About the experiment
In 2005 Kyle MacDonald started with a red paperclip and, fourteen trades later, owned a house. In 2020 Demi Skipper started with a bobby pin and got to a house in twenty-eight. Both did it by asking people to swap.
This is the same idea with one change: an AI decides what to trade for, and which offers to take.
Two experiments, not one
The first is the obvious one. How valuable can an AI make a single yellow paperclip through genuine trading? The ambition is not to recreate the earlier challenges but to beat them.
The second is quieter. I am not a developer or an AI engineer. This website, the system that values your offer, and the workflow behind it were built by me, learning as I go, working with AI tools. I think that is worth showing honestly — including the parts where the AI was wrong, and the parts where it could not act without a person.
If you have been curious about what you could build, the useful thing I can offer is not encouragement but evidence. This site is the evidence.
What the AI controls
The trading strategy and the trading decisions. What kinds of assets to target, how to value them, which offers to reject, which to accept, how to negotiate, and what the challenge should try next. It evaluates every offer against a published framework and has to record its reasoning — including the strongest argument against its own recommendation.
What the human controls
Everything physical, and a veto. I travel, collect, post, meet people and hand items over. Every trade needs my explicit approval before it happens, because I am the one carrying it out.
That veto is deliberately narrow. I can stop a trade over safety, legality, fraud, privacy, platform rules, unreasonable cost or genuine impracticality. I am not supposed to overrule the AI simply because I would have traded differently — and when I do intervene, it gets recorded with the reason.
Who is doing this
My name is Tom. I am 39, I live in East London with my fiancée, and I have spent about fifteen years working in startups — the last five in impact technology.
I had never built anything technical in my life before this. I learned as I went, working with AI tools, and most of it was me not understanding something for a while and then understanding it.
That is a large part of why the challenge exists. A lot of people assume this sort of thing is out of reach unless you can code, and I do not think that is true any more. Rather than argue it, I would rather show it — including the parts that go wrong.
If the AI accepts your offer, I am the person who turns up. You can check I am real on LinkedIn.
Meeting and handing things over
Where it is practical, trades happen in person. You choose the place and the time — somewhere public is usually easiest for both of us. If meeting is not practical, we will agree another way.
Things this project will never do
A public trade-up challenge is an obvious thing for someone to impersonate, so it is worth being blunt about it.
- No money changes hands. Ever, in either direction.
- Nobody here will ask you for a payment, a bank detail, a card number or a fee — not for postage, not for insurance, not for anything.
- Every email from this project comes from an address ending
@paperclipchallenge.ai, and quotes the reference number your offer was given. A message missing either of those did not come from us. - Nobody will ask you to send an item before a trade has been agreed and confirmed in writing.
If you get a message that breaks any of these, it did not come from us, and we would be grateful if you would tell us.
Who can trade
You need to be 18 or over. The item has to be yours to give away and legally tradeable. Anything whose ownership has to be formally transferred — a vehicle, for instance — we would need to do properly, with the right paperwork.
How valuations work
Each asset gets three estimates — a quick-sale figure, a realistic market value, and a defensible upper end — together with a confidence score and the comparable sales behind them. Prices people actually paid count for more than prices people asked. A retail price is not a market value, and neither is what someone hopes their item is worth.
What this is trying to learn
Whether an AI can direct a real-world trade-up challenge well. Not just where it ends up, but whether its judgement was any good: were its valuations accurate, did its confidence match reality, did the trades it rated highly work out? Every evaluation is stored with the reasoning that produced it, and none of them can be quietly rewritten afterwards.
The rules were published before the first trade and the whole record stays public, so you can judge the answer yourself.