I taught an AI to make TikToks without me.

Earlier this year I started a small experiment: an AI-generated character called Carl, posting short videos on TikTok.

Eleven plus episodes later, the views are moderate. Carl is not famous.

That was never really the point.

The first videos were about topics I found interesting. Interesting to me, kinda boring for TikTok. So the experiment drifted, and I ended up creating a character that kinda looks like me, building the videos around him instead.

That part was fun.

But the real product was never the videos. It was the machine behind them.

The first episodes were manual work. Write the script, generate the scenes, pull everything into CapCut, cut, sync, export, post. Fun once. Tedious by episode three.

So after every episode I automated the part that annoyed me most.

• First the scripts. Drafted in Carl's voice, I just approve.
• Then the scenes. Each shot generated from the script, with the character kept consistent.
• Then the assembly. The cutting and the editing itself. The part I was sure would stay manual went last.
• Then the posting.

By the last episodes the pipeline could take an idea and come back with a finished, published video. Script, scenes, stitching, captions, upload. Mostly autonomous. I stayed in the loop to approve things, not to do them.

That was the goal I actually cared about, and I hit it: a tool that starts from an idea and ships the whole video on its own.

So where does this leave social content? Sky is the limit now.

An episode costs close to nothing. Anyone can run multiple accounts, multiple characters, in different directions, until something eventually sticks.

That is the game now. I built my engine for it.