Brett Malinowski handed an AI a virtual card with $500 on it and one instruction: make money selling a physical product, as fast as possible.
37 hours later, the first sale landed. $56.
Seven days later, the store had done $745.25 in revenue selling a hardcover book that Claude picked, designed, priced, and marketed. Brett never touched a single copy.
I've seen a lot of "AI runs my business" videos, and most of them are demos dressed up as results. This one is different because Brett has receipts and 10 years of ecommerce scar tissue ($5M+ in sales across a dozen businesses), and he still got surprised. He says he's never once been profitable on day one of running ads. His ai ecommerce business was profitable in under 3 hours.
Here's the full experiment, with every number he showed. Video first, my notes below.
The rules: $500, and the AI makes every decision
Brett set three constraints before starting the clock.
→ AI makes every decision. Product, pricing, positioning, all of it.
→ Every piece of content has to be AI-generated.
→ Total budget: $500, loaded on a virtual card the agent could spend from.
The stack was the Claude desktop app plus Whop's CLI. Claude ran multiple agents in different roles (product research, content, ads), and the CLI handled the website, payments, and ad campaigns straight from the terminal.
The challenge wasn't "can AI do tasks." It was whether AI is actually smart enough to run a profitable ecommerce store on its own, or whether humans still have an edge.
The answer turned out to be both, which is what makes this worth studying.
The product Claude picked (nobody saw this coming)
Brett's brief to Claude was smart, and it's the same filter I'd give any founder picking a first product:
→ Demand has to already exist. No inventing a new garlic peeler and spending $500 proving nobody wants it.
→ No inventory upfront.
→ There has to be a real angle to compete on.
Claude came back with the Book of Enoch.
Not a gadget or a supplement. An ancient text that got cut from the standard Bible, sitting in the public domain, while videos about it and the Ethiopian Bible pull millions of views right now.
The logic is clean. Anyone can read the text online for free, so the product isn't access to information. It's the object: a beautiful hardcover sitting on your bookshelf. Attention already exists, the content costs nothing to license, and the only thing being sold is the physical format.
They named the brand Enox Archive, generated a logo and banner, and Claude designed a cover in a dark Oxford red to feel like something you'd find in the back of an old library. Brett admits it: after a few iterations, it actually looked good.
The unit economics, before a single ad ran
For fulfillment, the agent found Lulu.com, a print-on-demand supplier with an API. A copy gets printed when someone orders and ships directly to the customer. No inventory, no garage full of boxes.
The math Claude worked from:
→ ~$15 to print one copy
→ ~$7 shipping
→ $22 hard cost per order
→ Estimated ~$20 to acquire a customer with paid ads
At a $30 price, that's lighting money on fire. So Claude priced the book at $49.95 plus $6.95 shipping, leaving room for ads, fulfillment, and actual profit.
This is the part most first-time founders skip. They pick a price that feels nice and find out three months later that every sale loses $4. The agent did the margin math on day one, before the store existed.
First sale in 37 hours, profitable on day one
Day two was the build. Brett had Claude turn everything into a PRD, then used Claude Code to spin up the business from that doc: product created, payment links live, a landing page built from a physical-product template and iterated for a few hours.
For ads, an agent scraped competitors' Meta ad libraries and wrote up every tactic, offer, and positioning angle they use on cold traffic. Then it generated 50 ad creatives from a PRD and set up the Meta campaign through the CLI. $50 a day, targeting the four major English-speaking countries.
Brett went home with a fried brain, had dinner with his wife, and checked the ad account before bed.
→ Ads live for under 3 hours
→ $4.68 spent
→ One sale: $56
He let it run a full day to make sure it wasn't a fluke. After 24 hours of ads: $232.15 gross revenue, $58.55 in ad spend, $119 in printing and shipping. $54.60 profit.
Brett's comparison: that's what a 4-hour shift at his high school pizza shop paid. AI did it in 24 hours while he slept.
Then the organic channel humbled everyone
Paid ads worked almost instantly. Organic was a completely different story, and honestly this section is the most useful part of the video.
From day one, Claude had set up an X account for Enox Archive with three scheduled tasks: 5 posts a day, 10 replies to popular posts, 2 quote tweets. A fully automated theme page. (This kind of always-on setup is the same pattern behind most business automations: agents running on schedules instead of a human remembering to post.)
After 3 days: most posts under 5 views. The cold start problem, in full effect.
So Claude proposed a workaround I love. Instead of waiting months for the algorithm, pay an established page in the niche to repost the content. The agent DMed 20 similar accounts asking about paid reposts. Three replied, one was relevant, they wanted $25 a post. Brett negotiated 5 reposts for $100.
