I Made $4M. Here’s The AI Automation Stack We Use transcript

In the last 24 months, our agency did around $4 million in revenue, and AI was a big part of how we were able to move that fast. But, not in the way people usually talk about it. It was not because ChatGPT wrote emails for us. It was because AI helped us build internal tools, test ideas faster, automate repetitive work, and turn workflows that used to take hours into systems that just run in the background. And in this video, I'm going to show you the actual AI automation stack that we use.

What we used to use before, what we use now, what has made us real money, and what I still would not fully automate. Let's get into it. All right. Most businesses are not bottlenecked by one thing. They're bottlenecked by a thousand small things. Someone has to update the sheet. Someone has to find content angles. Someone has to check what is trending. Someone has to review calls, and then someone has to build a prototype, and someone has to test a landing page, and the list goes on and on and on.

The dangerous part is that none of those tasks feel big enough to fix. But, when you add them all up together, they eat away at your company. That is where AI has helped us the most. Not replacing the business, but compressing the annoying parts of the business. AI does not replace judgment. AI compresses bottlenecks. Look, our stack has changed a lot. Earlier, we used to use a lot of Zapier, Notion, and tools like n8n for automations.

These tools were super useful at the time. But, recently, we have used them way less because tools like Hermes, Open Claw, and Claude changed what was possible. Our current stack is basically Claude, which builds prototypes and apps, Hermes, which we use for automations and workflows, Open Claw, which we use for agent workflows and operating system experiments, Notion for planning, content board, and knowledge base, and then tools like Google Drive, Sheets, Zapier, you know, the old ops layer.

That's still useful, but a lot less central. The biggest shift is Claude Code. Before, we would get quoted $20,000 to $25,000 for a prototype or an MVP web app. And then maybe another 5,000 for every new feature we wanted to test. But, now, with Claude Code, we can build a version in an afternoon. It might not be perfect. It might not be production ready on day one, but it's enough to test demand, and that changes everything because instead of asking, "Should we spend $25,000 on this idea?" we can ask, "Can we build a rough version today and see if anyone cares?" That is a completely different business rhythm.

The first workflow I want to show you is Founder Funnel. This is a product we are working on for founders who want to create better content, but do not have time to live on X all day. The problem is simple. Founders know they should post. They know they should react to what is happening in the market, but they're not going to sit on X 24/7 looking for trends. So, Founder Funnel helps surface what happened in the last 24 hours, what is relevant to you, what angles might be worth for you to respond to, but the important part is what it does not do.

It does not just auto write generic tweets in your voice because honestly, AI cannot truly write in your voice if it does not know what you think. It needs to ask questions. It needs your point of view. It needs your decision-making. So, the workflow is not trend to AI tweet to post. The workflow is trend to question to your opinion to stronger content on X. That is the difference between AI slop and an AI-assisted content system.

The tool does the research and the prompting. It guides you through the right content piece. You as a human still bring the judgment. And that's what it looks like completely vibe coded with Claude code. I come in. I decide to write a post. Say I want to react to a trending post. It will show me the different trends trending right now. I can select one and decide to generate content around it. It'll go around, formulate questions that it needs my answer, my point of view to come up with a genuinely good draft.

And once it has that, it will start asking me those questions one by one. And there you go. It starts asking me different questions. Do you think Anthropic's model of sending applied AI engineers directly into Blackstone's 200 plus portfolio companies actually threatens McKinsey Deloitte? So, it'll ask me relevant questions to that specific trend or news item. I'll just click uh skip on all of these. Uh it's going to give me something very generic because I didn't give it information.

Uh let's go for a short tweet for this one. Generate a draft. And it'll start drafting it right now. Mind you, it'll put in my thoughts. I just skipped everything for the sake of this tutorial. And there we go. Anthropic just built a $1.5 billion consulting arm to replace Deloitte and McKinsey. Ruthless, but the economics make sense. And now I have a tweet I can use to reply to or participate in this massive news that just came out in the AI space.

The second workflow is our clipping platform. This is at clips.unfungible.xyz. I built this with Claude Code and Open Claw a couple of months ago, and it has had a very real impact on our business. This one workflow brings in roughly 20 to 40,000 a month for us in revenue. That is what I mean when I say AI is not just a productivity toy. If you use it correctly, it can create internal software that turns into actual revenue.

And again, the point is not that Claude magically built a perfect business. The point is that the cost of testing the idea collapsed. It vanished to zero. Before, something like this would have been a long development cycle. Now, we can build, test, improve, and only invest more time when the market shows us it's something worth it, it's something that actually wants and craves. That is the real leverage. And this is how the Clips platform works.

You can sign up as a clipper, or if you're a brand, then we onboard you into the specific brand portal. But if you're a clipper, you log on, you see all the active campaigns, you can participate on them. This is the admin view, so this is only our view. No one else gets to see it, but we're able to manage campaigns, we're able to see everything we need as admins from the back end. And this is here you go, the revenue so far this month, $16,000, which is great to see from this one app.

The third workflow is our sales automation pipeline. My co-founder Leon has talked about this in more detail, but the high-level version is this. The system finds leads, it scores them, it gives us a way to contact them, we reach out, it puts them into our CRM, then after calls, it pulls in the transcript from those calls, it reviews how our closer did, it gives that closer feedback, and it recommends next steps. That is the perfect example of how we think about automation.

We're not automating the human relationship between us and potential client, we're automating the parts around that relationship. Things like lead research, scoring, CRM updates, call reviews, follow-up suggestions. The closer still has to sell, the founder still has to make decisions, the team still has to understand the client, but the system removes a lot of the admin drag around the process. So, the stack is not about having the most amount of tools, it's about knowing what each tool is for, and using it to the best of your ability.

Claude is for building, for example. Hermes is for automating workflows and helping with operations. Open Claw is for agent workflows and experiments where we want an assistant that can actually do things. Notion is for planning, boards, and shared context. Google Drive and Sheets and things like Zapier, they're still matter, but they're a lot less in the center of our system like they used to be. And the mental model I use is very simple.

If you pay yourself, say 10,000 a month, your time is roughly $62 an hour. So, ask yourself as a business owner, would I pay someone $62 an hour to fill out the school Google Sheet? Would I pay someone $62 an hour to copy data from one tool to another? Would I pay someone $62 an hour to search X for trends? Most of the time that answer is going to be obviously no. Those are the things you should automate first, not the core judgment, not the relationship, not the final decision-making.

Automate the grunt work around those things. That is the leverage. I do also want to be honest about what I would not fully automate just yet. I would not fully automate the critical decisions that you still have to make in your business. I would not let AI decide the entire strategy without human review. I would not let AI represent us to clients without oversight or at all. I would not let AI approve major financial or operational decisions on its own.

But I will absolutely use AI to prepare the data. I will use it to suggest options. I'll use it to show me competitor patterns. I'll use it to summarize calls, to build first prototypes, to find the next bottleneck. That is the balance between, you know, having an AI that helps you and having an AI that does everything for you. AI does the heavy lifting. You, as a human, you still get to make the call. So, that's it.

That's the AI automation stack behind the last 24 months of our business. If you want me to make a more tactical video where I break down one of those workflows step-by-step, comment which one you want, whether it's the founder funnel, the Clips platform, the sales automation pipeline, or our Hermes workflows, whichever one it is. And if you're building your own AI automation stack, start with one boring task that wastes your time every single week.

Do not start with the most complicated part of your business. Start with the thing you should never have been doing manually in the first place. As always, subscribe if you enjoyed this video, and I'll see you in the next one.