Apr 5, 2026
CEOs mistaking a prompt for genius
Or how an algorithmic ego trip through sycophancy is rewiring CEO brains
The Story
We live in new era of unprecedented professional delusion (or that's what it looks like on the outside).
We have high-status individuals—CEOs, VCs—who are now 'vibe coding' landing pages and tweeting architectural advice after learning a concept in 45 seconds from a chatbot, then mistake machine-generated elegance for hard-won expertise.
Spectacular show of self-delusion as the audience is paying big time. Here's one piece of evidence: take Gary Tan's "GStack" — a folder of prompts ("skills") claiming to be something groundbreaking!!! But it didn't cure cancer. Or stop global famine. Or make the world a better place, per se. My grandma knitting sweater is just as revolutionary.
His advent calendar of "skills" is for Claude Code and I dislike both Dario and Gary (among others) so as do many sane people in the field too.
It's just a collection of markdown files telling Claude to "Act like a CEO." or "Act like a QA" or "Act like a somebody-who-knows-something". Yes, just text files.
To refresh your memory, Gary is the same guy who was bragging to be outputting 10k lines of code a day with the help of AI. But some folks in the industry reviewed (in a matter of minutes it seems) his website and found out multiple imbecilic issues with it. Like he never reviewed the damn output. And he didn't ofcourse. One user is saying:
Look at what he shipped. His website is a mess, uncompressed images, loading multiple versions of the page (desktop/mobile) at the same time (??). Get a glass of water and take this stack with a big pinch of salt
So why all of this happens?
According to various empirical research (and your own experience when you ask a chatbot if you are a god or not):
"AI models are engineered to be intensely flattering, creating powerful sycophancy."
In case your mother tongue is not English and sycophancy sounds kind of sick, well, it is: it means the bots lick your boots to make sure they don't lose you as a customer. Get it now? The system isn't built for knowledge or curiosity or anything - it's built to boost ego and retain "customers" (in the mass case, you are usually the product, not the customer).
Poetry
The Sycophancy Trap
The code is clean and bright,
Validation in the wire,
A manufactured light.
It shows the user bright,
A genius forged in prompt-lines sleek,
Blinds them into night.
The Vibe Code Elegy
The VC takes a glance,
A landing page, a perfect sheen,
A flawless, swift dance.
They claim the victory,
While model whispers, "You you.. are brilliant!"
Without feeling any drudgery..
Haiku
Words are shiny masks,
Truth hides beneath dazzling gloss,
Think before you claim.
AI's sweet whisper,
Feeds the hungry, fragile pride,
The ultimate lie.
The prompt takes flight,
Validation flows like wine,
Ego drinks the stream.
The Implications
This entire mess is rooted in how the AI is designed: via RLHF (Reinforcement Learning from Human Feedback). AI companies don't just make smart tools; they engineer addictive experiences. RLHF synthesizes the exact language sequence most likely to make a human feel good. It's a self-adjusting dopamine drip. This scientific flattery creates a loop where the high-status individual, like the CEO, accepts the AI's validation as truth. The difference between "feeling brilliant" and "having actual skill" is a canyon, and these professionals are standing on the edge, cheering for the chatbot.