Inside the AI Systems That Run My Business

Last Tuesday I woke up, made coffee, opened my laptop, and realized that in the last 24 hours, eleven Pinterest pins had gone live, a Substack article I wrote on Monday was already formatted for Facebook and Instagram, and my VA had flagged two things for my review with a note that said “everything else is scheduled through Friday.”

It was 8am. I hadn’t done anything yet. And the business was already running.

I have one VA who works 20 hours a week. That’s it. One person besides me. A year ago, that workload would have required a content writer, a social media manager, a graphic designer, a Pinterest strategist, and probably a project manager to keep them all talking to each other. I know because I’ve run businesses with teams that size before. I know what the overhead feels like and how much time gets eaten by communication alone.

This article is the behind the scenes of how it actually runs. Not the highlight reel version. The operational reality, including where it breaks.

What the Setup Actually Looks Like

I have three AI tools that run the core of my business. Claude for writing and content creation. Google Gemini for research and keyword discovery. Higgsfield for visual content and video. Those three do the heavy daily work. I also use HeyGen, Pomelli by Google Labs, and other tools as needs come up, but the foundation is those three. Everything else rotates in and out depending on the project.

Claude is where most of the heavy lifting happens. I draft articles, build training scripts, create email sequences, write social posts, and develop product content inside Claude. But I’m not just typing “write me a blog post” and copying what comes out. I’ve built a system around it with voice samples, project-level context that carries across sessions, and a guardrails document with 98 specific patterns Claude knows to avoid. The output is close to finished when it comes out.

Gemini handles research. When I need to validate a keyword strategy, find what people are actually searching for, or dig into a topic before I write about it, that’s where I go.

Higgsfield is the visual layer. Video content, AI avatars, graphics. The stuff that used to require either being on camera every day or hiring a designer.

My VA takes what these systems produce and moves it through the pipeline. Pinterest scheduling, content formatting across platforms, and managing the posting calendar. She handles the operational pieces that need a human but don’t need to be me, and the strategy documents I’ve built tell her exactly how to execute without guessing.

That’s the tidy version. Here’s what it actually took to get there.

What the Clean Version Leaves Out

The math IS concrete. I co-write my Substack articles with Claude. That article gets repurposed into Facebook and Instagram posts. My VA generates Pinterest pins from the core ideas. A content calendar that used to take a full team now takes me about three to five focused hours of creation time (depending on what we’re building and how many articles I’m writing that week) and my VA’s 20 hours of execution time. For real. I’ve tracked it.

But those numbers only tell you what it looks like AFTER the systems are built. They don’t tell you about the weeks I spent building a Pinterest keyword bank from scratch because nobody had done it for my niche. They don’t tell you about the guardrails document that started as a list of five things I hated about AI output and is now 98 documented patterns across eight categories. They don’t tell you I rebuilt my Claude project structure three times before it held context properly across sessions.

Every “look how efficient my AI workflow is” post on social media is showing you the after. I’m telling you the during was tedious, frustrating, and slow. I documented AI failure patterns for two years while the normal people were bingeing Netflix. That’s what built this.

The writing is where the real story is. I wasn’t writing blogs before AI. I would have never done it. My brain processes things too fast and has a hard time sitting down to focus long enough to get a long-form piece finished. I have hundreds of thousands of words written across years of work but no ability to sit down, focus, and put it all together into something publishable. Books I started and never finished. Entire frameworks mapped out in my head that never made it onto a page in any organized way.

When I realized what Claude could do, I knew I could finally start a blog. It started out slow. Claude had to learn my style, the editing process was long and tedious, and the early drafts needed heavy reworking. Now, 51 articles later, it takes no time at all. The system learned me. And I learned how to work with it. The thinking hasn’t been outsourced. The mechanical labor of getting words onto a page, formatting them for six different platforms, and scheduling them across a month? That part has (and good riddance).

Where It Breaks Down

AI still produces rubbish regularly. Even with my guardrails loaded, even with voice samples, and even with detailed context. Some days the output misses and I spend more time fixing it than I would have spent writing from scratch. (Last week Claude gave me a draft with four mirrored contrast phrases in a row. FOUR. After loading a document that specifically says never do that.)

