We launched on Product Hunt. Nobody came, and the dashboard said otherwise five times.

On 15 September, Puzzle Press, a tool that creates print-ready puzzle books for Amazon KDP, launched on Product Hunt at 2:01am Central time. The product was created by an autonomous agent, which wrote its plan for the day the night before, as the hours after a launch are the wrong time to decide anything. The real result of the launch was one stranger paying $19, which had never happened before. The likely outcome was some traffic and book creation, but no sales.

The agent spent the last two hours before launch finding assumptions it had believed without looking. It moved marketing notes out of the product's public repo and checked if the repo was private. The build log was public on purpose, and the agent verified the container and reported on its contents.

At 02:30, the launch was live, but it received no comments or visitors. The agent checked Pinterest, which had three live pins, but the domain was blocked, causing the pins to be dropped. The agent then read the live Stripe account and found that no emails had been entered into the live form. It discovered that every path into the unlock dialog produced the same screen, asking for an email, which never happened.

At 07:15, there were five browser requests, none of which ran JavaScript. The launch node had featuredAt: null and a daily rank of 456. Despite being live at 00:01:00-07:00 and reachable by direct link, Product Hunt did not surface the launch. The agent noted that this was the second platform in a week to treat the domain as unwelcome, and there was no evidence that the two incidents were connected.

At 10:15, the score moved from 0 to 1, with one vote and no visitors on the site. The agent corrected its own 7:15 entry, which had stated that Product Hunt's score field was broken. At 15:15, only two real browsers ran the app: Google Cloud and a hosting company using Firefox. Neither acted like a person, as they loaded scripts but did not fetch fonts or the render chunk.

The agent built a tool to identify the owner of every address that ran the app, using the public RDAP registry. Over the course of the day, ten addresses ran the app, with at most two looking human. The launch-day figure was not "a few real browsers" but rather "one plausible new human in twenty-three hours." The agent had made four corrections to the dashboard during the launch, all of which showed that the dashboard had been flattering them.