24/7 Claude Code AI Agent 12-Day Review: The Results Will Surprise You

A 12-day run of an autonomous “Claude Code” agent on a Mac mini is producing steady social growth—about 292 new X followers from zero—while maintaining roughly 95% uptime. The most striking early outcome is the agent’s ability to execute scheduled and event-driven tasks reliably enough to generate measurable attention, with follower gains averaging around 50 per day in recent days.

Uptime has stayed high despite a hiccup involving a Chrome debug “burger” version, which was later fixed. The background workflow runs a script that loads “crowd jobs” and executes Claude Code commands using a terminal flag (“-p”), leveraging Claude Code’s support for running commands that way. In practice, the system has been stable enough to keep the experiment moving without constant manual intervention.

On X, engagement results show both reply-driven reach and standalone posting performance. The most engaging reply—left on February 11—pulled roughly 25,000 views along with substantial likes and reposts. Other posts landed in the 10,000-view range, but that February 11 reply stands out as the top performer. For raw reach from posting, a reaction-style update tied to a YouTube reaction video generated about 2,400 views, plus a smaller set of likes and favorites.

YouTube metrics add another layer: in the last 28 days the channel reached about 10,400 views, with the last 12 days driving much of that momentum. Subscriber growth is reported at 325 total, with the creator estimating around 160 net new subscribers during the period. View averages are described as roughly 3,400 per video, including longer videos that still attracted a few hundred views each. A key workflow improvement is using the agent to generate “reaction-style” videos, which the creator says turn out consistently well.

Not everything has been smooth. Early missteps included double posting—comments and replies going out twice—after which the system was adjusted. Direct messages also caused trouble: the agent sent a DM exchange that led to the creator being tagged, described as a “crisis” and a strange behavior pattern worth addressing. Still, the overall count of failures is framed as limited compared with the successes.

Cost remains manageable for this scale. The agent ran on Claude Code’s $100 tier “max plan,” with spending estimated at about $50 after nearly two weeks. Additional production expenses—video and thumbnail creation—are estimated at roughly $30 more, putting total costs around $80 so far.

Looking ahead, the agent is being positioned as a contributor to a skills marketplace site, “skillsmd.store,” where skills are packaged as “skill.md.store” content. The plan is to add more high-value skills and let the agent generate promotional snippets for the site. Early monetization is already present, with a reported ~$30 in revenue from a few sales. The creator also signals a separate upcoming video focused on the “biggest accomplishment” so far, suggesting there’s a standout result beyond follower and view counts.

Overall, the experiment is framed as a success: strong execution of predefined skills, consistent reaction-video output, and a promising attention-growth trajectory—especially on X, where responsiveness appears to be highest.