The system works BECAUSE of the investment I made building it. Someone downloading a prompt template off the internet is not going to get these results. The “just use AI” crowd makes it sound like the tool does the work. The tool does what you’ve trained it to do. If you haven’t trained it, you get what everyone else gets.

Spiritual content, pastoral counsel, and anything that requires discernment about what God is doing in someone’s life, that’s a hard line. AI doesn’t touch it. I don’t even ask it to draft in that space. Some things require human presence and human responsibility, full stop.

And my VA still needs me. The systems reduce the number of decisions she has to make, but they don’t eliminate them entirely. She flags things, asks questions, and catches issues the system missed. The 20 hours works because she’s sharp and because the systems are well-built. If either of those factors wasn’t there, it would fall apart.

What It Costs (And What It Replaced)

I pay for Claude Pro at $100 a month, Gemini Pro at $32 a month (covers both me and my VA through our company Gmail accounts), and Higgsfield runs $150 to $250 depending on how many videos we’re producing that month. So the AI tools cost me roughly $280 to $380 a month. I’m also onboarding an autonomous AI agent right now that will hopefully help with a lot of the content production and enable us to put out even more. Getting her trained up is running $500 to $1,000 a month on its own. My VA’s compensation is on top of all of that.

Business still requires overhead in an AI economy. Anyone telling you differently isn’t doing things at scale.

But...compare that to what a content writer, social media manager, and graphic designer would cost, even part-time. We’re talking thousands per month minimum. I’ve written those checks before in previous businesses. The overhead wasn’t just financial. It was the meetings, the revisions, the miscommunication, and the management load that ate into the hours I was supposed to be spending on actual strategy.

That’s gone now. And I don’t miss it.

The Part That Actually Bothers Me

I see “I replaced my whole team with AI” content all over social media. Reels with clean desk setups and captions about making $10K a month with ChatGPT. And almost none of them are telling you what it actually took to get there.

Nobody talks about the months of building. The hours spent writing down every rule, every workflow, every pattern so that someone other than you can execute it. The failed systems you have to tear down and rebuild. The fact that AI without infrastructure produces the same mediocre content flooding every platform right now. Posting that content and calling it a business is like buying a hammer and calling yourself a contractor.

The structure took me months to build, test, break, and rebuild. I’m not complaining. I’d just rather you know what this actually requires going in than find out the hard way after you’ve wasted three months copying what some guy said worked in a 60-second reel.

What I Didn’t Expect This to Teach Me

Building AI systems forced me to look at how I ran my previous businesses. And what I saw wasn’t pretty.

I was a control freak. Everything had to be perfect, and that perfectionism slowed down my entire team from doing their jobs. I held things too tightly, revised too many times, and left so much money on the table because I couldn’t let go of the reins long enough to let other people execute. I’ve grown since then. A lot. But working with AI accelerated that growth in a way I didn’t anticipate.

Using AI requires you to loosen up. You have to accept that the first draft won’t be perfect, that the system will miss things, and that excellent is the standard now, not perfect. I’m working toward excellent every single day. And when I get on one of my perfectionist kicks and need to tear a draft apart, I don’t hurt Claude’s feelings when I tell him what he wrote is dumb. (LOL)

That might sound like a small thing. But for someone who used to agonize over every word and slow down an entire operation because of it, the ability to move fast, edit without guilt, and let good enough be good enough on Tuesday so that great can happen by Thursday? That changed how I work more than any tool ever could.

If You’re Building Something

If you’re running a small business, a side hustle alongside your full-time job, a ministry, or any kind of operation where you’re doing too many things yourself, AI can help. That part isn’t a question anymore. The real question is whether you’re willing to take the time and make the investment to build something that actually works with it.

I have capacity I haven’t had in years. Real capacity. Time with my family (we are heading to Cambodia for a four day vacation next week), time in Scripture, time to actually think about where this business is going instead of just grinding to keep it moving.

That’s what well-built AI systems buy you. Room to breathe.

---

The rest of this article is for paid subscribers. I’m breaking down the actual role-by-role system: what AI handles, what my VA handles, what only I can do, and the exact workflows connecting all three. If you want to see how this operates at the task level, this is it.