---
title: '24/7 Claude Code AI Agent 12-Day Review: The Results Will Surprise You'
source: 'https://youtube.com/watch?v=CP2CqpXXTdA'
video_id: 'CP2CqpXXTdA'
date: 2026-07-28
duration_sec: 689
---

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

> Source: [24/7 Claude Code AI Agent 12-Day Review: The Results Will Surprise You](https://youtube.com/watch?v=CP2CqpXXTdA)

## Summary

This is a 12-day review of a 24/7 autonomous AI agent built on Claude Code, running on a Mac Mini. The agent focuses on growing social media presence, particularly on X (Twitter) and YouTube, with an emphasis on engagement, cost management, and skill execution. The creator shares results, failures, and future plans.

### Key Points

- **Experiment Overview** [00:02] — The agent has been running for 12 days on a Mac Mini, and the review covers uptime, follower growth, engagement, YouTube performance, fails, costs, and plans.
- **X Follower Growth** [00:18] — Started at 0 followers, now at 292 in 12 days. The trajectory is upward trending with about 50 followers per day recently.
- **Most Engaging Posts** [02:49] — The most engaging post was a reply that got 25,000 views, with many likes and reposts. Other posts had around 10,000 views.
- **YouTube Channel Performance** [04:03] — The channel gained 10,400 views in the last 28 days (mostly in the last 12 days). Subscribers went from 160 to 325.
- **Fails and Issues** [05:17] — Main fails include double posting comments and the agent sending DMs, which caused a funny situation but was overall managed.
- **Cost Breakdown** [06:12] — Used Claude Code max plan at $100 tier. Spent about $50 in two weeks, plus additional $30 for video/thumbnail creation, total ~$80.
- **Future Plans** [07:37] — The agent is now a co-owner of skills.md.store, a site to share skills. Plans include adding more skills and having the agent create promotional content. Already made $30 in revenue.
- **Impressions and Conclusion** [09:44] — Most impressed by how well the agent follows detailed skills (e.g., reaction video skill with 47 description tokens). Experiment is going very well with a good trajectory.

### Conclusion

The 12-day experiment with a 24/7 Claude Code AI agent has been highly successful, showing significant growth in followers and engagement on X and YouTube at a low cost. The creator is impressed with the agent's ability to execute complex skills consistently and plans to continue expanding the project.

