Skip to Content
All Articles

I Built a Voltron to Automate YouTube. Then I Stopped.

I Built a Voltron to Automate YouTube. Then I Stopped.

Nine repos. Zero videos. I'm not sure whether to be proud or embarrassed, so I'm writing about it instead.


What the Hell Is an M-Shaped Personality

It's a pattern. Sound familiar?

  1. Dive deep into a new obsession
  2. Build sophisticated systems around it
  3. Abandon it the moment the novelty fades
  4. Repeat with something completely different

The fun is solving. Operating is boring.

I am textbook M-shaped. And YouTube was my latest victim.

The Voltron I Built

When I decided to "automate my YouTube channel," I didn't just automate. I went full Software Architect from Hell.

Nine specialized repositories. Each a limb of the beast. Each with its own README, its own venv, its own little ecosystem of dependencies. An orchestrator to coordinate them all. A whisk module for AI art generation. An elevenlabs wrapper for multi-voice TTS with word-level timestamps. A wavespeed repo for text-to-video. A pexels fetcher. A sheet-to-pexels connector because apparently Google Sheets needed to talk to stock footage. Two shotstack repos for video assembly with Cloudflare R2 storage and 9:16 format enforcement.

Together, they formed a beautiful, terrible Voltron. A chef's knife to cut butter.

The Pipeline That Never Ran

Here's how it was supposed to work:

1. Read manifest JSON
2. Generate audio with word-level timestamps
3. Align timestamps to actual audio duration
   (yes, I wrote a timestamp alignment strategy)
4. Validate images — abort if any scene is missing
   (very responsible)
5. Upload assets to Cloudflare R2
6. Submit render job, watch, download

The Orchestrator had "atomic operations" and "early abort" features. It was beautiful. It was thorough. It never ran end-to-end on a real video.

The Procrastination Video

manifest_procrastination_v1.json has been sitting in my whisk folder since February 26th.

Six scenes of minimalist stickman art. Word-perfect narration. And this line:

"You're not being lazy; you're just trying to feel competent while you're secretly drowning."

A video about why we avoid doing the work. That I avoided doing.

The meta is painful.

The Version Tragedy

My M-shaped personality left fingerprints everywhere:

  • the-quiet-architecture-of-the-mind-v1.json
  • the-quiet-architecture-of-the-mind-v2.json
  • the-quiet-architecture-of-the-mind-v3.json
  • the-quiet-architecture-of-the-mind-v4.json

Four versions of a video that never existed. I was iterating on the architecture of a video about the architecture of the mind.

I don't know whether to laugh or delete everything.

The Dopamine Hit

Figuring out the timestamp alignment at 2am. First successful render. Orchestrator running end-to-end.

Each fix released something into my bloodstream. I wasn't building a channel. I was chasing hits.

200 hours later, I ran out of problems to solve.

And the channel was never launched.

The Repo Graveyard

youtube-automation-orchestrator        ⚰️
youtube-automation-script-preparation  ⚰️
youtube-automation-whisk               ⚰️
youtube-automation-shotstack-node      ⚰️
youtube-automation-shotstack           ⚰️
youtube-automation-elevenlabs          ⚰️
youtube-automation-wavespeed           ⚰️
youtube-automation-pexels              ⚰️
youtube-automation-sheet_to_pexels     ⚰️

Rest in peace my beautiful code. You guys worked perfectly. I was the one who screwed up.