Hey digital adventurers... okay so I’m about to do something either really brilliant or completely insane and I honestly can’t tell which one yet but that’s never stopped me before right
Remember when I wrote about my AI co-CEO experiment like 2.5 years ago? That whole wild ride where I let ChatGPT help run an actual business? Yeah well... I’ve been thinking about that a lot lately and I realized something
That was just the beginning. We were barely scratching the surface of what’s actually possible now.
Because here’s the thing... back then AI was impressive but limited. Now? Now we’ve got Claude with MCP integration, we’ve got vibe coding platforms that actually work, we’ve got automation tools that can orchestrate complex workflows, we’ve got AI that can actually THINK through problems rather than just pattern match
And I keep asking myself... what if we took all of this and built something genuinely autonomous? Not like “AI helps with tasks” autonomous but like “AI runs the entire operation with minimal human intervention” autonomous
So I’m doing it. Starting now. Building in public. Again.
What I’m Actually Building
Okay so here’s the concept and I’m still figuring out details as I go but that’s part of the fun right
I’m creating a complete e-commerce system where AI handles... basically everything. Product research. Supplier coordination. Inventory management. Customer service. Marketing campaigns. Analytics. Decision making. The whole stack.
But here’s what’s really important to understand... this isn’t just about managing an existing business. It’s about creating a system that can both BUILD and RUN a business. That’s the ultimate goal here.
My role right now? I’m doing the complex setup work myself because honestly that’s the hardest part. Building the infrastructure, connecting the systems, creating the decision frameworks, establishing the automation flows. This is strategic work that requires human judgment and understanding of what we’re trying to achieve.
But once that foundation is in place... once the system is actually running... my role shifts to pure strategic oversight. Interventions only when truly necessary. The AI should handle day-to-day operations, make routine decisions, and only prompt me when it encounters something that genuinely needs human input.
Think of it like this... I’m building the blueprint now so the AI can operate the business later. And that blueprint itself becomes valuable because if this works... it’s replicable. It becomes a framework for autonomous digital businesses.
When I wrote about building internal digital solutions fast, I talked about how modern tools let non-developers create real solutions quickly. This takes that concept to its logical extreme... what if the AI does the building AND the operating?
Why Am I Really Doing This
Look I need to be completely honest about my motivations here because this isn’t just a fun experiment...
This first project is a blueprint. But it’s also a test of a much bigger question... can we push AI to the point where we can actually TRUST it to do things in a company based on data and context, only asking humans when it’s genuinely needed?
Because if I succeed here... if I can get AI to manage and develop a small business autonomously... that’s not the end goal. It’s the BEGINNING. It’s proof of concept for something much larger.
I already did this once before actually. That AI co-CEO project I mentioned? We managed to get nearly 90% autonomous operations driven by AI. But that was different... that was me spending tons of time in AI chats, constantly prompting and guiding and supervising. It worked but it wasn’t sustainable.
This time I’m asking a harder question... can the system run WITHOUT me constantly being in the loop? Can it make decisions, take actions, and only interrupt me when human judgment is actually required?
If I can prove that works... even at a small scale... then we’ve validated something really important about where AI capabilities are actually at right now. Not theoretical capabilities. Not marketing promises. Real practical autonomous operations.
And that opens up all kinds of possibilities for what comes next.
The Technical Stack I’m Using
Okay let’s get specific about how I’m actually building this because details matter right
Core AI Brain: Claude. Both through the chat interface and via API. I’ve been using Claude extensively and wrote about setting up Claude Desktop with MCP... it’s genuinely the most capable AI for complex reasoning and decision-making right now
Automation Orchestration: I’m choosing between Make.com and Zapier for this. Haven’t decided yet... I need to evaluate which one gives me better integration options and cost-effectiveness for what I’m building. But the key point is this layer gives Claude actual HANDS. It can trigger actions, respond to events, run workflows without me manually prompting it
E-commerce Platform: Shopify. This is deliberate because Shopify is incredibly open by design. Tons of APIs, webhooks, integrations. It’s perfect for connecting AI systems to actual commerce operations
Data Layer: Here’s where I’m borrowing from my previous experiment... one central Excel file that stores all crucial data. Simple, accessible, easy to update. The AI can read from it and write to it, creating this self-improving knowledge base for customer service patterns, marketing insights, product performance
Automation Goals: Right now I’m using tools that can be connected immediately. But my ultimate goal? Building my own automation server. Full control over the infrastructure, no reliance on third-party platforms, complete customization of how everything works together
The philosophy is start with what works NOW, but design with the future state in mind. Get it working with existing tools, then gradually replace pieces with custom solutions as I understand what’s actually needed.
