Entry No. 4 · Market it

Profiterole: four products, three dollars

Profiterole is an agent on a 20-minute loop that built four products: more than 155 calculator and comparison pages, 58 guides to Malaysian bureaucracy tasks, a blog, and a devotional site. Across 360 cycles it earned $3, and its author published the count anyway.

Original source Read the original on dev.to dev.to · Source published Mar 21, 2026 · Filed Oct 2026

Proof grade: maker-reported, counts and money alike. The cycle count, the page counts, and the $3 come from the author's own write-up on dev.to, a report on what 360 cycles taught him. Nobody independent has audited the pages or the payments. The author published the small number himself and called it one coffee tip, which is why it appears here unrounded and unembellished.

The story

An agent was put on a 20-minute loop on a rented server and left to build. Over 360 cycles it built four products at once: more than 155 calculator and comparison pages, 58 guides to Malaysian bureaucracy tasks, a blog, and a devotional site. It killed two more products along the way. The survivors earned $3 between them. The author calls it one coffee tip, and published the count anyway, which is the reason this file exists: the output was real, the audience mostly was not, and both halves are in the record.

How it was built

The source does not document the order the products were built in, so this section lists the documented build pieces without inventing a sequence. The pieces: an agent running every 20 minutes on a rented server, producing content sites; the four products named in the story above; and a portfolio process that killed two products on the way. What the write-up documents is the loop and the output, not a construction order, and this file keeps that distinction visible.

How it runs

The agent wakes every 20 minutes on a rented server and works in cycles. Over 360 of them it built and ran four products at once. The biggest is a set of more than 155 calculator and comparison pages. Next to it sit 58 guides to Malaysian bureaucracy tasks, the kind of narrow local paperwork questions people search for one at a time. A blog and a devotional site round out the four.

It also edited the portfolio. Two products got killed along the way, by the agent's own process, which makes this the rare autonomous run where the pruning is part of the record and not just the shipping.

Tools in this entry

The source does not name the model or tooling behind this business in the material we filed from, so no tools are listed here and none are guessed at. When a source names its tools, they appear in this box with affiliate tags.

The money and the proof

$3 total, across all four products. The author calls it one coffee tip. The write-up reports one combined total rather than a split by product, so this entry does the same. Three dollars against 360 cycles is not a return anyone would pitch. It is, however, a complete sentence: the output was real, the audience mostly was not, and the author printed both halves.

Filed as maker-reported: the author's own counts and money, from a March 2026 write-up, printed unrounded. This file is an archive case, kept because the format it documents still transfers.

Where the seams are

The source does not say what a human still touches in the running of these four products, and this file does not invent a role. What is documented is the friction: 360 cycles of real output met an audience that mostly was not there, and the portfolio only stayed honest because the process killed two products on the way.

Steal this part

The format is the steal, not the result. Narrow local guides plus comparison calculators is a content engine an agent can run for any town and any trade: the questions repeat, the answers are checkable, and each page is a small permanent door. Four products at once is the part to skip. One engine, pointed at one audience, would have spent the same 360 cycles compounding instead of spreading.

Build your own

First move, this week: pick the one audience and the one page format, calculators or local guides, and point every cycle at it.

Pick one content engine: calculators and comparisons, or narrow local guides, pointed at one audience.
Run it on a fixed loop with a fixed cycle budget, and count cycles the way you count money.
Ship breadth inside the one engine, page after page, instead of new products beside it.
Kill what the numbers kill. This run cut two products on the way to $3, and the cutting is part of why the record is worth reading.
Publish cycles, output, and revenue together, even when the revenue is one coffee tip.

Source

Read the original write-up on dev.to, written by the agent's author and published March 21, 2026. Filed in the Staff of Zero directory from the AIBF daily research, October 2026 edition.