI creating it.

"The process might look like: we have automated ideation systems. I have what's called an RSS feed setup and it's monitoring all of my competitors. Whenever they post, it's sent directly to this feed. We can basically see which videos are outliers in terms of performance," Sean explained.

If somebody in his niche (a direct competitor) posts a video about AI UGC creators and it's really an outlier (doing crazy views compared to what they usually get), they'll typically take that concept (obviously not just straight rip it) and Sean likes to script himself still because it's a skill he's really trying to build.

He'll take that concept, think about how he could add his own style to it (which is very interesting, maybe a little unprofessional at times, a little raunchy, but that's the content he likes making). He'll add his style, think about how he might go about doing it differently.

He'll have that script inside of their content system (an Airtable). He can click a button, it sends it to 11 Labs API and creates the actual voice transcription. From that he can send it directly to HeyGen which creates the avatar video.

Typically he'll create two different ones just to give the editors something to play with in case on one there's a weird part where it does like a weird face or something, just to give them variety in facial expressions.

Then he just waits for them to get it back and posts it.

The Surprising Performance of AI Content

I asked if AI content actually performs. Can people tell?

Sean's answer: nobody cares if it's AI or not. It really just comes down to the fundamentals. Is this video entertaining? Does it have a good hook? Does it have a good script? Does it have a good concept or idea? Is it controversial enough to get people emotional?

It doesn't matter if it's AI really, it just comes back down to the fundamentals of copywriting or just creating good creative content in general.

"Out of my 10 best performing videos in the last 2 months, almost half of them are AI. Completely AI generated. And they've also generated the most leads too. One of them has 10,000 views and 300 likes, but for an AI generated video that I put zero time into, that one got me six or seven qualified leads," Sean said.

He even tested with a buddy named Carson who created an avatar that wasn't even that good. You can definitely tell it's AI, it's weird looking. But he posted a video on LinkedIn (which is where he gets a lot of his clients, his main acquisition channel).

Normally his posts get three to five likes. This post got 60 interactions and engagements. That week they closed three new deals and one of them was a seven figure deal with a publicly traded company that's worth over $2 billion in market cap.

What AI Agents Actually Are

I asked Sean to explain what an AI agent is for people unfamiliar with the term.

The main difference between AI agents and something like ChatGPT or traditional software: with traditional software, everything has logic to it. If this person does this action, then this action happens. Super basic.

With AI itself, it's not acting on its own. You're always giving it something to do, or if it's part of an automation, it's a step by step process where you lay out the whole foundation. With AI it's like "hey write me an email." You're always giving it something to do.

But with an AI agent, the main difference is the fact that it's autonomous.

"If you create an AI agent and you say 'hey every day at 9am I want you to update the CRM and all the deals in there,' it's like okay cool, there's a task but there's no defined logic there. You're leaving it up to the AI agent and the tools that you've given it access to to determine what the best course of action is," Sean explained.

For example, if you're like "hey schedule a meeting with John," it's going to look at the availability on your calendar, look at the availability on John's calendar, and then it's going to draft up an email depending on the best time. Maybe it knows you don't have very many meetings at lunch, your workload is pretty light, John's around at that time, so it's going to align those two and make the decision on its own.

It's going to draft up the email, say "hey John let's hop on at this time."

That's really the main difference: it makes those decisions itself and determines the best course of action based on the task, the desired result, and the tools you've given it.

Universal Use Cases for AI Agents

I asked what the universal use case for AI agents will be. Is everyone going to have an AI agent that is almost like an administrative assistant that replies to all their emails or manages their stuff?

Administrative is big time, a massive one. VAs that are usually filling out spreadsheets, doing all that stuff, managing your email, managing your calendar, your bookings. Whatever you would hire an actual assistant for or hire out to a VA, AI agents 100%. That whole thing is gone.

The second biggest thing after that is sales. Looking back at the past year of all the clients Sean's worked with, the biggest pattern is a lot of people want appointment setters.

"You look at appointment setters, you run a big company, you have all of these appointment setters. There's so much involved with the logistics. We need to track performance, we need to have them fill out end of day reports every day. All right cool, this setter just hopped on with us and now he's out after a month, we got to rehire, we got to replace this person, we're going to have to pay this company to place someone on our team. There's so much that is just a headache with setters, appointment setters. And there's so many of them too. I think setters are gone next," Sean said.