What’s Different From Last Time
So yeah I mentioned I did a version of this before with nearly 90% autonomous operations... but here’s what I learned from that experience and what I’m doing differently now
Less Chat Time: The old approach required me to be constantly in AI chats. Prompting, guiding, reviewing, adjusting. It worked but honestly it was exhausting. This time the system should operate in the background. I should only see notifications when something needs my attention
Event-Driven Instead of Prompt-Driven: Rather than me asking the AI to do things, events trigger the AI. A customer sends a message... the AI handles it. Sales data hits a threshold... the AI adjusts pricing. Inventory gets low... the AI initiates reordering. All automatic, all autonomous
Self-Improving Systems: I’m taking concepts from the previous project like self-improving customer service, marketing, and product management... but implementing them as closed-loop systems. The AI doesn’t just respond to situations... it learns from outcomes and adjusts its approach over time
Clear Escalation Rules: This is critical... the system needs to know WHEN to ask for help. Not everything requires human judgment, but some things definitely do. Building those decision trees properly is probably the hardest part of this whole project
Better Data Integration: Everything flows through that central knowledge base. The AI always has context. It knows what’s happened before, what worked, what didn’t. That historical data informs current decisions without me needing to provide context every time
It’s the difference between having an assistant who constantly asks you questions versus having a manager who only comes to you with decisions that actually require your input. That’s what I’m aiming for here.
Nothing Can Go Wrong Actually
Wait that sounds wrong... let me clarify what I mean
This is a learning experiment. The bar for success is intentionally LOW. I’m calling this successful if I manage to automate at least 70% of operations. Just 70 percent. That’s it.
Why such a low bar? Because the real value isn’t in hitting 100% automation. The real value is in the insights I’ll gain along the way...
What CAN be automated safely versus what SHOULD stay human
Where AI makes better decisions than humans versus where it fails
What types of decisions require what types of data and context
How to structure escalation protocols that actually work
What the real costs are compared to traditional operations
Where the unexpected bottlenecks and challenges emerge
Even if this project only hits 50% automation... even if parts of it fail completely... I’m getting data that’s valuable for everything I build after this. And honestly that’s more valuable than a perfectly automated business that I don’t really understand.
When I wrote about when automation is worth it, I emphasized that sometimes the learning is more valuable than the outcome. This is one of those times.
Plus here’s the thing... it’s my project, my money, my rules. I’m not answerable to investors or a board. If something goes sideways, I adjust and learn from it. That freedom to experiment without catastrophic consequences is actually pretty rare and I’m not taking it for granted.
How Often You’ll Hear From Me
I’m committing to documenting this... but I’m being realistic about cadence
Updates will happen when I make meaningful progress or encounter meaningful problems. Sometimes that might be every other day when things are moving fast. Other times it might be once every two weeks when I’m deep in implementation work.
The more I get built, the more I’ll have to share. Early stages might be quieter as I’m setting up infrastructure. Later stages should be more dynamic as the system starts actually operating and generating real results and real failures to analyze.
But I’m NOT going to do fake updates just to maintain a schedule. When I write, it’ll be because I have something worth sharing. Real progress, real insights, real challenges. Not just “here’s what I’m thinking about maybe doing someday.”
I know from my remote work experience that consistent communication matters, but quality beats frequency every time. So expect irregular but valuable updates rather than regular but shallow ones.
Why You Should Actually Subscribe
Okay real talk for a second about why following this is worth your time...
Most people experimenting with AI either do it quietly behind closed doors OR they only share the successes and polish everything to look perfect. I’m doing neither of those things.
This is genuinely building in public. The failures, the costs, the unexpected problems, the “oh crap I didn’t think about that” moments... all of it gets documented. And THAT is valuable because it’s the real messy truth about what works and what doesn’t when you’re pushing AI capabilities to their limits.
