Promptly Wrong .ai
Season one/42 stories/3 minutes each

It's usually
not the AI.

Four friends quit their jobs to ride the AI wave. None of them is an expert. In every story a real business walks in with a real problem, and they get it wrong with total confidence — until someone stumbles onto the boring fix that actually works.

The joke is the delivery. The payload is realistic expectations — 42 of them, one per episode.

Read episode 01 → All 42 episodes
For anyone who wants to use AI well, and know what it's really doing · no jargon survives edit
Session transcript ep. 01/scroll ↓
The client
My grandmother grew the tomatoes behind the house in Naples and crushed them by hand. From June to September the kitchen smelled of nothing else. I want a website that feels like standing in that kitchen, aged nine, being handed the wooden spoon.
The CEO nodded the way a jeweler nods at gravel. His gift, he had never said this aloud out of modesty, was compression.
The CEO types
make restaurant website goodsend ↵
AI
Generic pasta stock photo draft A Taste of Tradition. Family recipes made with love.
make it authenticsend ↵
AI
Tuscan-themed stock photo draft Authentic Italian Cuisine. An Authentic Experience. Authentically Crafted, Authentically Yours.
My grandmother was Neapolitan. This is Tuscany. To her these were enemy hills.

What you know by the end of season one

Not how AI works. What to expect from it — which is the part that costs people money.

What it's good at
Bulk text, tagging, sorting, first drafts, translation you can check.

The wins in this series are real and almost always unglamorous — the fix that ships Tuesday and invoices Wednesday. A third of the episodes end in one.

Where it breaks
Facts, sources, memory, arithmetic, and anything that happened lately.

It sounds exactly as certain when it's wrong. You'll learn the tells — and the specific habits that catch them before a client, a judge, or your family does.

When to keep a human
Anything that gets signed, and any sentence that has to be true.

No doom, no hype, no vocabulary quiz. Just calibrated expectations, delivered as jokes you'll repeat at dinner.

House rules

Three of them, running under every story — they're why the jokes hold up as information.

01

The AI behaves like the real thing

Never smarter, never dumber, never magic. Every AI failure here is one you can reproduce yourself. The exaggeration lives entirely in the humans.

02

A third of them end in a win

This isn't an anti-AI show. It works — usually in the most unglamorous way available, discovered by the person least excited about it.

03

The joke lands on the person

Somebody asked badly, trusted blindly, or dismissed too fast. The prompt is the human's contribution — and they are wrong promptly.

The firm

Nobody here is the villain. Everybody here is a little bit the problem.

The CEO
visionary
The CEO
Founder · fluent in jargon

Believes AI is magic and that a four-word prompt makes him a visionary. Never admits a mistake — every flaw becomes a feature. Genuinely kind, but only when disguised as strategy.

The IT Guy
the only dev
The IT Guy
The only real developer

Right about the mechanics nearly always, right about what the client needs nearly never. Buries the correct answer in a flat aside nobody hears until it's almost too late.

The Operator
ships things
The Operator
Ops · has actually shipped things

Not technical, not impressed. If it works well enough to invoice by Thursday, how it works is somebody else's hobby. Keeps landing on the fix that saves the client, for schedule reasons.

The Believer
it understands
The Believer
Client-facing · universally liked

Reads intention and feeling into everything the AI writes, defends it like a friend, and takes it personally when the IT Guy calls it math. Right about people, wrong about the machine.

The Client
a new one each story
The Client
Different every episode

Competent at their own work, new to AI, in the door because everyone told them they need it. States the problem wrong, sits unmoved through the demo, and gets the last line.

Season one

42 episodes, one blog page.

Read all 42 episodes → the blog
The author the author
Who is writing this

I build with these systems for a living. I also watch them faceplant, daily, in ways nobody warns you about.

Every failure in season one is one I have watched happen to a real person with a real deadline. The names are invented. The mistakes are not. This is the AI literacy course I wish existed for my family — no slide decks, no doom, no vocabulary quiz.

By day I run an Applied AI Studio, where organisations pay me to have these arguments before they become episodes. Season one is what happens when nobody calls first.

Founder · no idea what I am doing Meaning explorer Swims in research papers, drowns in the AI wave Have typed "make it better" and lived with it

Where this is going

Now — season one
For beginners

42 stories for people who use AI but don't build it. What it does, where it breaks, and what never to hand it.

Next — season two
For people who build it

Same rules, aimed at the technical crowd: context, evaluation, agents, and the failures that only ever show up in production.

Then — season three
For researchers

Further up the stack — benchmarks, claims, what a result actually proves, and the distance between a paper and a product.

Also in the works
Comics, video, and a season per industry.

The same cast and the same house rules in new formats — four-panel strips, filmed sketches — and pointed at one trade at a time: healthcare, law, retail, logistics, education. Every industry has its own expensive misunderstanding.