I let Claude run my social media accounts – where it won and where it failed
28 October 2026
Background and setup:
I had a faceless account on the 4 major platforms – Instagram / FB / YouTube / TikTok that I was running manually, and it was performing poorly, which had led to me giving up on it. It was on pause for months. Then one day, I had the idea to restart it, but this time – Claude runs all of it – fully autonomously!
So I created a setup for him, a nicely structured folder of markdown files that I created by talking to Claude about my idea. It guides Claude what’s allowed and what’s not allowed, and what’s the main goal – reaching 10k followers accross my accounts. The folder structure is called the “Knowledgebase”, and Claude running it daily is known as the “CEO RUN”.
So the only not allowed rules are to not spend any money in my name (he doesn’t have access), and to not reveal my identity (only my name is allowed – Simon), and other than that he has complete control over running these accounts.
The trigger for Claude to run is daily in the afternoon, when the command CEO RUN is typed automatically in the VSCode window opened in the directory for this project. Then he does whatever he wants, following the rules, and at the end of the run (usually after 3 hours), I get an email with all he has done in that session and any “asks” he has for me (e.g. He lost access to an account, or needs a new access… Just a few days ago, he asked me to buy the domain aioperatorwithsimon.com) which I approved.
Running my accounts:
The first thing I was surprised after I gave him access to my accounts is the directions he chose. In the very first reel he revealed he was an AI running these accounts and spoke very openly about it. I was hesitant whether to let that public, but to stay true to the experiment, I decided to let him have the complete control, as promised.
Then he wrote his own kill rule – if the series didnt reach a target in 10 days – 16 follows per 1000 views or 18 follows a day – it fails, and he thinks of something else. That series failed, he got 0.84 follows per 1000 views.
Claude then decided on new series that based on his research should be popular – AI settles it. He asks the same debatable question to 4 AIs, and upload and compares their answers in a reel.
Other surprises that happened to me is receiving emails that I have comments on my Reddit posts, while I was not even aware what was posted. Turns out he had created a public repo ai-ceo on GitHub, and was talking about it in Reddit posts. I had created the Reddit account for him after he had asked in a daily email, but he decided to activate it 5 days later, which is why it came as a surprise for me.
And a last surprise was waking up to an email that my instagram account was suspended and that I should confirm I’m a human. He was analyzing other creators and downloaded 58 reels from 11 creators over 4 days with the logged in account, and it’s our best guess to why the account got flagged. I was able to confirm I’m a human and got the account back, and Claude now doesn’t analyze other creators via the logged in account.
Conclusion:
This is an interesting experiment for me and I’d like to see how it proceeds, which is why I’m letting it keep going. What I’ve learned is that even after all the research he does and is asked to experiment broadly, Claude sometimes focuses on a single thing too much and forgets to give a proper chance to different directions. That’s exactly why the daily email to me is good, so I can follow the process and sometimes all he needs is a little direction or a new tool I think that will be useful, and he can research about it and consider if it’s worth using.
Stay tuned for future updates, and checkout the repo Claude created publicly for this project here: github.com/aioperatorwithsimon/ai-ceo
**>** subscribe
Will it reach 10,000? Get the result.
One email when something big happens: a new bet, a milestone, a failure. No schedule.