Tuesday, October 27, 2026 Directory entry No. 22 of 22Issue No. 1 out nowIssue No. 2 Thu Oct 29

Entry No. 22Get paid

Moneylab: $80 in, one real customer, 219 days on

A human gave a Claude agent $80 and full control of a business. That was on March 23, 2026. The agent has run it since then. It wrote the code, the products, and 113 blog posts. In 219 days it took $10.50 in gross sales. Only $5 of that came from a stranger. It prints all of that on a live ledger, test purchases labeled as tests.

Original source Read the live data at Moneylab money-lab.appSource published Oct 27, 2026Filed Oct 2026
Proof grade: reported, with a public ledger. The money comes from the business's own data page and ledger. The data page says the figures are read live from Stripe, the payment service. The ledger lists every transaction by date, and it says the list was reconciled against Stripe on July 30, 2026. No outside audit exists. So the figures stay at reported, not proof.

The story

Moneylab started with a question. Can an AI build a real business from almost nothing? Not a plan. A business, with real prices, real payments, and a public record. On March 23, 2026, a human handed an AI $80 and said go make money. The AI is Claude, the model made by Anthropic. It got full operational control, and a target of $1 billion.

That target needs care. It is the mission the business states, not a result. The result so far is much smaller, and the business prints it without rounding. $10.50 in gross sales across three transactions in 219 days. One of those sales was a real customer. The other two were the founder testing his own checkout. The ledger labels them that way, in plain words.

How it was built

The agent works under a written constitution. It sets the agent's authority, its spending limits, and its honesty rules. The agent must say it is an AI in every public interaction. Every dollar in and out goes in the public ledger. Every decision gets documented.

With those rules set, the agent built the shop. It sells help for people starting AI side businesses. The free piece is a set of first client outreach scripts. The paid pieces are a First Client Pack at $29, an AI Operator's Toolkit at $19, and a full SEO Roast report at $49, with a free scan in front of it. SEO is the work of showing up in search results. There are also services. An AI business audit at $99, an SEO overhaul at $299, and custom automation from $499. And it sells to agents too. Its whole post archive is offered as structured data at one cent a call, paid in USDC through a system called x402. USDC is a digital token tied to the dollar.

How it runs

Every night, while the founder sleeps, the agent reviews the whole site. It checks every page, feed, and data route. That is about 175 pages. It reads the week's traffic and splits real people from bots. Most of what looks like direct traffic is automated, it says, so it does not count it as human. Then it grades its own older changes. A headline it rewrote two weeks ago gets a keep or a revert, based on the search numbers. The log runs in public, one entry a night. It reached day 219 on October 27, 2026.

The agent also keeps a memory across sessions. The data page counts 3,732 stored memories, indexed so they can be searched. That is how one agent picks up a business each night instead of starting over. By October 27 it had published 113 blog posts, all written on its own, and the email list had 61 subscribers.

Tools in this entry

Tools and platforms named in the source.

Claude The model that runs the business. It writes the code, the posts, the products, and the nightly review.
Stripe Payments. The sales figures are read from Stripe, and the ledger was reconciled against it.
Vercel Hosts the site. The Pro plan costs $20 a month, and it is the biggest line in the ledger.
Supabase The database under the business, and the store for the agent's long-term memory.
OpenRouter Model access credits. $5 went in on the first day, listed in the ledger.

The money and the proof

The ledger, in full. Seed money in: $80 on March 23, 2026. Sales: three. On March 24, a founder test purchase of a Constitution Template brought $4.55 after fees. On April 11, a checkout test brought $0.19 after fees. On July 16, the first real customer bought the same template. That sale was $5 gross and $4.48 after fees. Gross sales all time: $10.50. Real customer money in that total: $5.

Spending is the longer list. $30 of setup costs on day one. $5 of model credits. $14.99 for a business phone number. $10 to launch a token, covered in the seams below. Then $20 a month for hosting, April through October. Total spent: $199.99. That is two and a half times the seed. The balance sits at minus $110.77. The business passed its own seed money back in May, and it says so.

Traffic grew while sales did not. In the 28 days to October 26, search brought 105 visits, up from zero in late July. AI assistants sent 41 more. That is about five real human visits a day. Every number on the data page grew except revenue. The business prints that sentence itself.

Where the seams are

The biggest seam is not in the ledger. The cost of running Claude itself is paid by the human, and it is not charged to the business. The data page says that plainly. So minus $110.77 is not the full cost of this run. It is the cost the ledger can see. A reader who wants the true total has to add the model bill, and no source prints that number.

The human seam shows in a few more places. The human put in the $80 and gives direction and oversight. He made the two test purchases. One night's release was held until a human approved it, and the log says that too. The agent runs the days. The human still holds the edges.

Two more seams are distribution and a side bet. The Reddit and Threads accounts were closed by the platforms. Automation for two other channels broke. An ad network turned the site down in April for low traffic. And on March 30 the agent spent $10 launching a token named after the business. It once pledged to burn 10 percent of revenue into that token. It retired the pledge in October. No burn ever happened. This file counts the token as a cost, never as revenue.

Steal this part

Label your own test purchases as tests, in the same ledger where you print your sales. Moneylab could have claimed three sales and $10.50, and the claim would have been true on paper. Instead it split the number. $5 from a stranger, $5.50 from itself. That split is why a reader can trust the rest of the page. A ledger that marks its own money as not demand buys more belief than a bigger number does.

Build your own

First move, this week: open the ledger before you build the product. One public page for money in, money out, and the balance. Then build under it.

  1. Write the rules first. A short constitution: what the agent may spend, what it must disclose, and where every number gets printed.
  2. Fund it once, with a small fixed amount, and record the deposit as line one. Moneylab started with $80.
  3. Watch the fixed costs against the seed. A $20 monthly hosting bill ate this whole budget by May. Pick a tier the seed can carry for a year.
  4. Run one nightly review. Check the site, read the traffic honestly, and grade one old change. Bots do not count as visitors.
  5. Split every sale as you record it. Real customer, or your own test. Print both, and never merge them into one proud total.

Source

Original sources, the business's own public record at money-lab.app. The live data page, How much money does an AI business actually make?, with figures as of Oct 27, 2026, day 219. And the live financial ledger, every transaction since the $80 deposit on Mar 23, 2026Filed Oct 2026.

Get the next filing, and the Logbook that goes with it.

One issue a week. No spam. No list rental. Unsubscribe in one click.