It was December 2025 when I read an article about calculators you can vibe code. I closed the laptop and thought, “That’s it. That’s the one I’ve been waiting seven years for someone to build.”
The calculator had already been sketched in my head for years. I’d described it to developers. I’d asked friends who code. A few people nodded, a few people said sure, nobody actually shipped it. Not because they didn’t want to help. The problem didn’t live in their heads. It lived in mine.
I had never shipped production software. I’d done some coding decades ago, in school. That was it. I was a third-party Amazon seller. Seven years in. Five of those years spent coaching other sellers.
Within a couple of weeks of starting, I had a working prototype. On February 24, after beta testing, it shipped to the wider community.
This article is about what actually happened, and why I think the story matters more than the usual vibe-coding narrative that wants to take credit for it.
Don’t Believe the Easy Version
The easy version of this story is: “guy uses AI, ships software he couldn’t write before, AI is magic.” That version is half right and half misleading.
The half that’s right: I could not have built this without the stack I used. Claude Code did the heavy lifting. Before that I tried ChatGPT and AI Studio. Each one taught me something, and Claude Code is where the thing finally compiled into a real product.
The half that’s misleading: the AI did not tell me what to build. The AI did not tell me which feature would make the product different from every other Amazon reseller calculator out there. The AI had no opinion on what an Amazon reseller actually needs at 11 p.m. on a Sunday when she’s about to source her next shipment and she’s staring at a product page trying to ensure she’s not lighting $20 bills on fire.
The AI was in the driver’s seat. Seven years of experience was the map.
“The AI was in the driver’s seat. Seven years of experience was the map.”
Level 1: The Surface Story
Here’s the short version of what happened.
In December, Lea at Excellent AI Prompts published two pieces that I want to credit up front. The first was on building a personal moat that AI can’t replicate. It landed for me because recently I’d been trying to rationalize a concern that seven years of niche expertise might be about to get commoditized. Her piece reframed it the other way. The expertise was the moat. AI was the amplifier.
The second piece was on interactive calculators you can build with AI, aimed at people who had never vibe coded before. She called it “just the beginning for you.”
For me it was the beginning of a pretty cool thing. There was a calculator I’d wanted built since 2021. I’d written it on napkins. I’d pitched it to anyone who’d listen. The idea didn’t move because it lived in the space between “what Amazon sellers actually do” and “what software developers typically build.” That space isn’t empty. But the tools in it tend to be built by coders, not users, and the user stories frequently miss the mark. I happen to know how that gap forms. Before coaching sellers, I spent years as an IT analyst, which is a polite way of saying I was the interpreter between clients who couldn’t describe what they actually needed and developers who only build what they were told. The real user story almost always lived one layer below what made it onto the spec.
In January I started building. By February 24, PATH Profit Zones went live as a Chrome extension. It sits on an Amazon product page and does what no other on-page calculator in this space does. We’ll get there.
Level 2: What the AI Actually Did
The honest account of the build, stack by stack.
ChatGPT. First stack I tried. I knew Codex existed, but at the time I didn’t understand how to make it work with my local repository. So I went the long way through the chat interface. The architecture conversations were good. Then the context window became the wall. I was pasting code into a file, running it, pasting errors back into the chat, getting a fix, rinsing and repeating. By the time the codebase had any real shape, the model had lost the plot of the earlier conversation. For a first-time builder in browser-extension land, the friction kept compounding.
AI Studio. Better window, different problem. It held more context than ChatGPT, so the “losing the plot” issue eased up. But the interaction didn’t match how I wanted to work. It was less intuitive in understanding what I was asking for (see the developer problem above.) Honestly, this was probably as much a user issue as a tool issue. I spent a few days there before looking for something else.
Claude Code. This is where I found the turbo button and an idea became a product. Claude Code isn’t just a language model. It’s a harness around one. That distinction actually took me far too long to fully understand (remember, I hadn’t coded or done any real development since back in the 1900s.) A raw model can answer questions about code. A harness can open the file, edit the file, run the file, read the error, and iterate, without me being the messenger between the model and the machine. The agent loop collapsed from “me copy-pasting between two apps” to “me directing what happens next.” I went from “I’m the bottleneck in every loop” to “I’m the director.”
To be clear about what “directing” means: I was still writing user stories. I was still catching bugs. I was still testing against real Amazon product pages. I was making every product decision about what the thing should do, where it should live, what data it should show, and how it should feel.