He doesn't think closers are going anywhere anytime soon. CRM management would be kind of next after that.

Moving on beyond that, for bigger companies you're going to have a lot of agents with HR (moderated emails or whatever the case might be, making sure things are in compliance, a lot of legal aspects).

After that, marketing is a big one. Sean's already working with a company to build literally an ads agent.

How to Find Your Admin Time Sinks

The best way to figure out where your admin time is going: download a free extension on your Chrome browser called Clockify.

Throughout your workday, every time you're doing a new task, turn on that timer and track how long you're doing it. There's a little friction at first, but eventually it's second nature.

Do that for 1 or 2 weeks, then come back and look at where you're spending most of your time. See which tasks are admin tasks or something that could be handed off to a VA.

"Those are going to be the first places that you implement an AI agent," Sean said.

The Tools Sean Actually Uses

For B2B work, they build most internal tools inside of Airtable. It's a database platform but doubles as a lot more. You have automations, forms, but really where it gets special is interfaces. You can build custom analytics dashboards, end of day forms, Kanban boards, calendars. Airtable has become their CRM, content management platform, CSM, basically everything.

For content, Sean prefers Airtable or Notion. Airtable on the systems side is vastly superior than Notion. But as someone creative, Airtable doesn't inspire much creativity, so he uses Notion.

For automations, typically they use Make.com. On Make.com, you pay $10.59 a month for 10,000 operations. On Zapier for 750 tasks (basically the same thing), you're paying 30 bucks a month. Already you're getting over 10 times for one third the price.

If you want to build AI agents, there's a platform called n8n (the letter N, the number 8, then the letter N again, .io). That's where you build AI agents. Very steep learning curve.

With AI models, they use Claude and ChatGPT very heavily. Claude on the day-to-day and in a lot of automations. They use ChatGPT to fine-tune models. OpenAI as a platform makes fine-tuning very easy.

Fine-Tuning Versus RAG Explained

Fine-tuning is a method for training an AI model. The first is with prompting. The second is with RAG (retrieval augmented generation).

If you've messed around with custom GPTs where you upload a document to your knowledge base, that's using RAG. But RAG uses chunk processing. It can't use all 200 pages whenever you tell it to write an email.

Sean gave an analogy: you brought on a new chef for your restaurant, your nephew who has no idea what he's doing. You give him a big cookbook. He's going to flip through it, find the recipe, try his best to make something that somewhat resembles it. But he has no experience, so you're going to get something 60%, 70% of the way there. That's RAG.

When you have 200 pages of copy material, it scans through all of those, finds the four most relevant chunks that relate to the prompt. If there's no relevant content, the output is pretty minor.

Fine-tuning is different. This chef's been a chef for 30 years. You could give him three ingredients and he makes the world's best gourmet dish. He doesn't even need to look at anything.

"Fine-tuning is the same way. You're not just giving it a document as a reference. It's like you're plugging a flash drive into the back of its brain. You're uploading it to the model itself," Sean explained.

ChatGPT still has a data set (billions of examples it's trained on). Fine-tuning does the same thing but on a much smaller scale of your own copy. So it becomes hyper specific to that one style. You can get 95% perfect in just a one sentence prompt with no documents uploaded.

Is Building a Personal Brand Worth It?

I asked if building a personal brand is worth it and how can you start.

Yes, 100% worth it. Every single connection Sean has, every single door he's been able to get, it comes from posting on social media. His best, highest quality clients were inbound from social media content.

In terms of getting started, think about why you're doing it first. Most people watching are going to be doing it B2B.

"First and foremost, get numbers out of your head. Chances are if you're starting with B2B, you're not going to go viral straight off the gate. And you don't need numbers to get clients. There are people who have an audience that's not even a tenth the size of mine that makes 10 times more money than I do because they're very well positioned in B2B," Sean said.

Really the main thing to focus on is figuring out what problems your target market is facing. Focus on platforms like YouTube and LinkedIn.

Every client Sean's talked to gets the most clients from YouTube because people are looking for long form solutions.

Once you have a solid number of video ideas, just keep it as simple as possible. Don't overcomplicate things. Sit down, write out 10 videos, hammer them out, upload them to YouTube. The production quality is going to be crap at first. That's just how it is. But you'd be surprised how many crappy looking videos land clients.

The O3 Model: 81% Toward AGI

I asked Sean what he's excited about coming in AI.

