AI Experiments: A $1K Business Attempt and YouTube Uploads on Autopilot

Codie Sanchez tried to build a $1000 business in one day with ChatGPT Operator. Swyx had Claude upload his YouTube videos while he watched. Both learned something.

Last updated: February 11, 2026

TL;DR

@Codie_Sanchez tested ChatGPT Operator for marketplace flipping — it found items, messaged sellers, scheduled couriers. Not fully automated, but “changed the game.”
@swyx had Claude Cowork upload 4 YouTube videos autonomously — including trimming silences by click-and-drag.
The pattern: AI agents doing real computer work, not just answering questions. Messy, but working.

Two experiments. Two honest reports. Neither was perfect. Both were impressive.

@Codie_Sanchez tried to build a $1000 business in one day

The challenge: Can ChatGPT Operator bootstrap a business from scratch?

She pointed it at Facebook Marketplace flipping — buy low, sell high. The AI:

Found free or underpriced items
Messaged sellers and set up pickup times
Scheduled couriers to deliver items
Researched resale values
Created new listings

“At one point I even watched it do a Captcha saying it wasn’t a robot 👀”

Her honest assessment:

Was it fully automated? No.
Could it scale to 7 figures? Probably not.
Did it change the game? Absolutely.

The AI handled the tedious parts — research, messaging, scheduling. She handled the judgment calls.

→ Source

@swyx had Claude upload his YouTube videos

Not a simple upload. A 9-stage autonomous plan:

Scan files for 4 Zoom recordings
Watch the videos (analyze selected frames)
Open YouTube to the right channel
Upload each video
Title and describe them
Trim the silences — by click and drag
Drive the whole plan to completion

He watched it happen. Claude controlled his computer, navigated YouTube Studio, and edited videos.

“If you are not using this for a routine knowledge work task once a day you are probably miscalibrated for HOW MUCH BETTER THIS THING IS”

The key insight: he could interject and change plans mid-way. When he underspecified something, he just… said so. Claude adapted.

→ Source

What these experiments reveal

AI isn’t magic. Codie’s business experiment required constant human judgment. Swyx had to review irreversible actions manually.

But the mundane parts — the clicking, typing, searching, scheduling — that’s increasingly handled.

The shift: AI moving from “answer my question” to “do my task.” Messy now. Getting better fast.

What boring task would you hand off if you could?

FAQ

Did the $1000 business experiment actually work?

Partially. The AI found items, messaged sellers, scheduled pickups, and created listings. But it wasn't fully automated — Codie had to intervene. Her honest take: 'Could this scale into a 7-figure business? Probably not. But it definitely changed the game.'

What is Claude Cowork?

Anthropic's computer use feature that lets Claude control your computer — clicking, typing, navigating apps. Swyx used it to upload videos to YouTube, including trimming silences by click-and-drag.

Should I be using these tools daily?

Swyx thinks so: 'If you are not using this for a routine knowledge work task once a day you are probably miscalibrated for HOW MUCH BETTER THIS THING IS.' Start with something boring and repetitive.