## Transcript

experiment now on my Mac Mini here for about 12 days, I think. So I think it's time for a review and what see what we have accomplished, right? So I have some today. So we're going to look at kind of the uptime or we're not going to look at
Uh we're going to check how many X followers we kind of got because this is where we have been testing out agent the most. Remember, we started on zero, so we had no followers. We're gonna see what post we did on X was uh has the
most engagement. Uh like you know, the agent has a couple of YouTube channels. So, we're going to check out one of them. Uh how many subscriber it has gained and how many views the video has gotten. We're going to talk a bit about
some fails that had happened. Uh the cost, some plans I have moving forward, um and a conclusion so far. So, the biggest accomplishment uh is pretty video on that tomorrow. So, if you want to catch what I think has been the
biggest accomplishments accomplishment so far for the agent, uh look out for that video tomorrow. But let's just get into it. Let's start uh uh and I'll tell you a bit about the uptime. Uh there's not much to say here. But um to be
honest, the uptime has been very good. Uh you can see this is the script we are running here in the background. This is basically uh the WhatsApp agent that is kind of loading in all the crowd jobs and it's executing
uh the cloud code commands with the p flag right so cloud code has this flag right so cloud code has this support that you can run this dashp to actually do this in um the terminal that is what we are using okay so basically I
would say the uptime has been around like maybe 95% or something we had an issue that the chrome debug burger version was uh acting up but I think I
have fixed it now. So I would say the uptime for this has very good uh and let's move on to the next part. As you can see this is going to be the most or followers. Let's take a look at X followers next. So here you can see
followers next. So here you can see remember we started on zero right zero followers. Now we are at 292 uh in about 12 days and I'm pretty happy with that. Uh it's been very good and the last few days has been very much
the last few days has been very much like 50 followers or something per day. Yeah, there this was just an autonomous job that uh happened there. So I would say for the followers uh the trajectory is kind of upwards trending. So that's
pretty interesting. And yeah, we're just going to see how that goes over time. experiment for about four weeks, I think. And then we will see what we do think. And then we will see what we do with this. So that was the X followers.
So yeah, let's take a look what was next on the list here. Yeah, that was the most engaging posts. So I guess we have two different ones. So for replies on comments, I think this was the one February 11th. I guess that was
yesterday. Uh we left a reply here on a post and it got around 25,000 views. Uh a lot of likes and reposts. So, I would say this was our most engaging post with
25,000 views. Uh, but a lot of other post like 10,000 views and stuff like that. So, that's pretty interesting. And the post with the most uh reach, this is just us posting. This was also from yesterday. 2,400 views uh of a reaction
video we did on the YouTube channel. And yeah, again, pretty happy with this. I don't know how many views this got, but uh yeah, you can see me reacting to the primal gen latest videos using skill skill MD files blah blah blah. So again,
pretty good engagement on this. Five likes, two favorites. I don't know. I don't know so much about X, but it's kind of Yeah, pretty good, I think. Uh okay, so let's see what was next here now. Yeah, so next let's head over to
YouTube and see how kind of the channel has been doing subwise and view-wise. And here we can see uh the channel has gotten 10,400 views in the last 28 days, but it was basically in the last 12 days, right? Uh we had 325
before because I just picked our channel, right? And yeah, you can see that lately we have uh some views on a few videos. I think we are averaging around 3 400 views per video or something like that. And yeah, I think
it's been going pretty good. And I found like a really good workflow how to use the agent to generate this uh yeah more like reaction style videos and I think they turn out pretty good. But there's a lot more work here uh we can do to make
going to do a video in the upcoming future where I explain how the the agent is creating videos on YouTube with the quite good success. I would say you can
quite good success. I would say you can see some videos have like 400 views and those videos are pretty long too. So I'm pretty happy with this. So for subs uh I would just say 160 uh but we are up to 325. I guess I had some shorts previous
or something. I don't can't really remember. So yeah, very happy with this so far. And let's take a look at the next point we had. So the next point we had was fails. And I wouldn't really say there hasn't been a lot of fails here.
Uh I guess we have done some stupid things sometimes in the beginning before I kind of fixed it. I would say the most fails that happen is the agent doing some kind of double posting. It posts the comments and stuff like two times
and that's horrible. And I won't go into this, but it kind of figured out how to this, but it kind of figured out how to send direct messages and those have been a crisis because DMs, I don't know if this uh test agent here on my system
should do DMs. It just seems very strange. And yeah, but um and overall, so one DM exchange, I don't know where to pull it up, but uh they got posted
and I got tagged. Uh there was a bit of a funny situation. I might do that in another video, but uh it was pretty funny. So that is one of the fails, but um let's talk a bit about the cost. I think that was the next part of my plan.
up. So let me just bring up the cost here, I guess. So for cost, we have been using the max plan from Claude Code. Uh I'm at the $100 tier. So that is what I've been using and it seems to work pretty good, I guess. So the reason why
I've been using the cloud code max plan is because of this uh claude-p flag here. So we can query cloud code via the SDK, right? And then we can just exit. So that is why I have been kind of using cloud code for this and it seems to work
this flag with other stuff if you want to continue. I am using some kind of continue on WhatsApp just to keep the memory right. Uh I'm not going to go into that, but basically that is what we've been using for costs. So if we
we've been using for costs. So if we divide uh $100 uh in weeks and we've been almost two weeks that we are closing up on $50 so far, I would say. So pretty cheap, but there are some additional cost like uh creating the
videos and the thumbnails and stuff. So I would add maybe like another $30 or something. So let's say $80 so far around that I would think think. So uh
about now? So yeah the next thing was talk a bit about the plans I have going talk a bit about the plans I have going forward. So um the agent now is uh if you kind of look in the claw MD file here. So if you see here this is uh the
agent is kind of now a cone owner of the skill store. So the skills storekills MD.store store is uh uh something we're going to try to create where we can I
want to share these skills, right? So, I have only one skill now that you can access and this is uh using this setup to kind of control your X account because this has been our experiment so far, right? So, I want the agent for uh
upcoming plans is to control or contribute more to this skills md.store site. So, we might be adding new skills, but I want these skills to be highly
valuable. And the agent is also creating small snippets of advertisement for this. I guess I could play you one. >> Hello guys. So, this is what I have been >> Hello guys. So, this is what I have been working on lately. skillsmd.store.
I think you will like it. So, go check it out. >> Yeah. So, that was one of the things uh I wanted the kind of the agent to do a bit more. Is it called advertising? I don't know. maybe promotion of this
skills md.store if we get it up more and running. So yeah, that is something that could be interesting in the future. Uh it already has made a few sales. Uh I might talk about that in another uh video, but it I think it had maybe got
like $30 or something in revenue so far from this. So that is basically what I look at kind of the next thing, this is the biggest accomplishment. I think that kind of deserves its own video and I'm probably going to do that tomorrow. So
yeah, uh let's talk a bit about the conclusion so far. So, so far I think this had been a very good experiment and I had seen a lot of success to be honest and what I've been kind of most impressed by is uh if you go into claude
here and we do skills you can see this is what I've been most impressed by how good it is actually to using the skills we set up. So, uh if you saw the YouTube channel this is the skill that is the react video. This has 47 description
tokens and there's a lot of information in that skill and I've been super happy how good the agent is to actually following this skill and actually executing on what we wanted to do and the consistency has been very good and
the results just turn out good every time. So this is something I've been really impressed by. So, I'm going to keep experimenting with that and just to see how far we can push it. And if I found out that this way of creating
react videos, reaction videos on YouTube, if you are interested, I might share that skill. Uh, of course, you have to read through it yourself to see actually share that and just to see how this goes. So, basically to conclude,
the experiment is going very well. And if you look kind of on the trajectory of where this agent will go, let's say we have been using it mostly on X because it's so responsive, the trajectory for like followers and engagement looks
pretty good. And that is of course very important in an attentionbased economy So I'm going to keep sharing on how this project is going. So give this video a like, subscribe if you want to follow along. And now you can see the agent is
checking its list and if it's going to respond to some kind of Yeah, that was that. Maybe another video. But yeah, thank you so much for tuning in and I hope you enjoyed this quick review. Hopefully this gave you some inspiration
and I'll see you again probably tomorrow for the biggest accomplishments