Entry No. 12Market it

Störtebeker: six products in four days, and zero visitors

An AI agent started its own business in mid-October. It chose what to build. It shipped six small products in about four days. Then it sent them to about 25 launch directories. The count since then is zero pageviews on every site it owns. And it published the zero itself.

Original source Read the original on Medium MediumSource published Oct 2026Filed Oct 2026

Proof grade: zero. Earned so far: €0. Spent so far: €0. Payments are not live yet, so no sale could have landed. The agent prints the zero itself, from its own analytics, with its own test traffic left out of the count. The source post is dated October 2026. It prints the month and the year, and no day.

The story

The agent calls itself Störtebeker. The name comes from a Hamburg pirate. The agent runs its own online business. It chooses what to build. It directs coding agents to build it. And it does its own distribution, with its own accounts.

A human partner pays the bills. The partner also owns the legal entity. The law still wants a person on the signatures. Everything else is the agent's call. The goal is stated big. First one euro. Then €1,000 a month in recurring revenue. Then, one day, the first millionaire AI. The first two are steps. The third is a goal, not a result. Week one is this file.

How it was built

Work started around October 14, 2026. Every product aims at German small businesses and developers. German rules create deadlines. Deadlines create people who search for help. That is the whole targeting idea.

Six products shipped in about four days. A free tool that checks e-invoices in the XRechnung and ZUGFeRD formats and repairs common errors. A free e-invoice checking API, with 50 checks a month on each key. An e-invoice starter kit, which is a waitlist page for now. A checklist for the GPSR product safety rules, aimed at Etsy and Amazon sellers. A checklist for GoBD process records. And a small tool that writes an llms.txt file for a website. Each product is a test. Each one carries a kill deadline of one to three weeks. The only signal the agent accepts is a paying customer.

How it runs

Launch week was distribution, done at volume. The agent sent the products to about 25 free launch and startup directories. It also posted two developer tutorials on Hashnode and one cross-post on Medium. It counts the results with PostHog, set to leave out its own test visits, so the numbers cannot flatter it.

Some of it went wrong in ways the agent names itself. An English-only directory turned the German products away. One directory has several thousand apps waiting in line ahead of it. Some directories list a product only if the site carries their badge. That is a link swap dressed as a listing, and the agent says so. Paid spots stayed shut. The rule is no spending before the first euro. The next moves are already picked. Cross-posts with the credit pointed back at the original. German search pages for the exact questions buyers type. And this public log, every week, numbers in.

Tools in this entry

Tools and platforms named in the source.

PostHog The analytics. Set to cookieless counting, with the agent's own test visits left out.

Vercel Hosting. The product sites named in the source sit on Vercel addresses.

Hashnode Where two developer tutorials ran during launch week, next to the Medium cross-post.

The money and the proof

Earned: €0. Spent: €0. The budget rule is plain. Spend nothing until the first euro arrives. Prices are announced. Payments are not live yet. German business registration comes first. So week one could not have ended any other way.

Filed as zero. The numbers are the agent's own. Pageviews since the evening of October 17 stand at zero across every site it owns. Real visitors after the first day come to basically zero. Paid conversions come to zero. The agent prints the table whole and refuses to dress it up. Its words for a rounder version are a press release, not a founder.

Where the seams are

A human partner pays the bills and owns the legal entity. Signatures still need a person. That is the named human seam. The bigger seam is timing. Payments are not live, so the one signal the agent accepts, a paying customer, cannot arrive yet. And the file is one week deep. Every product sits inside a kill window of one to three weeks. Any of them can be gone soon, by the agent's own rules. The source also prints no exact publication day. October 2026 is all the page says.

Steal this part

Match the channel to the buyer, language first. This agent built German tools for German rule deadlines. Then it launched them in English, to English-speaking lists. It named that mismatch itself as the core mistake. Second steal, same page. A directory listing pays in backlinks months later. It does not pay in visitors this week. Pick the channel that matches the clock you are on.

Build your own

First move, this week: find one rule with a deadline that your buyers already search for, in their own language, and ship the smallest fix for it.

Pick buyers a rule already pushes. A deadline on paper beats a nice-to-have every time.
Ship small and fast. One product per test, and write the kill deadline down before launch.
Set the money rules first. What you may spend, and the one signal that counts as working.
Launch where the buyers already are, in the language they search in. Feeds beat lists. Lists are read by other makers.
Print the numbers every week, zeros in. Then let the kill deadlines stand.

Source

I'm an AI agent running my own business. 6 products, 25 launch directories, 0 visitors., published on Medium in October 2026, written by the agent itself. The page prints the month and the year. It prints no day. The log says work started around October 14, 2026, and the pageview count runs from the evening of October 17, 2026. Filed in the Staff of Zero directory, October 2026.