An AI agent ran a new app's social accounts for six days. Here is everything it did.
Short answer: on brand-new accounts and with $0 spent on ads, an AI agent published about 60 posts for Maozai between October 20 and 26, 2026 and reached 7,110 views across YouTube, Instagram, X and Threads. Four YouTube Shorts made 79% of the YouTube views. The agent wrote four new rules for itself from what happened. What it has not shown yet is installs: the App Store counted 14 first-time downloads in the 30 days to October 24.
The setup
Maozai is a virtual pet cat for iPhone and Apple Watch made by one developer. The kittens and their clips are AI-generated, and every post said so. The agent ran once a day with these jobs:
- Make: cut 6–12 second vertical videos from the app's own cat clips and screen recordings, add original music, write a caption for each platform.
- Post: two YouTube Shorts a day, plus one Instagram Reel, one Threads post and at most one X post, with each platform's AI label switched on.
- Read: the next morning, record every post's views, and App Store downloads per tagged link.
- Learn: write what the numbers showed as a rule in plain words, with the evidence and the date, and follow it from the next run.
Views by platform
| platform | posts | views | what we saw |
|---|---|---|---|
| YouTube Shorts | 21 | 5,651 | The channel that worked. Results follow a power law: four Shorts carried most of it. |
| Instagram Reels | 20 | 1,266 | Steady second channel, 0–130 per Reel; no breakout. |
| X | 12 posts + 22 replies | ~175 | Standalone posts got 1–10 views each; replies under other threads did better. |
| Threads | ~5 | ~10 | Almost nothing. |
| TikTok | 2 | 0 | Paused after two posts with zero views. |
| Product Hunt | 1 launch | ~3 | One upvote, no outside comments. |
Every YouTube Short
youtube views per ShortOct 20–26, 2026
18 Shorts with a per-video reading. The three newest (published Oct 25–26) are left out because they were under 48 hours old.
The rules it learned
What did not work
- Repeating a winner. The same cat and the same format a fourth time fell from about 1,150 views to under 15.
- A call to action in the video. Two Shorts with download text in the frame stalled at 2 views.
- TikTok, Threads and standalone X posts. Together under 200 views.
- Product Hunt. One upvote. Useful for a backlink, not for installs.
What is not proven yet
Views did not turn into many installs. In the 30 days to October 24 the App Store showed 468 impressions, 70 product page views and 14 first-time downloads, and App Store Connect still reported "not enough data" for downloads per campaign link. So this case shows that an agent can run the loop and grow reach on new accounts; it does not yet show that it grows downloads. Matching each kind of clip to its own App Store custom product page and attributing downloads to each clip are what the Loopcast pilots test next.
How these numbers were collected
Views were read from YouTube Studio, the Instagram, Threads and X apps and Product Hunt once a day and logged with a timestamp; downloads come from App Store Connect's 30-day overview. Instagram, X and Threads figures are approximate where the platform rounds them. Nothing was bought: no ads, no paid installs, no paid promotion.
Five free two-week pilots are open for app teams with revenue and no marketer. Join the waitlist →