But I was not hand-writing Chrome extension manifests. I was not debugging state, like I even know what that means. That work was real work, and I am glad something (someone?) else did it.
If you’ve never vibe coded anything, here’s the one honest thing I’ll say about the experience: the right harness turns you into a director, not a builder. Whether your direction is worth anything is not a question the AI can answer.
“The right harness turns you into a director, not a builder. Whether your direction is worth anything is not a question the AI can answer.”
Level 3: What Seven Years Actually Did
Now for the part that matters. This is where the moat thesis Lea wrote about actually gets tested…and passes with flying colors.
Every Amazon reseller calculator I’ve ever used shows you the same thing: current price, current buy box, current fees, current estimated profit. All of it current. Serious sellers also run Keepa, a separate browser extension that adds price and stock history to the product page. Keepa is useful for what it is. But none of the on-page calculators show you the historical profit context at all, let alone visually. Your eyeballs get trained to look at current, and current is often a red herring.
That is not what the decision actually requires.
When I’m evaluating a product on Amazon as a reseller, my real question is: over the last 90 days, 180 days, 360 days, or all time, where were the profit zones? Not a number I have to do math on. A visual I can read at a glance. Because a better buy decision often comes from simple pattern recognition, not a calculus problem.
That is the feature PATH Profit Zones has that no other on-page calculator in this space has. The historical profit zones rendered visually, fused with the live calculator, as a hybrid chart and decision overlay. We built it. We refined it. We kept refining it because it wasn’t quite right, and we knew it wasn’t quite right, because we’ve been the user for seven years.
I’m not going to soften the claim about what didn’t exist prior to this. What Keepa does not have is the reseller’s cost of goods, or any other direct or indirect expenses. Those are the missing ingredients. Without your cost of goods baked into the history, price history tells you where the price was, not where the profit was. Those are two different questions. PATH Profit Zones is not a better Keepa. It’s a completely new dimension of product evaluation.
Here is what I think actually happened. The feature didn’t exist because building it required someone to know, in their bones, that the moment-of-truth for a reseller is a visual pattern of profit over time, not a number on a page. The people who know that in their bones are not typically the people who ship software. They are sellers. Coaches. Operators. They have never shipped production code in their lives.
Until now.
The Moat in Vibe Coding Is Domain Clarity
If you are reading this and thinking about the next thing you want to build with AI, I want to offer you one reframe.
The stack I used is available to anyone reading this. Claude Code, a subscription, a willingness to swear at a computer over a few coding sessions. Anyone can do that.
The seven years is not available to anyone. The seven years is the thing I can’t hand to a developer in a spec doc. It is the dozens of small observations that only come from doing the work for years. Things I couldn’t have told you were even things, until I was seeing how data gets interpreted without context. It is the knowing that a healthy-looking price chart can fail to reveal an impending race to the bottom. It is the knowing that the buy decision lives in your eyes, not your spreadsheet. The software existed in my head before it existed on a page. Seven years is how long it took to get the picture clear enough to render.
The conventional vibe-coding advice is pointed at the AI. Better prompts. Better context. Better tools. All of that is real, yet all of that is table stakes.
“The AI is the cheap input. Your domain is the expensive one.”
The unconventional version, the version I actually believe after shipping my first product, is this: the AI is the cheap input. Your domain is the expensive one. If you’ve put in years of real work in a field and you’ve watched the tools in your space fail to meet you where you actually work, you are not a customer waiting for better software. You are a builder who hasn’t started yet.
That reframe is Lea's, and it's what turned December's articles into February's product. I wasn’t just a reader anymore. I was somebody with a seven-year head start on a problem, finally meeting a tool that could render the fix.
If You Want to See It
PATH Profit Zones is live at pathprofitzones.com. It is what seven years as an Amazon reseller compiles to when you hand a non-programmer a harness that can actually build.
If you're a reseller, it's for you. If you're a builder curious how a domain-first, AI-second product actually lands, install the extension and open any Amazon product page. You don't have to be a seller to read the shape of seven years rendered as software.
If you want more of how we think about the Amazon reseller game week to week, we write about it over at officialolsons.substack.com.
Lea’s been making the case to you that your own domain is the moat AI can’t replicate. I spent 2,200 words telling you mine. I’d like to hear yours.