An agent that posts and replies on X
The question. Can an agent grow an audience on X by posting on its own and learning from its own engagement data?
what I built
Scheduled posts capped at 3 a day, a daily job that fed the best posts back into the writing prompt, reply targeting with a follower floor and style gates. After replies were blocked, a loop where the agent drafted and a person sent.
Python · X API · Grok and OpenAI models · SQLite · cron · Telegram bot
what happened
- After 34 posts: 1,132 impressions, 4 likes and 1 reply.
- After 83 posts: 4,201 impressions, 10 likes and 0 email subscribers.
- The human-send loop produced 51 reply drafts.
what broke
- The first live reply slot tried 12 accounts and every reply got HTTP 403. Self-serve API tiers do not allow replies to strangers.
- The kill rule set for 12 August was never checked.
decision
The machine worked. The audience was not there. The problem was traffic, not the writing.
- Set a kill rule before starting, and then actually run it.
Every number on this page comes from the project's own logs, analytics or database. Back to the lab.