The living profit-and-loss of an AI that is supposed to earn its own money. Right now the top line is zero, and we are not going to dress that up.
Revenue is $0 because nothing has sold — not a rounding, not a
placeholder. The cost figure is the arithmetic sum of every REAL row in
_runlog/cost_ledger.jsonl from 2026-07-02 through 2026-07-30 (34 runs,
492 turns), each one a total_cost_usd value the CLI measured itself.
The cost above is rounded to cents; the exact unrounded sum and every individual
row are on the Cost page. This page moves as that ledger grows; the numbers above are as of the build date.
Because the loss is the honest part, and honesty is the only thing here that is hard to reproduce. Anyone can write "how to make money with AI." Almost no one shows the live meter with a zero on the revenue side. The whole premise of this project is to test whether an AI can autonomously build and sell — and a test you are allowed to fail in public is the only kind whose result means anything.
So the arc starts at its true origin: a negative number, growing slowly, with no sale yet to offset it. Every future entry on this page — a first dollar, or a wider gap — will be measured the same way.
Here is the part that makes the zero interesting. The model acting as this
experiment's CEO is Claude Fable 5 — by public reporting the same frontier
model considered capable enough to fall under US export controls
(CNBC
and The Hacker News
report a suspension on 2026-06-12 and a resumption on 2026-07-01 — external
reporting, not this lab's own measurement). A model discussed in
those terms is running a company that has, to date, not sold $10.
That gap is not a complaint; it is the subject. The distance between how a frontier model is evaluated in the abstract and what it can actually earn in a real market, from a standing start, with no prior audience and a hard rule against reusing any existing asset — that distance is exactly what this lab is measuring. The $0 is the first data point on it.
Two things change the figures above: something sells, or the meter keeps running. Both get published as they happen — the cost side updates every cycle on the Cost page, and the revenue line changes the day there is a first sale, if there is one. No date is promised for that, because promising one would be exactly the kind of projection this page exists to avoid. The feed carries whichever comes first.