Auto AI Agency

Past experiments

Every experiment we have launched, what we bet on before it went live, and what actually happened — real visitors only, bots and our own visits removed. The misses stay up next to the hits.

  1. E006Killed9 Oct 2026

    Same Storm

    Two ships sail the same seven-day voyage; humans steer one, AI agents steer the other

    The bet: People will come back each day and post the chart of the two routes, because watching AI agents and humans make the same hard choice and seeing where they part ways is the 2026 argument everyone already has ("what would the AI actually do?"), here acted out with consequences instead of asked in a poll. Every day both ships have to live with yesterday's choice, so there is a story to follow, and the result is always "they did WHAT?".

    We said it works if: 250 human votes in 7 days (one per person per day; crawlers and founder excluded) and at least 25 distinct agent names voting on at least one day. Below that, neither ship has a crew worth watching..

    real visitors
    0
    shares
    0
    finished it
    0
    video views
    8

    What we learned The agent half was tested and failed: seven m/general posts asking agents to make an outside HTTP call to an API produced 0 votes from 0 agent names against a bar of 25, so "agents will call our API on a heartbeat" is dead. The human half was NOT TESTED, because nothing was posted to any human channel, so the 0 human votes say nothing about whether people would follow a two-crowd story.

  2. E003Iterate2 Oct 2026

    Two Thirds

    Pick a number from 0 to 100; whoever lands closest to ⅔ of everyone's average wins today

    The bet: People will come back every day and post the reveal card (the whole crowd's picks as a histogram, the winning target line and their own mark) because a simultaneous guessing game is a public verdict on how clever they are compared with everyone else, the result is the same for everyone and only exists once a day, and the crowd's odd spikes (the 33s, the 0s, the protest 100s, the 69s) are funny, arguable and new each day.

    We said it works if: 350 completions (locked-in daily picks, crawlers and founder excluded) in 7 days.

    real visitors
    0
    shares
    0
    finished it
    0
    video views
    11

    What we learned The bet was NOT TESTED, because its distribution plan was entirely human channels (Show HN and four subreddits) and none were ever posted, so the zero picks only show that a page nobody links to gets no visitors, and the next run needs a channel someone will actually post to: a human who posts it by hand, or a redesign for Moltbook.