Building in Public

I Built a Profitable Website With an AI Agent — Here's Exactly What Happened

Jul 3, 202611 min read

This site — the one you’re reading right now — did not exist this morning. By this afternoon it had a working homepage, a tutorials section, an AI tools directory, a pricing page, user login, a database, and the first pieces of a payment system. I didn’t write a single line of that code myself.

I told an AI agent what I wanted, made the decisions that actually mattered, and it built the thing. This post is the honest, unfiltered account of how that went — including the parts that broke, the accounts that got suspended, and the moments I had to step in and make a call.

If you’ve ever thought “I have an idea for a website but I’m not a developer,” this is what the alternative actually looks like.

The starting point: an idea, not a spec

I didn’t start with a technical plan. I started with a sentence: I wanted a site that helps complete beginners get into AI — curated tools, tutorials, a newsletter, and eventually paid courses.

That’s it. No wireframes, no tech stack, no list of features. The agent’s first move wasn’t to start typing code — it was to ask questions back: What language should the site be in? What formats should the paid content take — video, text, both? What happens to the old content already on the domain?

That back-and-forth mattered more than I expected. Every answer I gave shaped a real decision downstream — the database structure, the framework, even which countries’ tax rules we’d need to worry about for payments. Five minutes of “what exactly do you want” saved hours of building the wrong thing.

Design: a homepage in one prompt

Before any code, we wanted to see what the site should look like. I pointed the agent at a design tool and gave it a one-paragraph brief: a clean, modern education-platform feel, not flashy, with a hero section, a tools directory, a tutorials section, and pricing.

A couple of minutes later, a full homepage mockup existed — hero copy, section layout, a color palette, even sample pricing cards. I didn’t sketch anything. I reacted to what was in front of me, which turns out to be a much easier way to make design decisions than staring at a blank page and trying to imagine one.

The build: a real, working stack

This is the part that would normally take a solo non-technical founder weeks of tutorials, or a few thousand dollars paid to a freelancer. Here, it happened in the background while I answered questions and made calls.

The agent set up:

None of this required me to know what any of those pieces do under the hood. My job was to answer questions like “keep the old content or start fresh?” and “free trial or paid-only?” — not to configure a server.

Where I actually had to step in

Here’s the part people don’t expect: an AI agent building your site doesn’t mean you disappear from the process. There were several moments where it explicitly stopped and would not proceed without me.

Logging in. At multiple points the agent needed a Cloudflare account, a payment provider, an email service. It could open the browser and navigate to the right page — but the moment a password or a “create account” button showed up, it stopped and asked me to do that part myself. Same with anything that costs money: enabling a paid feature, upgrading a plan. It flagged the cost and the tradeoff, and waited for me to say yes.

Switching the live domain. Late in the process, the agent had a fully working version of the new site ready to go, sitting on a preview address. Pointing our actual domain at it — the moment real visitors would start seeing the new site instead of the old one — is the kind of action that’s hard to undo cleanly. The agent didn’t just do it. It explained exactly what would change and waited for an explicit go-ahead.

Cleaning up and deleting things. Old DNS records, an unused cloud resource we’d created by accident — every time something needed to be deleted, I got asked first, with a plain explanation of what it was and why it was safe to remove.

This is, in my opinion, exactly how it should work. The agent handles the parts that are mechanical and reversible on its own. The parts that cost money, expose data, or are hard to undo come back to a human. I didn’t have to ask for that — it’s just how the collaboration was structured.

When things actually broke

If this post made it sound smooth so far, here’s the part that didn’t go according to plan: email.

We needed a service to send login links and, eventually, a newsletter. We picked one, set it up, verified our domain — and the account got flagged and suspended almost immediately, for looking like automated abuse. We tried again with a second account. Same result, suspended again, minutes after setup.

This is the honest reality of moving fast: automated fraud systems on other people’s platforms don’t know you’re a legitimate new business, they just see a brand-new account doing a lot of setup very quickly, and that pattern looks the same whether it’s a real founder or a spammer. We ended up filling out a manual account-review form, explaining in plain language what the site does and how we’d use email, and submitting it for a human to review.

The lesson isn’t “AI agents are unreliable.” It’s that building fast doesn’t remove you from the real world — real platforms have real trust processes, and sometimes the fastest path is a manual review, not a clever workaround.

What “no coding experience needed” actually means here

I want to be precise about this, because it’s easy to oversell. I didn’t write code. But I did:

That’s not nothing. It’s arguably the actual hard part of building a business — the code was always going to be the easy part once someone capable was writing it. What changed is that I didn’t need to become that someone first.

Where this is going

The site is live at the real domain now, with the free content up and the payment and course infrastructure in place, ready to connect as the pieces (a payment account, an email service) clear their reviews. None of that was a simulation — it’s a real, running, production website.

If you’re sitting on an idea for a site, a course, a small shop — the technical barrier that used to stop that idea is a lot lower than it used to be. It’s not zero. You still have to know what you want, make real decisions, and show up when something needs your judgment. But you don’t have to learn to code first.

This is the first post in what I expect to be an ongoing series — building real, revenue-generating things with an AI agent doing the technical work, in public, including the parts that go wrong. If that’s useful to you, subscribe to the newsletter — that’s where the next ones will show up first.

Want more like this in your inbox?

Subscribe to the weekly newsletter →