I Run My Store Like a Toyota Factory. Here’s Why It Actually Works.

Genius Loci · Field Notes

Somewhere around product number four hundred, I stopped thinking about Genius Loci as a website, I knew it was turning into a factory. 

That probably sounds excessive for a Shopify store, but once you’re dealing with thousands of products, landing pages, suppliers, product data, photos, pricing, SEO, and all the other little things that have to be correct, it stops being a normal website project. It becomes a production problem.

You can’t sit and handwrite four thousand product pages. No thanks.

The obvious answer is automation. Let AI do it.

The problem is that AI can produce absolute BS at an incredible rate.

If the system gets something wrong, it doesn’t just make one bad page. It can make five hundred bad pages before you even realize what happened. Wrong fitment, made-up benefits, fake urgency, invented reviews, broken links—the whole catalog can become confidently incorrect overnight. This can quickly create trust issues. 

I would genuinely rather publish one honest page than one hundred pages that quietly lie to people just to get a click. Toyota taught me to value and honor the customer and that is what I will do. 

TPS is the solution. 

Why I started thinking about lean production

The Toyota Production System was designed around a problem that sounds remarkably similar to what I was dealing with:

How do you make a large number of products efficiently without stopping constantly to inspect or shipping garbage? 

Most of what I saw in the world of “AI content at scale” chose one extreme or the other.

Either everything required so much manual review that automation barely saved any time, or the system generated thousands of pages and simply hoped they were good enough all while taking my money. That is "muda" - waste. 

With TPS quality isn't treated as something you check after production. It is built directly into the production process.

A few Toyota concepts immediately made sense to me for what I am building: 

• Jidoka: Quality gets checked at the station where the work happens, not at the very end.

• Andon: If something is wrong, the line stops. The problem doesn’t get quietly passed to the next station.

• Genchi Genbutsu: Go and see the real thing for yourself. Don’t make decisions based entirely on reports and summaries.

• Heijunka: Produce in small, steady batches instead of giant, chaotic runs.

• Kaizen: Keep making small improvements instead of constantly tearing everything down and starting over.

Obviously, I’m not building Camrys.

But the underlying problem is almost identical: make a lot of something, keep the process moving, and don’t let quality disappear just because the numbers get bigger.

Every product goes through the same line

The first important thing I built wasn’t actually the page generator, it was the production line around it.

Every product has to follow the same basic path. The system finds an opportunity, decides whether it is actually worth pursuing, checks the real product data, creates a draft, inspects the result, and only then allows it to move toward publication.

Nothing is supposed to teleport from “this might be interesting” straight to “this is now live on the store.”

If every product follows a completely different process, then every mistake becomes its own little mystery. You end up digging through one-off prompts, random scripts, old conversations, and automation logic trying to figure out where something went wrong.

I wanted one production line I could understand, improve, and run four thousand times. That doesn’t mean every product page has to look identical. It means the truth-checking, decision-making, and quality-control process stays consistent even when the final product is different.

The system is not allowed to lie

The most important part of the entire line is the quality gate.

This is where the Toyota idea of Jidoka becomes very literal. If the system detects a serious problem, it is supposed to stop.Not “make a note and keep going.”Stop.

A page should not invent a five-star review that doesn’t exist. It should not create a fake countdown timer. It should not claim a warranty, guarantee, fitment, discount, or benefit that nobody actually confirmed. Through my use of LLMs it has become very obvious that they love to take creative freedom all of the time. 

The goal was to build in quality into the processes so that would not allow this junk data to pass into the next step of the process. 

If a page is missing something important or starts making things up, I want it blocked with a clear explanation of what went wrong. That is infinitely more useful than a beautiful page that is quietly dishonest.

Actually looking at the page matters

This is where Genchi Genbutsu, or “go and see,” has become a real part of how I work.

I’ve spent a lot of time building things with AI, modifying software, working on Shopify themes, creating products, making content, and trying to connect all of these systems together. One lesson keeps coming back:

Never completely trust anything, go and see for yourself. 

I’ve had plenty of moments where a tool told me something was fixed. The code looked right, the report sounded convincing and everything was supposedly working.

Then I opened the actual page on my phone and it was still broken. Without Genchi Genbutsu there is no proof... and the TPS framework would fail. 

Small batches are faster than one giant mess

When people hear “AI automation,” they assume the goal should be to process the entire catalog overnight. Technically, you could try.

My system intentionally works in small batches. It takes on a manageable number of products, runs them through the line, checks the results, and then does another batch.

That’s the Heijunka part of the system: steady, leveled production instead of one enormous run followed by three days of figuring out what exploded. Large production runs hide problems. 

There is also a very practical reason for this: I am not running a data center.

A lot of this is designed to run on my own hardware, with local models and relatively normal equipment. I would rather build something that can run reliably every day than something that looks incredible in a demonstration and cooks itself the first time I give it real work. The "Jarvis" style Ai dashboards that the vibecode LARPers are posting about now are the antithesis of what I made. Built in quality is key, fancy things for social media are pointless. 

Tools that exceed your real capacity are not actually useful.  

Kaizen 

I have spent months finding small problems, fixing them, and feeding those fixes back into the system. 

Continuous improvement is a philosophy of being, in a sense, so why not bake it into my self-learning, self-improving virtual production line? 

Why I actually care about this

I’m not building Genius Loci just to see how many pages I can generate.

I’m trying to build something that combines the different things I actually know how to do: product development, automotive knowledge, design, marketing, content, e-commerce, software, and physical manufacturing.

The automation is there to help me do more with those skills. It is not supposed to replace judgment or turn the store into a giant pile of generated words.

I want the system to run largely on its own. I want it to find opportunities, create products and pages, learn from what performs, and keep the production loop moving without needing me to personally push every button.

I am still very far from what I would consider "complete". There is still much to improve in the UI/ UX experience as well as tons of back end functions that need to be programmed in. 

Toyota’s entire production system is built around the idea that you should plan for the long term and implement changes methodically and I intend to emulate that as well. 

Genius Loci is turning into much more than a just a store. Right now its an agentic problem -solution testing system that surfaces solutions in the form of products based on social and market signals. A self improving loop that allows for the organic discovery of genuinely useful physical artifacts was the goal of the GeniusLociSuperOS.

This is simply one faculty of what I am building here. Stay tuned to see what I come up with next as I create the spirit of the place.