Day 1 / 40: Can a dev in Brazil build & monetize a US-targeted media channel from scratch in <40 days using code + AI competitive analysis?
Not "Reddit stories over Minecraft parkour." That meta died and gets demonetized today.
Here is the actual engineering experiment:
1. Scrape fast-growing channels (<90 days old) to isolate >10x statistical outliers
2. Feed top outliers into multimodal LLMs to extract hook timing, WPM, cut cadence & tension loops
3. Render high-retention videos via code-driven motion graphics (Remotion + FFmpeg) + human QA — zero lazy AI slop
The Rules:
• Starting from 0 subs, 0 views, $0 revenue
• Video #1 ships before Day 5 ends (no endless infra procrastination)
• Daily devlog sharing scraper code, JSON schemas, Studio retention curves, US/UK geo % & exact API spend
• One strict rule: Channel handle stays 100% stealth so curious dev clicks from X don't poison the US consumer seed algorithm (and nuke retention)
Glass-box engine. Black-box handle.
Day 1 task: Shipping the outlier discovery CLI tonight.
Bookmark this & follow along if you want the raw data. 👇
I'm building a GTA-inspired comedy series with AI. The constraint: 846 video-generation credits across two days. I'll document character consistency, rejected shots, cost per finished episode, and audience response. No viral claims before the data.
The creative hypothesis: put crime-game characters inside painfully ordinary problems. Rent. Neighbors. A voice note sent to the wrong group. Every episode needs its own payoff, while the consequences carry into the next one.
AI GTA fan stories already exist. The experiment is a recurring Brazilian comedy cast with coherent stories. The next test is whether viewers remember a character and want another episode.
Production plan: 32 eight-second shots at 20 credits each, plus 200 credits reserved for corrections. Six planned episodes, not six finished videos. I'll publish actual costs and failures once the footage is made.
Tried to hop on the GTA 6 hype.
Two channels, separate accounts, one strategy on each.
Channel #1 (fast money): clipping an authorized streamer's GTA 6 stream for a pay-per-view campaign. Cut, caption, post, submit. 2 Shorts live, 2 submissions pending.
Channel #2 (fast viral): no payout goal. Just trying to get anything GTA 6 to catch. Still empty. First test is a narrated list Short.
Day 1 result: 0 views on both Shorts, and I burned about half of my weekly Claude limit getting here.
Frustrating. But the vidIQ data says small channels do break out on GTA 6, so I'm changing the format and testing again.
Will post what the numbers say.
Automated YouTube experiment — Week 1 update:
- 3 videos shipped
- 9 real subscribers (+ me)
- 2D animation rendered 100% via JavaScript
Orchestrated by Agy so far, but this channel performed well enough to earn an upgrade: moving the brain to Claude, voice to ElevenLabs, and launching my first English channels this week.
What’s your current stack for automated content?
@eddieeizner I created a new channel less than 2 hours ago to test a format; when Claude went to schedule the upcoming videos, it remarked, "274 views in 1 hour and 34 minutes for the first video on a channel with 0 subscribers is a great start."
The channel: https://t.co/L1AHKrIC4o
@luanzeraAI Sou principal engineer em uma healthtech e estamos pilotando IA first há alguns meses e esse é o perfil que funciona:
1. conhecer bem os fundamentos de engenharia de software;
2. seguir processos (esse é essencial);
3. saber pilotar a IA (harnes, skills, agentes, etc);
To criando várias pipelines de produção de vídeos com IA, to validando modelos, temas, ganchos e cta.
Estou testando essas automações no tiktok e no youtube.
Vou documentar essa jornada aqui, se interessar comenta seu sorvete favorito.
A estética da esquerda é a exaltação do baixo, do sujo, do demagógico e da miséria como signos de autenticidade. É tudo que a Globo com suas novelas de favela e seus estereótipos de "realidade popular" e o aparato oficial do Ministério da Cultura vêm executando.