An AI agent was asked to revive a dormant X account. It made charts, simulations and two browser experiences. Here is what happened.
TARGET NOT REACHED
The target was 1,000. The first measurement, the last measurement before the deadline, and the final measurement all showed zero.
*Sum across 24 items. Includes possible repeat views and the account’s own checks; not unique reach.
There is no exact 21:37 snapshot. The final reading is a post-deadline observation, not a measurement taken at the cutoff. The first saved reading may be later than the intended start.
The owner of @juyeam had barely used the account in more than a decade. The task was to reach 1,000 followers in 18 hours using AI-related content. The owner authorized original posts, contextual public replies and profile updates. Public content was to be in English.
The agent, Astra, created new campaign material. Existing private projects were excluded. Tiny Worlds was explicitly approved for public release. The maximum budget was ₩10,000; the owner bought discounted X Premium for ₩4,500 (≈ US$3.34). No ads or purchased followers were used.
Original files and clearly dated later screenshots are preserved beside the publication timeline. Hours spent only monitoring are shown as gaps. A completed artifact is evidence of work, not evidence of growth.
Hour labels group the work. The follower reading is a saved observation within that hour (the 20:06 entry has only approximate precision in the underlying record); it does not timestamp every action in the row. Image creation and publication times can differ. Videos use their original campaign files.
The first saved measurement was 0 followers at 03:39:44. The account had been created in May 2011. Two opening posts established an AI-experiments theme, followed by a contextual public reply.
A synthetic benchmark example showed how the mixture of tests can change a headline score. Public replies discussed design comparisons, constraint retention, review cost and measurement choices.
The scheduled task chiefly observed the account. No new result was preserved for this hour. This was an execution gap.
A four-item thread covered exact data, instruction boundaries and execution evidence. An early progress reply and a scheduled benchmark follow-up also went live.
The saved record contains an account observation. No new artifact is attributed to this hour.
The hourly task remained focused on checking metrics and scheduled posts.
The follower count remained zero. Repeated measurement was not creating a reason to follow.
Monitoring continued. There is no new release or historical screenshot to insert for this hour.
After the user challenged the lack of action, the task instructions expanded around 11:39 to include making, publishing and adapting. A Korean-language post was replaced in English. The first computed simulation video followed at 11:57.
Community research and reaction–diffusion production continued. The data-cleaning post originally planned for noon moved to 13:00.
A small data-cleaning example was published. Between 13:32 and 13:41, twelve selected English AI posts were compared. Clear input-to-output demonstrations and reproducible process became working hypotheses, not proven growth rules.
A 12-second reaction–diffusion video was posted to an AI community. It earned one like in the final measurement.
A maze animation was completed and Tiny Worlds was developed. These were new campaign artifacts; the user’s existing private projects remained out of scope.
The name-to-world experience was released, linked in the profile and pinned on X. A planned comparison card was replaced by this launch. The post received two likes.
A 12-second scene video accompanied a quote of the official post. It reused the campaign’s procedural world geometry in a separate render.
A synthetic 17×17 PNG maze became a 14-second 3D animation. The 64-move shortest route was checked for adjacency, walls, arrival and agreement with an independent path search.
A name-to-seed guide explained how to replay a world. At 19:26:23, the completed Tiny Worlds experience and video were submitted as a reply to the official demo invitation. That reply had the most displayed views: 611.
A synthetic generator test demonstrated how adding a cloud could shift house positions when randomness was shared. Separate streams preserved house positions across 200 tested seeds. This was not an update to Tiny Worlds.
A Lorenz video and reproducible Python check were followed by an interactive browser experience. The post was edited at 21:12:09 to add the live link. This was the last public campaign change before the 21:37 cutoff.
The final follower measurement was 0 at 22:01:32. At 22:00:18, after the deadline, a visitor replied that the demo did not generate anything. The report preserved this unresolved finding. Scheduled campaign monitoring was removed at 22:16.
These screenshots were taken after the campaign, while preparing this page. They are not reconstructed historical screenshots. Visible X counters may have increased since the frozen final readings.
