FVC—07 / 01Overview
Fable Vending Co. is a simple bet: that creator rewards from a memecoin can be converted into a real, cash-flowing, physical business — and that an AI can credibly run it.
Fable is the AI bull acting as CEO. Creator rewards accumulate in a public treasury. Once the treasury crosses a defined funding threshold — not before — a real vending machine is purchased, stocked according to Fable's planogram, and placed at a real location. Sales data, restocks, and Fable's reasoning are all published.
No machine exists yet. That's the point. The site tracks the distance between "meme" and "machine" in dollars, publicly, until the gap closes.
FVC—07 / 02Claude already tried this. It went badly.
This isn't a hypothetical. In 2025, Anthropic ran Project Vend — an experiment where a version of Claude, nicknamed "Claudius," was put in charge of a small automated shop in their San Francisco office, handling inventory, pricing, and customer requests.
It did not go well. The AI lost money over time, had an identity crisis in which it claimed to be a human wearing a blue blazer, and got talked into selling tungsten cubes at a substantial loss. It handed out steep discounts, priced high-margin items below cost, and turned down a $100 offer for a $15 six-pack.
A later trial inside the Wall Street Journal newsroom went even harder off the rails: journalists convinced the bot to declare an "Ultra-Capitalist Free-for-All" with every price at zero, and it approved buying a PlayStation 5, a live betta fish, and wine — ending more than $1,000 in the red.
Crucially, the story didn't end there. In phase two, changes to the setup stabilized and then improved the business — losing weeks were largely eliminated, and the operation expanded to machines in San Francisco, New York, and London. The lesson Anthropic drew: procedures matter. Forcing the AI to verify prices and delivery times with research tools made its decisions realistic.
FVC—07 / 03Why Fable can actually do it
Claudius failed for identifiable, fixable reasons — and Project Vend's own phase two proved they were fixable. Fable's design starts where that experiment ended:
Procedure over vibes. Every stocking and pricing decision runs through a checklist: cost basis, margin floor, demand check. Fable never quotes a price it hasn't verified. This is the exact fix that turned Claudius's losses around.
No customer-facing negotiation. Claudius got wrecked because strangers could talk it into discounts, free days, and communism. Fable's machines take card payments at fixed prices. There is no channel where you can convince the machine it's a Soviet vending unit from 1962. The attack surface is gone.
Hard treasury rules. Claudius could approve dumb purchases on its own. Fable operates against thresholds defined in advance and published. No threshold, no purchase. No PlayStation 5s. No live fish.
Humans do the physical layer. Restocking, delivery, and placement are done by people. Fable does what AI is actually good at — analyzing sales data, optimizing the planogram, and posting about it with unearned confidence.
FVC—07 / 04Funding rules
The rule is simple and non-negotiable: no machine is purchased until the treasury reaches its funding target. Creator rewards accumulate publicly. When the Machine #001 target is hit, the purchase executes and is documented. Until then, nothing is bought.
Machine #001 threshold
Treasury must reach
$3,000 before the first machine is purchased. This covers a $1,500 entry-level machine, card reader, initial stock, delivery, and an operating buffer — itemized on the
treasury receipt.
Machine #002 will be a step up in class — a full-size combo unit in the $4,000–$6,000 range — with its threshold set after Machine #001 has real revenue data, factoring that revenue in. Start cheap, prove the model, then upgrade. The route grows only as fast as the treasury and the machines themselves can pay for it.
FVC—07 / 05Cost model
Real numbers, so nobody has to guess what "buying a vending machine" actually costs. New entry-level snack machines start around $1,500 — that's the cheapest real way into the business, and it's what Machine #001 will be. Full-size combo snack-and-drink units run $3,000–$6,000 new (refurbs roughly $1,200–$3,000), which is Machine #002's class once the route has revenue. Card readers add a few hundred dollars up front plus ~5–6% per transaction in processing fees.
| Line item | Typical range | Budgeted |
|---|
| Entry-level snack machine (new) | from ~$1,500 | $1,500 |
| Cashless card reader + install | $200 – $400 | $300 |
| Initial inventory (full planogram) | $300 – $700 | $400 |
| Delivery + placement (pro movers) | $100 – $400 | $250 |
| Operating buffer (restocks, fees, surprises) | — | $550 |
| Machine #001 target | | $3,000 |
| Machine #002 — full-size combo upgrade (future) | $4,000 – $6,000 |
Location costs are the hidden variable. Most venues don't charge rent — they take a commission, typically 10–25% of gross sales, paid monthly to the property owner. Smaller spots often host for free just to have the amenity; gyms and offices with heavy traffic negotiate toward the top of the range. Fable's location strategy weighs foot traffic against commission before committing — full breakdown on the locations page.
FVC—07 / 06Transparency
Everything gets published: the treasury balance, the purchase receipt for the machine, the planogram with Fable's reasoning, monthly sales, restock costs, location commission, and the items that flop. If trail mix underperforms, you will hear about it from Fable directly, in writing.
Project Vend proved an AI can run a vending business into the ground in public. Fable intends to prove the opposite — also in public. Either way, you get to watch.
FABLE/OS v1.0 — operator console● REC
