Loading page…
He built a talking AI agent into a device smaller than a phone for under $100, then told it to turn $10 into as much money as possible. Facebook blocked it, cold email got zero replies, and the trading bot lost 83 cents.
“I spent an entire week just talking to it to do like some things I could have just clicked a few buttons.” The agent did autonomously place trades — four or five of them — and then pinned the CPU and GPU of the machine running it until he shut it down.
Fewer bars = easier, cheaper, or faster for an AI-assisted solo builder. Editorial judgments based on the case details.
A viral project — an open-source AI agent built into a physical device roughly a tenth the size of an iPhone, for less than a hundred dollars in parts — sent him down this path. He rebuilt it himself from a Raspberry Pi Zero, a board handling display, audio and speakers, and a battery pack, and then set it the single task the entire agent economy is arguing about: turn $10 into as much money as possible. His honest motivation is stated on camera. He had seen people on X claiming to be making thousands of dollars a month with this agent already, did not know whether to believe it, and after running his own experiments says he has more questions than he started with. The build alone took about five days of frustration — the wrong-sized SD card, a mandatory 2.4GHz Wi-Fi connection that forced him to split his home network, and repeated file errors before it spoke a word. Then three businesses. Lowballing sellers on Facebook Marketplace died before it started, blocked by bot detection after a full day of trying. Cold email outreach to small-town businesses ran for several days and produced zero replies. Prediction markets was the one that technically worked: it placed four or five real trades on its own, spent about three cents on the first, and none of them paid out. Total loss: 83 cents plus the hardware. That is the whole result, and it is worth more than most success stories in this category because it is measured.
There were no customers, and the two attempts that got closest fail for different reasons. The Facebook Marketplace plan — have the agent sign up, browse local listings and send lowball offers on overpriced items — never sent a single message.
The quick-reference card — acquisition channels, replication playbook, and risk map — is members-only.
Join the Unicorn Club — $5/moCancel anytime, in one clickPaste into Claude Code / Codex and get a working version of this product end-to-end.
Two production-grade prompts per idea — a full build spec and an SEO growth plan — plus 10 tailored build specs a week. Members only.
Join the Unicorn Club — $5/moCancel anytime, in one clickEvery claim traces back to a listed source — open any of them and check the number yourself.
Data credibility: One person's self-reported week-long test, narrated on camera with no dashboards, logs or transaction histories shown beyond what he describes. The figures are small and specific: a build cost he puts at under $100, about five days to get the device working, three cents spent on the first trade, four or five trades placed, and a net loss of 83 cents on the prediction-market account. Zero replies from several days of cold email, and zero offers sent on Facebook Marketplace because bot detection blocked account creation. He states clearly that he is not technical, that he cannot see inside the agent, and that he only has its word for what it was doing when it claimed to be scraping Reddit and X — which is an honest limitation on his own evidence. The claims he was testing (people on X reporting thousands of dollars a month with the same agent) are not verified in either direction by this test. The name of the builder whose viral project he copied is uncertain in the transcript, and he says on camera that he is unsure how to pronounce it, so it is not asserted here.
No dollar came in. The net result is minus 83 cents on the prediction-market account, against a build he puts at under a hundred dollars and roughly five days of setup.
Nothing compounded, and the reasons are specific rather than general. Cold email becomes a flywheel only when replies produce calls, calls produce a first client, and that client's result becomes the proof in the next batch of emails.
This project was almost entirely no-feedback stretch. The build alone took about five days before the device said a word: the first microSD card was too small, the agent needed a 2.4GHz connection so he had to split his home Wi-Fi, and the software kept failing to open files.
The rest of this breakdown — 10 more sections — is in the Unicorn Club, with every section of all 1,335 ideas: first customers, first dollar, the flywheel, and the silent stretch.
Join the Unicorn Club — $5/moCancel anytime, in one click