A number of years ago, while I was still running Kardish, I went to the people behind our fancy, state of the art ERP and point-of-sale systems with a list of things I knew the business needed — smarter inventory replenishment, logistics optimization, logic that understood how our products actually moved through nine stores and a warehouse. The answer came back in two parts. First: that's not possible with your system. Second: if you really want it, it's custom software — integrations, consultants, programmers, project managers — somewhere around $80,000.
So I built it myself, with AI.
Not a prototype. Not a demo. Working tools, with real replenishment and inventory optimization logic built in, that produced consistent results we ran the business on. I built them alone, at a fraction of a fraction of that quote, on top of — and running parallel to — systems whose own vendors had told me it couldn't be done.
That's the experience I want to describe in this piece — because most of what's written about AI comes from technologists, and most of it is about the tools. This is about what happens when the tools land in the hands of someone who has spent twenty years inside the operations of a real business.
What the stack actually looks like
I was an early adopter, and I've stayed hands-on daily ever since. My working stack isn't one product — it's several chatbots (they have different strengths, and I use them against each other), image and video generation, audio and speech tools, coding tools, and agents I've set up to handle multi-step work.
A typical day runs through all of it: research, drafting, analysis, building. Marketing work that used to take a team. Financial modelling I used to need help with. Code I would once have outsourced, built and shipped in an afternoon.
The piece I find most valuable, though, isn't any single tool — it's what I've been able to encode. Over twenty years I accumulated a stack of strategic and operating frameworks: systems we used at Kardish, structures I was taught by external consultants, theory from a long shelf of business books. I've built that thinking into reusable tools I can plug any business or idea into — and get meaningful, customized strategic work out the other side. I've run full business plans for new ideas this way. The frameworks were always good. They were just trapped in binders and in my head. AI made them executable.
The video my mother believed
Here's one from the early days that still makes me laugh. I wanted to see whether I could produce a sample marketing video — this was when video generation was crude and there was no editing inside the tools at all. So I chained them together: chatbots to write the scripts, then chatbots again to write image prompts optimized for the specific image models I was using, then image generation to create healthy, wellness-minded people, then video tools to bring those stills to life in short lifestyle clips. I stitched the clips together and laid AI-generated background music underneath.
The result looked real enough that I sent it to my parents. My mom watched it and said of one of the women, "Oh, I really like her" — not knowing there was no her.
The point isn't the trick. The point is that a retailer in Ottawa produced agency-grade creative by understanding the work well enough to break it into pieces and route each piece to the right tool. That's an operating skill, not a technical one. It's supply chain thinking, applied to content.
What AI gets wrong — and why that's where the value is
I want to be honest about the other side, because this is what separates real use from hype: AI gets things wrong. In the early days of coding with it, it got a lot wrong.
What I learned is that the work happens in a loop. The code breaks; you hunt down the error; you copy the failing piece and the logs and feed them back in; you let the model help diagnose its own mistakes. These tools are remarkably good at reading logs — better than they were at writing the code in the first place, back then. Sometimes the loop converged quickly. Sometimes it got stuck, and there were long nights where I had to dig through the code myself and figure it out. I learned the basics of programming that way — not from a course, but from refusing to let a broken tool stay broken.
That loop is the whole game, and it's why operating experience matters more with AI, not less. The model produces output with total confidence whether it's right or wrong. Someone still has to know what right looks like. Twenty years of running a business is, among other things, twenty years of looking at a confident answer — from a vendor, a consultant, a spreadsheet — and knowing when it doesn't smell right. That judgment doesn't get automated. It gets multiplied.
The combination
I think about it this way: AI didn't change what I bring to a business. It changed the output volume of one person who brings it. An operator who knows what needs to be built, paired with tools that compress the building — that's a fundamentally different proposition than either one alone.
The inventory optimization system the vendors said was impossible is my favourite proof, but the website you're reading this on is proof too. I built it with AI. Judgment decided what it should be; the tools did the labour; the loop fixed what was wrong.
That's what AI looks like in the hands of an operator who's done the work.