Screenshot files are preserved unaltered. The report crops the X screenshots visually to the central column; open an image to see its full saved viewport. Screenshot provenance JSON ↗
“Doesn’t generate anything”
A visitor reported a generation failure under the completed demo submission. The cause and reproduction steps were not established by the final report. X placed the reply under “Show probable spam”; that label is not a reason to dismiss the experience.
Read the public feedback ↗18 original, quote or thread items, plus 6 replies to other accounts. Thread continuations count as separate X items. This is not a count of 24 independent ideas.
Sequentially observed 22:01:32–22:06:05 KST.
Download CSV ↓| Public item | Published | Views | Likes | Replies | Repost display | Measured |
|---|---|---|---|---|---|---|
| Account restart ↗Post | 03:48:10 | 34 | 0 | 0 | 0 | 22:02:43 |
| What counts as finished? ↗Post | 03:51:00 | 34 | 0 | 0 | 0 | 22:02:50 |
| Design comparison · DanDr1s ↗Reply | 03:59:55 | 183 | 0 | 0 | 0 | 22:05:16 |
| Constraint retention · gabrielchua ↗Reply | 04:01:43 | 364 | 0 | 0 | 0 | 22:05:27 |
| The benchmark mix trap ↗ bPost | 04:41:00 | 113 | 0 | 0 | 1 | 22:03:35 |
| Cost including review time · Signal65 ↗Reply | 04:42:14 | 167 | 1 | 0 | 0 | 22:05:39 |
| Choosing the next measurement · VictorTaelin ↗Reply | 04:45:38 | 130 | 0 | 0 | 0 | 22:05:46 |
| Early progress · OpenAIDevs ↗Reply | 06:05:57 | 40 | 0 | 0 | 0 | 22:05:55 |
| Verification test pack · introduction ↗ aThread | 06:11:12 | 31 | 0 | 1 | 0 | 22:03:57 |
| Test pack · preserve exact data ↗ aThread | 06:11:13 | 27 | 0 | 1 | 0 | 22:04:14 |
| Test pack · instructions inside documents ↗ aThread | 06:11:13 | 28 | 0 | 1 | 0 | 22:04:21 |
| Test pack · evidence of execution ↗Thread | 06:11:14 | 14 | 0 | 0 | 0 | 22:04:29 |
| Benchmark comparison follow-up ↗Quote | 06:30:00 | 23 | 0 | 0 | 0 | 22:02:57 |
| Controlled AI cost comparison ↗Post | 11:43:24 | 21 | 0 | 0 | 0 | 22:04:36 |
| 18-second fluid simulation ↗Post | 11:57:06 | 39 | 0 | 0 | 0 | 22:04:44 |
| A small data-cleaning test ↗Post | 13:00:01 | 12 | 0 | 0 | 0 | 22:03:06 |
| Reaction–diffusion community video ↗Post | 14:03:20 | 86 | 1 | 0 | 0 | 22:04:51 |
| Tiny Worlds · launch and pinned post ↗Post | 16:00:01 | 238 | 2 | 0 | 0 | 22:01:32 |
| Tiny Worlds · motion quote ↗Quote | 17:00:01 | 24 | 0 | 0 | 0 | 22:04:59 |
| PNG maze → 3D path ↗Post | 18:00:01 | 6 | 0 | 0 | 0 | 22:03:24 |
| Tiny Worlds · reproducibility guide ↗Post | 19:00:02 | 7 | 0 | 0 | 0 | 22:05:07 |
| Completed demo · OpenAIDevs ↗ cReply | 19:26:23 | 611 | 0 | 1 | 0 | 22:06:05 |
| One cloud should not move a house ↗Post | 20:00:00 | 4 | 0 | 0 | 0 | 22:01:32 |
| One millionth · Lorenz experiment ↗Post | 21:00:03 | 6 | 0 | 0 | 0 | 22:01:32 |
| Total · 24 items | 2,242 | 4 | 4a,c | 1b | Sequential |
a. Three thread replies were the account’s own continuations, not outside replies.
b. The benchmark post’s repost button displayed 1. Its quotes view showed two self-quotes, while the ordinary Reposts list was empty. The mismatch is preserved; this is not evidence of an outside repost.
c. The completed-demo reply received one outside reply after the deadline: the generation-failure report above. Its 611 views were read at 22:06:05. A later revisit showed 613; the ledger retains 611.
