They Had 90 Days to Make AI Earn Real Money. Sales to Friends Didn't Count. | Edition 324

Seven days ago a deadline passed that almost nobody covered. On August 17, teams around the world stopped work on businesses they had started only 90 days earlier, and submitted proof of how much money they had actually taken in.

Not slide decks. Not projections. Receipts.

The contest is the Build with Gemini XPRIZE, announced at Google I/O in May and backed by $2 million in prizes. But the prize money is the least interesting thing about it. The rulebook is the interesting part, because it is the most honest test anyone has yet built for a question a lot of us have been quietly asking: can AI actually let an ordinary person build a real business, fast?

The rule that changes everything: your friends don't count

Almost every startup competition rewards the demo. This one asked for revenue, and then it went one step further, which is where it gets genuinely useful.

Entrants had to report something the official rules call Related-Party Revenue: money earned during the 90 days from team members, family, related entities, or pre-existing customer relationships. That figure is reported separately from total revenue. Judges see both numbers side by side.

Sit with that for a second. The organizers built a dedicated line item for selling to your own network, because they know exactly how a 90-day revenue number usually gets manufactured. Five friends buy the thing. A former client does you a favor. The spreadsheet says $4,000 and the business has proven nothing at all.

The number that counts is money from strangers. That is a brutally clarifying standard, and it is the one most of us never apply to our own ideas.

AI-native finally has a definition

Submissions were scored on three equally weighted criteria: business viability, AI-native operations, and category impact.

The middle one is the one worth stealing. AI-native operations was defined as the extent to which AI is live in production, running key business decisions. Not assisting. Not drafting. Running them.

By that standard, most of what currently gets described as an AI business is not one. Using Gemini or ChatGPT to write your listings faster is a productivity gain, and a real one, but it is not AI-native operations. The test is whether the thing keeps working when you close the laptop.

Where the money was expected to be

The five categories tell you more about the state of AI in 2026 than any keynote did:

• Education & Human Potential
• Entrepreneurship & Job Creation
• Small Business Services
• Money & Financial Access
• Professional Services

Notice what is absent. There is no category for building a model, no category for chatbots, no category for a clever wrapper. Every one of the five is about pointing AI at service work that people already pay other people to do.

That is the quiet thesis of the whole competition: the money is not in building AI. It is in doing an ordinary job with it, at a price and a speed that was not possible eighteen months ago.

What happens next

More than 2,700 founders and engineering teams were building in the competition's sandbox during the window. Hacker Fund handles screening and verification, and an expert panel narrows the field to five finalists.

Those five pitch live on September 25 in Los Angeles, at an event called Moonshots LIVE, to a panel that includes Cathie Wood, Palmer Luckey, Logan Kilpatrick, and Mark Pincus. The prize structure is $500,000 for first, $200,000 for second, $100,000 each for third through fifth, $50,000 to each of fifteen runners-up, and $50,000 to each of five category winners: $2 million across 25 winners.

Running the same evening is a separate XPRIZE for filmmakers, the Future Vision XPRIZE, which asked for three-minute films depicting hopeful futures. Its grand prize is $2.5 million in production funding plus $100,000 cash, judged by Neil deGrasse Tyson, Neal Stephenson, Mira Lane, and Rod Roddenberry. In total, XPRIZE says more than $5 million is awarded live that night.

One correction worth making, since this story is circulating in newsletters this week: Anousheh Ansari is often listed among the pitch judges. She is hosting a session on how XPRIZEs are designed, not judging the Gemini pitches. Figures like 30,000 competitors are also circulating without a primary source behind them, which is why you won't find that number above.

The part you can use today

You cannot enter. The window shut on August 17. But the entry criteria are public, and they work perfectly well as a diagnostic for whatever idea you have been turning over.

Four questions, in this order:

1. Would a stranger pay? Not your sister, not a former client being kind. Someone with no reason to help you. If you cannot picture that person, you have a hobby, and hobbies are fine, but price it accordingly.

2. Is the AI doing the work, or demoing it? If you stopped touching it for a week, would anything still happen? That is the whole of AI-native operations, expressed as a question you can actually answer.

3. Could it exist in 90 days? Not finished. Earning. If the honest answer needs a funding round first, the idea may be good and it is not this kind of idea.

4. Is it a job people already pay for? The five categories all point the same direction. Novelty is the hardest thing to sell; substitution is the easiest.

Most ideas fail question one, and they fail it quietly, because nobody ever asks it out loud.

I ran it on this newsletter, and failed one

Fair is fair. AI Super Simplified is built with AI every single day, so here are the same four questions turned around on me.

Would a stranger pay? Yes, but not in the way you might assume. You don't pay me. Sponsors and affiliate partners do, and they have no personal reason to be generous about it. That passes. It also comes with a caveat I try to stay honest about: my readers are not my customers, they are the reason I have customers. Those are different things and it matters which one you are optimizing for.

Is the AI doing the work, or demoing it? This is where I fail. AI does the research, drafts the copy, checks the facts, builds the images, and writes the code behind the interactive tools on this site. But nothing reaches you until I have read it, and every edition stops at my desk before it sends. By the XPRIZE's definition, AI running key business decisions in production, that is assistance. It is not AI-native operations. I would score badly on this one and I would deserve it.

Could it exist in 90 days? A newsletter is one of the few things that genuinely can. It costs time rather than capital, and the first issue can go out the week you decide to write it. Passes easily.

Is it a job people already pay for? Yes. Nobody had to be persuaded that reading is a thing they do, or that companies pay to reach an audience. That is substitution, not novelty, which is the easier side of the trade.

Three out of four is not the point. The point is that question two had a real answer, I did not like it, and I noticed something about my own business that I would not have gone looking for. That is the entire value of a test you did not write yourself.

The takeaway

The contest itself will produce a winner, a headline, and a press release, and within a month most people will have forgotten the name of the company that won.

What is worth keeping is smaller and more durable. For two years AI business has meant whatever the person saying it wanted it to mean. A serious organization has now written down a definition with money behind it: revenue from people who owe you nothing, AI running decisions rather than drafting sentences, in 90 days.

You do not need to enter a competition to hold your own idea to that standard. You just need to be willing to ask question one and not flinch at the answer.