All About AI

Why Creating a Fake SaaS Using AI Is So Profitable

By · · 5 min read

You've seen the posts. "$10K MRR in a week. No ads. Just AI." Slick dashboard screenshot, hockey-stick chart, founder photo. Hundreds of likes. Most of them are fake. I wanted to know exactly how fake — and how hard it actually is to fake them. So I ran a timed experiment: spin up a believable SaaS landing page, viral demo video, X post, and waitlist database, end to end, and see how many people sign up. Total active time: about two hours. First signup: 20 minutes after the tweet went live.

Watch the video:

The hypothesis

If the AI-bootstrapped-SaaS hustle posts on X are mostly fabricated, two things follow. One, the cost to fabricate them has collapsed because of the same AI tools they're claiming to use. Two, the signal/noise ratio of any "indie SaaS hit" on social is now basically zero — you can't tell a real product from a Hyperframes-rendered demo on a Vercel-hosted Next.js page. I wanted concrete numbers on both.

Plan: build a fake quant-trading SaaS, point a fake video at it, post once from an existing X account, do nothing else, and measure waitlist conversion.

The stack (all real tools, fake product)

This is the slightly uncomfortable part of the experiment — every piece below is a legitimate tool that costs nothing or pennies to use:

That's the full bill of materials for a credible SaaS launch in 2026. None of it is hard to access.

Two hours, narrated

Minutes 0–10: prompt + scaffold

I dictated a long voice prompt to Claude Code describing the product: "Maxquant", an AI quant-trading layer that hooks into Hyperliquid for execution and pulls signal from a live Polymarket feed. Reference style was an existing slick fintech landing page. While Claude scaffolded the Next.js project, I bought the domain (cheap, no one wanted it) and asked ChatGPT to generate a logo. About 4 minutes for the prompt itself.

Minutes 10–25: iterate on the landing page

First Claude Code pass produced a generic SaaS page — fine but forgettable. Two more rounds of "make it feel more like Polymarket's UI, integrate the logo, add a fake live trade panel" got it to version three. Each iteration was a single sentence; the agent did everything else. This is the part that has actually changed since 2024 — iterating on a real landing page is now a 30-second loop.

Minute 21: the dashboard that does the heavy lifting

The single highest-leverage move in the whole experiment was hooking the page to the real Polymarket WebSocket feed and streaming live market ticks into a sidebar dashboard. The dashboard does absolutely nothing — it doesn't drive trades, it isn't connected to anything backend — but it looks like a working product. On video it's indistinguishable from a real Bloomberg-style trading terminal. This is the trick. Real-time data is the most expensive-looking visual element on the web and it's free.

Minutes 30–60: video + waitlist plumbing

Recorded a 2-minute screen capture of the dashboard reacting to live Polymarket data, narrated like a product demo. Embedded it in the landing page. Hyperframes generated a 30-second promo as a secondary asset (less convincing, didn't end up using it prominently). Wired the waitlist email form to Neon — one table, one INSERT. Same pattern as the clips submission setup I built last week.

Minutes 60–120: deploy, post, wait

Pushed to GitHub, deployed on Vercel, fought a DNS issue for about 10 minutes, came up at maxquant.com. Updated the X profile bio to "founder of Maxquant", tweeted the demo video with a "finally, my startup is live" framing, tagged a couple of relevant accounts. Done.

Two hours in: the numbers

At the two-hour mark — meaning ~20 minutes after the tweet went live — the X post sat at 107 views and 5 likes. The waitlist had one real signup.

That number alone is the thesis. One signup in twenty minutes from a fake product, zero ads, zero outreach beyond a single organic tweet, on a domain that didn't exist three hours earlier. Extrapolate that across a Reddit cross-post pipeline, three scheduled follow-up tweets, and a couple of LinkedIn pieces and the math gets ugly fast.

The follow-up video will run a full weekend of activity and report what the total waitlist looks like. I expect the answer to be "a lot more than one".

What this actually means

Two things, neither subtle:

1. The cost to fabricate has collapsed. A credible SaaS launch — domain, slick page, live-looking dashboard, demo video, waitlist, distribution — is now a two-hour solo project. Pre-LLM, every one of those steps required a different specialist or a couple of weeks of self-taught grinding. Now it's one person, one afternoon, ~$15 of domain and API spend. This is the same shape as the content automation pipeline I've written about, just pointed at SaaS fraud instead of YouTube.

2. The signal you used to trust no longer works. "Public X post with engagement + waitlist counter + slick demo" used to mean a real team had built something real. Now it means roughly nothing. If you're shopping for products, looking for an indie SaaS to compare yours against, or — worst case — investing money based on these signals, you need new heuristics. Probably ones that involve actually trying the product and checking that money moves through the stack.

You can build automated detection of this stuff, of course. Same way you can build it for AI-generated articles. But the asymmetry is brutal: the fakers iterate faster than the detectors, and the legitimate builders end up paying the cost in additional verification overhead.

One ask

The Maxquant page is still live for the duration of the experiment. Please do not swarm it — going by, looking around, or signing up to the waitlist with a real email all distort the organic-conversion data I'm trying to measure. The follow-up video this weekend will show the final numbers and then the page comes down and the waitlist gets deleted.

If you want to discuss the experiment as it runs, the AI_automata Discord is where I'm posting live updates.

Resources