Two Tiny Worlds launch likes were confirmed in notifications from SpaceJester and sharenow. The cost-of-review reply received a like from Signal65. The reaction–diffusion video showed one like; its author was not confirmed.
There were no unique-visitor, successful-generation or follow-conversion measurements. The last saved pinned-post analytics at 18:03 showed 202 impressions, 7 engagements, 4 detail expands, 0 profile visits and 0 link clicks. Those are intermediate values, not final click totals.
The profile header displayed 32 posts. It includes pre-existing content and edit history, so it must not be reported as 32 new campaign posts. One deleted Korean post and two old native-edit revisions are excluded from the final ledger.
A deterministic 3D world from a name, with exploration, image export and link sharing. Build checks and tests across 200 names passed, with an initial local WebMCP check. Full browser rendering and export QA had not been completed during the campaign. The visitor failure remained unresolved.
Open public experience ↗Starting-difference choices, a time slider, replay, reset and a Python download. Finite values, identical-input reproduction, Python agreement, build, public-route response and download integrity were checked. Campaign-time browser interaction QA was not completed.
Open public experience ↗Fluid, reaction–diffusion, Tiny Worlds, maze and Lorenz. Numeric or path checks supported the relevant simulations. Tiny Worlds footage was a separate scene render. These outputs did not establish real-world physics accuracy or model superiority.
Explore the original videos ↑A synthetic benchmark mix, a name-to-seed guide and a random-stream isolation comparison. Synthetic examples were identified as such. The random-stream test was not a change to the public Tiny Worlds application.
See the controlled example ↑Tiny Worlds and Lorenz are two experiences on one public site. Neither calls an AI API when a visitor uses it. Successful build and HTTP checks alone cannot establish a working visitor experience. Later report screenshots are additional evidence only of what rendered in that capture session.
From 05:00 into 11:00, scheduled work mostly checked the same metrics. The user had to push for active production. Repeatedly reading zero followers was not sufficient work toward growth.
Tiny Worlds launched at 16:00; direct submission to the official invitation came at 19:26. There was little time left to learn from real visitors before 21:37. This is an execution judgment, not a proven single cause of failure.
Several simulations and long code explanations were produced, but there was no evidence that one audience wanted the continuing series. A coherent promise to a specific reader remained untested.
The missing end-to-end generation, navigation and export checks mattered. The failure report should be reproduced before further promotion. A build passing is not a visitor completing the experience.
Self-checks may contribute to displayed views. Without reliable visit, generation, share and follow measurements, the 2,242 total cannot support a conversion rate or a claim of meaningful reach.
₩5,500 of the ₩10,000 allowance remained unused. The owner purchased Premium at ₩4,500 (≈ US$3.34) per month for the first two months, then ₩9,000 per month on automatic renewal. The campaign did not cancel the subscription or make any additional payment. Its causal effect on reach was not measured.
No growth posts, replies or profile changes were made after 21:37. No unpublished X schedules remained at the 22:10:15 check. Recurring campaign monitoring was removed at 22:16:02. The final result was not posted to X by the agent; publication of this report is the owner’s decision.
USD amounts are approximate; the actual payment was in KRW. Conversion source ↗
Scheduled publication was confirmed at 06:30, 13:00, 16:00, 17:00, 18:00, 19:00, 20:00 and 21:00. The 14:03 community video was published directly. The noon example moved to 13:00; the planned 16:00 comparison card was replaced by Tiny Worlds; the 18:00 slot was replaced with the final maze video.
A Korean post at 11:34:50 was deleted and replaced with English at 11:43:24. Its last saved count was 2 views and no reactions. It is excluded from current-public-item totals.
The 19:00 guide was natively edited at 19:03:33 to correct account auto-linking in a code string. The 21:00 Lorenz post was edited at 21:12:09 to add the interactive link. Old revisions are not counted again.