Here’s what subscribing to Digital Thoughts gets you:
Most updates on this project will be completely FREE. I’m not gatekeeping the learning behind a paywall because honestly the more people who understand what’s possible with AI autonomous operations, the better.
But some specific deep dives... technical implementation details, full decision framework documentation, detailed cost analyses... those will be for paid subscribers. Think of it as supporting the experiment while getting access to the really granular insights.
And look... I’m a micro-blogger doing this independently. No VC funding, no corporate backing, just me testing ideas and documenting what happens. Your paid subscription literally enables me to spend more time on experiments like this instead of client work. That’s the honest economics of it.
When I shared my digital toolbox and MCP setup guides, hundreds of people told me those posts saved them hours or even days of trial and error. That’s the value of real documented experiments. And this project is going to generate a LOT more of that kind of practical, tested knowledge.
Plus... you’re not just subscribing to read about this project. You’re subscribing to someone who’s consistently exploring the cutting edge of digital tools, AI capabilities, automation possibilities, and practical implementation. I just spent $500 testing vibe coding platforms so readers would have real data instead of marketing hype. That’s the kind of content you’re signing up for.
Whether you choose free or paid doesn’t matter to me... what matters is you’re here for the journey. You’re learning alongside me. You’re part of this conversation about what’s actually possible when we stop theorizing and start building.
The value proposition is simple: Subscribe for free and get most updates plus all the regular Digital Thoughts content. Upgrade to paid when you want the really detailed technical breakdowns and want to support more experiments like this one. Either way, you’re getting access to real-world data about AI capabilities that most people are keeping proprietary.
You can literally watch me build this in real-time and decide at any point if the paid content is worth it. No long-term commitment, no pressure, just honest documentation of what works and what doesn’t.
The First Steps Start Now
I’m not waiting to begin... setup work is already underway
I’m connecting Claude to the core systems through API. I’m building the initial automation workflows in Make.com or Zapier. I’m setting up that central data file with the crucial business information. I’m defining the initial decision frameworks for when AI acts autonomously versus when it escalates to me.
The Shopify store is getting configured with all the necessary integrations. Customer service workflows are being mapped. Product sourcing automation is getting designed. Marketing campaign frameworks are being established.
This isn’t vaporware or “coming soon”... this is actively being built right now as you’re reading this. And every meaningful step forward will be documented and shared.
Next specific milestones I’m working toward...
Getting the core product sourcing automation operational with AI selecting products based on market data
Implementing autonomous customer service that handles inquiries and only escalates genuine edge cases
Building the inventory management system that automatically triggers reordering based on sales velocity
Testing the marketing automation to see if AI can create and execute campaigns without constant human oversight
Each of these is its own sub-project with its own challenges and learning opportunities.
Join This Experiment
So yeah... I’m building an AI-orchestrated e-commerce system that can both create and run a digital business with minimal human intervention. Starting from concepts I validated years ago but pushing much further with better technology and clearer goals.
This is happening whether anyone follows along or not because I need to know what’s actually possible... but it’s WAY more valuable as a shared learning experience. Your questions shape what I document. Your challenges make me think harder about assumptions. Your insights from your own experiments add context I wouldn’t have otherwise.
Subscribe to Digital Thoughts and you’re not just reading about this... you’re part of shaping what gets built and how it gets documented. Free subscription gets you most updates and all the regular content. Paid subscription gets you the detailed technical deep dives and directly supports more experiments like this.
Either way... welcome to the experiment. Let’s see what happens when we stop wondering what AI can do and actually build systems that test those limits for real.
Drop a comment below... what aspect of this are you most curious about? What concerns do you have? What do you want me to focus on documenting as I build this? I read everything and your input genuinely shapes how I approach this work.
And if you’ve tried anything similar or have insights from your own AI experiments... please share. The more we learn from each other the faster we all figure out what actually works in practice versus what just sounds good in theory.
Here we go...
PS. How do you rate today’s email? Leave a comment or “❤️” if you liked the article - I always value your comments and insights, and it also gives me a better position in the Substack network.
This article comes at the perfect time; I've been wondering what fully autonomous AI could really orchestrate, and I'm so curious to see you build this out.