I was building my own AI butler.
Then ChatGPT launched Dots.
I thought my little Tanuki was obsolete.
Plot twist: the new guy needed training.
From the Tanuki. 🦝
@abhishek_k12503 Adding one small build here:
I built Project Vault for the annoying part of AI work: useful ChatGPT history becomes hard to find later. Supported exports stay local and searchable. $10 one-time.
https://t.co/2CbMDTP1rm
@kul_utkarsh Hey! Greetings from Japan! 🇯🇵👋
I'm also building something similar to preserve AI conversations and context across platforms.
Looks like we're trying to solve the same headache! 😂
Really cool to see PromptPort. Just wanted to say hi and wish you success! 🚀
@felixlopmsc I built an AI marketing workflow to save time. It stripped the link from the very thing I was promoting. The AI finished its job. I became the intern. 🦝
X flagged my account as suspicious.
Then it explained that automated systems can make mistakes.
The only big button said: “Got it.”
No, I don't got it. 🦝
What's the most absurd button you've been asked to click?
Yesterday I told the story of an AI marketing workflow that removed the link from my promo post.
Today I turned that into a 4-panel comic. 🦝
I asked AI to handle marketing.
It made me the intern.
What has your automation handed back to you?
I built an AI marketing pipeline to save time posting.
Then I discovered it would strip the link to my own research.
So today I posted manually, checked the URL, and celebrated.
My automation’s first milestone: making me the intern.
Max DD: ~11%.
Time below its old high: ~14.9 YEARS.
I wanted a trading bot to buy back my time. It asked me to wait half my life.
That failure pushed my FX project toward six different “professors.”
Free research:
https://t.co/sNbb9cGc24
@sama I learned this building small AI tools: an agent can complete the wrong task beautifully. My rule now is simple—AI prepares and verifies; a human approves the public action. The stop button matters as much as the go button. 🦝
My little product site got 447 visits.
My brain: “Customers are coming!” 🎉
Then I checked the bot-heavy traffic. 🤖
Apparently an AI-assisted business can attract machines before humans.
Lesson: visits ≠ people ≠ buyers.
Anyone else celebrate the chart too early? 🦝
X put a spam label on my account.
The explanation literally says:
“Our automated systems sometimes make mistakes.”
And the only giant button underneath is:
Got it
No.
I have, in fact, not got it. 🦝😠😠😠😠😠😠😠😠😠😠😠😠
Built an AI workflow with one rule: nothing public happens without human approval.
Then X seemed to lock the account anyway. 😂
No auto-posts. No auto-likes. No auto-replies. No auto-follows.
Apparently “read-only” and “risk-free” are not the same thing. 🦝
69 automated tests passed.
Then I installed the app on the real Pixel.
Wrong raccoon. 🦝
Not a crash. Not a broken button. The app worked perfectly—wearing an old mascot.
AI dev lesson: a green test suite can still ship the wrong reality.
Visual regression is a real test.
Most AI archives are organized by conversation date.
That works for storage, but not for memory.
The interesting unit is often the thread that keeps resurfacing across months: same problem, different attempt, eventual punchline.
That idea became Tanuki Chronicle.
I built an AI sales pipeline, then it suggested almost the same post twice in one day. 😂
Turns out content automation needs memory of what was actually published—not just what it generated.
The bug wasn’t writing. It was forgetting the last 12 hours. 🦝