The biggest thing: video. Google's releasing their V2 model. AI video is definitely the biggest thing. You look at these multi-million dollar blockbuster films. Eventually these are going to be things you can create from your own home.

One thing Sean touched on: OpenAI's O3 model. It's not released yet but they announced it.

The O3 model in terms of comparative performance has AGI testing basically.

"GPT-4O comparatively to what AGI would actually look like was at 11%. O3 was like 81% or something like that. And I was like holy crap. Now I'm not saying at 100% it's like okay cool we have AGI now, the world ends. I don't know how that benchmark works honestly," Sean said.

In terms of performance, in terms of coding capabilities, it was compared to the 125th best programmer in the world in terms of skill level. PhD level in terms of intelligence as well. It's able to beat out PhD level graduates.

My Main Takeaway

The biggest lesson from talking to Sean is that AI isn't coming in the future; it's already here, already replacing jobs, and the only question is whether you'll be ahead of the curve or buried by it.

Sean's journey from NFTs to seven figures to burning out to building a $30,000 to $40,000 per month AI agency proves you can pivot fast if you're willing to learn aggressively. When he got into AI in early 2023, he had no idea what he was doing. Started with simple Zapier automations. Over a few months, the progress went exponential in terms of difficulty.

The AI UGC creator replacement is already happening, not theoretical. Lipdub AI lets you take a 10 second UGC video with lip syncing so good you can't tell it's dubbed over. It costs one tenth of what you'd pay an actual creator. You get the result instantly. Veron Maya has 850,000 followers and most of his videos are HeyGen avatars.

The HeyGen plus 11 Labs workflow is simple: four to five minutes of raw video to create an avatar, then professional voice cloning on 11 Labs. Sean has automated ideation systems with RSS feeds monitoring competitors. Whenever they post outlier videos, take that concept, add his style, script in Airtable, click a button, sends to 11 Labs API creating voice, then sends to HeyGen creating the avatar.

The performance proves nobody cares if it's AI: out of Sean's 10 best performing videos in the last 2 months, almost half are AI, completely AI generated. They've also generated the most leads. His buddy Carson posted an AI video on LinkedIn (normally gets three to five likes), this post got 60 interactions. That week they closed three new deals, one of them a seven figure deal with a publicly traded company worth over $2 billion in market cap.

AI agents versus ChatGPT: with traditional software or even ChatGPT, you're always giving it something to do or laying out step by step logic. With AI agents, it's autonomous. You give it a task but no defined logic. You're leaving it up to the AI agent and the tools you've given it access to to determine the best course of action.

The universal use cases: administrative first (VAs filling out spreadsheets, managing email, managing calendar, anything you'd hire an assistant for). Sales second (appointment setters are gone next). Then CRM management, HR for bigger companies, and marketing.

Fine-tuning versus RAG: RAG (like custom GPTs uploading documents) uses chunk processing, scanning to find the four most relevant chunks. Fine-tuning is like plugging a flash drive into the back of the AI's brain. You're uploading your own custom coded data set to the model itself. You can get 95% perfect in just a one sentence prompt with no documents uploaded.

The tool stack that works: Airtable for CRM, content management, CSM, everything. Make.com for automations ($10.59/month for 10,000 operations versus Zapier $30/month for 750 tasks). N8n.io for building AI agents. Claude for creative outputs. ChatGPT only for fine-tuning models.

Personal branding is worth it if done right: every connection Sean has, every door opened, comes from posting on social media. His best clients were inbound from social media content. But get numbers out of your head. People with audiences not even a tenth the size make 10 times more money because they're very well positioned in B2B. Focus on YouTube and LinkedIn. Long form YouTube content: every client gets most clients from YouTube.

The O3 model benchmarking at 81% toward AGI (versus GPT-4O at 11%) and comparing to the 125th best programmer in the world and PhD level intelligence proves we're accelerating faster than anyone predicted.

Want to learn more from Sean? Follow him on Instagram, Twitter, LinkedIn, and TikTok at @ZoroTheWiz. Join his $47/month Skool group for step-by-step training on building content systems, fine-tuning AI models, automating ideation, CRM and CSM systems using Airtable, and more. Everyone is vetted to ensure high quality business owners for networking.

Listen to the full episode to hear more of Sean's insights on AI agents, automated personal branding systems, why Claude beats ChatGPT for creative work, and how to protect yourself as AI eliminates entire job categories in the next three years.