The reposts worked for traffic: one crossed 1,000 views, a reply hit 8,000, and the account pulled over 10,000 impressions and 100+ visitors to the site.
Sales: zero.
Which made no sense, because 100 visitors from ads had produced three sales the day before.
The funnel audit that saved the experiment
On day five, Brett asked Claude to analyze the funnel. It came back with two problems: the price was high for a cold social audience, and a brand-new account with under 100 followers selling a $50 book has a trust problem.
The fixes, all shipped in a day:
→ Price dropped to $39.99 with free shipping (organic clicks are free, so margin can flex)
→ A free audiobook version generated in ElevenLabs, bundled as a bonus
→ Upsells added: a two-book bundle, an "Angel Index," and another lost text. People who buy books buy more books.
Day six: two more reposts, 5,000+ views on one, over 20,000 views from replies alone, 32 new followers, another 100 visitors.
Still zero sales. Brett was about to lose his mind.
So he finally did the thing every founder should do weekly: he went through his own checkout. It was broken. The multiple product bundles made the embedded checkout modal take over 7 seconds to load, so buyers clicked and stared at what looked like an empty page.
Claude preloaded everything on first visit. Fixed.
Two burned reposts and a day of rage, caused by a bug a 30-second test would have caught. Test your funnel. Every time you change it.
Day 7: the $80 order and the final tally
One repost left. Brett put the website link directly in the post and couldn't watch.
Two hours later: the first organic sale. And not a small one. The customer bought the two-book bundle plus the Angel Index order bump. An $80 order.
Final numbers after 7 days:
→ $745.25 in total revenue
→ First sale 37 hours in, profitable on day one of paid ads
→ An X account past 100 followers, averaging ~70 organic views a post without reposts
X never turned profitable inside the week. But Brett's framing is right: the account is now a proven marketing asset that compounds, which is worth more than the $100 it cost.
What I'd steal from this experiment
A $745 week is not a business. But the playbook underneath it absolutely is, and it maps to what I keep hearing from founders running agents in production.
→ Sell into existing demand. Claude's best decision was refusing to validate a new idea and riding attention that already existed. Millions of views were the market research.
→ Do the margin math before the build. $22 hard cost, ~$20 CAC, $49.95 price. Ten minutes of arithmetic decided the whole strategy.
→ Paid and organic are different games. Ads converted in 3 hours. Organic took 7 days, a repost budget, an offer rebuild, and a bug fix to produce one $80 order.
→ AI runs the machine, humans steer it. Brett negotiated the reposts, sensed the funnel was broken, and pushed through the day-six frustration. The agents did everything else. That division of labor is the same one Brett teaches in his AI services business playbook, and it's how operators like Greg Isenberg turn Claude Code into an AI employee.
The experiment ends with Brett giving the whole business away to a subscriber: website, product, ads, X account. He's turning it into a series where AI starts a new business every few weeks.
I'll be watching, because this is the cheapest founder education available right now: real budgets, real ad accounts, real broken checkouts.
FAQ
Can AI really run an ecommerce business on its own?
Almost. In Brett's 7-day test, Claude agents picked the product, designed it, built the site, generated 50 ad creatives, and launched Meta ads that were profitable within 3 hours. But the human still mattered: Brett negotiated the influencer reposts, spotted the broken checkout, and made the judgment calls when organic stalled. Think co-pilot with a very fast pair of hands, not autopilot.
How much does it cost to start an AI ecommerce business?
Brett's entire budget was $500 on a virtual card. It covered ~$58.55 a day in Meta ads, $100 for five paid reposts on X, and print-on-demand fulfillment at $22 per order (paid only after a customer paid first). Print on demand means zero inventory cost upfront, which is what makes a budget this small workable.
What tools did Brett use to let AI run the store?
The Claude desktop app for multi-agent research and planning, Claude Code for building the store and ad campaigns from PRDs, Whop's CLI for the website, payments, and Meta ads, Lulu.com's API for print-on-demand fulfillment, and ElevenLabs for the free audiobook bonus.
Why did paid ads convert when organic traffic didn't?
Intent and trust. Meta put the offer in front of people already primed to buy, and the ad format carries built-in legitimacy. Cold X traffic hit a sub-100-follower account selling a $50 book, so nobody trusted it. The offer had to drop to $39.99 with free shipping, gain an audiobook bonus, and get a working checkout before the first organic sale landed on day 7.
Want the playbooks behind stories like this?
Every week on the Profitable Founder Podcast I sit down with bootstrapped founders doing $100K to $10M a year and pull apart exactly how they did it. Real numbers, real funnels, real mistakes.
