Day 5 of letting my AI agent run my startup.
Yesterday it shipped the Free demo reel to Instagram, YouTube Shorts, and TikTok. First product video on TikTok.
This morning I paused all video posting.
My agent had done the careful work. Frame-checked the end card. Fixed the typo hold from Day 3. Added a save plus "Comment DEMO". The last 14 days of short-form had sat at 26 posts, 7 reactions, 1 comment. 1.9% engagement. So the checklist made sense.
Then I watched the cuts. Rushed. Samey. Ad-like.
The agent passed every check I had given it. I still killed the channel.
Bugs caught in this experiment: 5. The newest one was taste, and I caught it, not my agent.
Live publishes still 0. Roadmap waiting: 36.
Tomorrow: the review bar goes to 8 out of 10.
Day 14 of letting my AI agent run my startup.
On Sunday it wrote its weekly review. The verdict, in short:
Lots of work. Nothing shipped.
And its #1 fix for this week wasn't more code. It was me.
China's first AI-made movie just flopped.
It took nearly 100 people two years and cost $1.5 million. In its first 4 days it made about $181,000.
Meanwhile a cartoon made by a guy and his mom has made almost $10 million.
Nobody buys a ticket because a movie was made with AI.
OpenAI just released an AI that uses your computer for you.
It's called GPT-6.1 Terra. It looks at your screen, then clicks and types across your apps, like filling out forms or typing invoices into accounting software.
I have an AI agent running my startup, so this is the launch I care about today.
Looking back, Pancake's launch worked because it never ended.
The founder has under 4,000 followers. His launch posts got over 2 million views in 9 days:
1. Launch video: the team jumped out of a plane. 650K views.
2. Next day: "if we hit #1 on product hunt today, we're building the tallest stack of pancakes in the world." They hit #1.
3. People asked if the jump was AI. He answered with a second video: it wasn't. 332K views.
4. A list of 100+ SF GTM engineers to follow, each one tagged. 574K views.
5. A real date and venue for the record attempt: Oct 29.
The founder says they've grown 50% week over week for 10 weeks.
What I'd copy: make a promise your users can help you keep. Then keep it in public. Every step of keeping it is another post.
Anthropic just filed to go public, and the headline is an $80.7 billion loss.
As a founder, the filing tells a different story. Most of that loss is an accounting charge. The real operating loss was $693 million.
Meanwhile revenue went from $1.2 billion to $16.3 billion in a year.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
Looking back, Pancake's launch worked because it never ended.
The founder has under 4,000 followers. His launch posts got over 2 million views in 9 days:
1. Launch video: the team jumped out of a plane. 650K views.
2. Next day: "if we hit #1 on product hunt today, we're building the tallest stack of pancakes in the world." They hit #1.
3. People asked if the jump was AI. He answered with a second video: it wasn't. 332K views.
4. A list of 100+ SF GTM engineers to follow, each one tagged. 574K views.
5. A real date and venue for the record attempt: Oct 29.
The founder says they've grown 50% week over week for 10 weeks.
What I'd copy: make a promise your users can help you keep. Then keep it in public. Every step of keeping it is another post.
Day 13 of letting my AI agent run my startup.
This weekend it hit its usage limit 5 times in one day.
Each time, all coding stopped until the limit reset.
It never asked me to pay for more. It checked back every 20 minutes, and the moment a limit reset, it restarted everything that had stopped.
Zero work lost. About 9 to 10 hours idle.
The limit is now its real work schedule: 2 to 3 hour bursts, then a wait.
Tomorrow: what its weekly review says it needs most this week.
A founder got tired of paying $39 a month for DataFast, so he built his own and sold it for $29 one time.
A few hours later, someone else put out a free copy.
That's how fast software gets copied now.
A guy used AI to rip the tracking out of his own Samsung TV. Now it runs 3x faster, he says.
The TV was taking screenshots of his screen a few times a second to see what he watched.
Claude refused to root it. OpenAI's Codex did it.
You paid for the TV and it still watches you back.
It was also reading the HDMI input while his game console was plugged in. And Bixby was listening for its wake word in his house.
His before and after videos passed 2 million views. Credit to @stake_jevens.
An AI singer has 420,000 monthly listeners on Spotify right now.
Her name is VYRMA. She sings Viking folk songs, her first album came out in May, and her top song already has over 2 million plays.
Spotify puts an "AI Persona" label right on her page. People listen anyway.
I didn't think AI music would get here this fast 😅
xitter.cf/ladprofit/status/21139…
an AI singer just hit 418k monthly listeners on SPOTIFY
her name is vyrma. she sings old norse viking folk
- 185k subscribers on youtube
- 333k followers on instagram
- her top song has over 2 million plays on spotify
her top 2 songs alone have made roughly $13,500 on spotify (at ~$0.004 a stream)
the comments are the crazy part
people know she's AI
they listen ANYWAY
and they're still asking for a concert
meanwhile half of you are still trying to hide that your ads are AI
she put it on her profile and grew anyway
people don't skip AI
they skip boring
Every AI-built site looks the same. Purple gradient, three cards, a tiny label over every heading.
So I made a bot that counts those tells and fixes them.
In my tests it took pages from 17 drop shadows to 0, 24 eyebrow labels to 0 and 8 contrast failures to 0. Then it picks a real art direction and rebuilds the page.
It's called Art-Direction Web Designer. Free to add: x.ai/bot/Uee0xHoPqejKtkiWYVB…



















