Hi Reader -

I was doomscrolling the other day on LinkedIn.

Everyone was going on about RIP media buyers, RIP designers.

…and how to write better prompts. how to get better outputs.

Utterly fascinating stuff.

All that stuff LinkedIn thought leaders love to post about between their humble brags about revenue milestones.

And I noticed something that made me laugh.

All of the AI prompting advice was also just… management advice.

The exact same stuff people have been saying for 100+ years.

So I started running a little test.

I call it the AI manager test.

1. Take any piece of AI advice.
2. Swap “AI” for “your designer” or “your team.”
3. See if it still holds.

Turns out it always does. Every single time.

“Give clear instructions.”

“Provide context.”

“Show examples.”

“Explain the why.”

Groundbreaking stuff, right?

That’s not AI advice. That’s “how to get good work out of literally anyone” advice.

No shoot, Sherlock.

Here’s the fun part.

The people who are bad at managing humans?

Also terrible at getting good output from AI.

They just blame the tool now instead of the employee.

“AI is overhyped.”

“It just gives generic outputs.”

“It doesn’t understand what I want.”

No. Your brief was garbage. The AI is just honest about it.

Here’s what actually works.

For AI. For designers. For anyone making creative for you.

1. Build a reference library of performant ads.

Most people fill the swipe file with ads they like. Ads that won awards. Ads that made them feel something.

That’s not the point.

The swipe file is ads that actually scaled or had longevity for audiences like yours.

Doesn’t matter if you think they’re ugly.

Doesn’t matter if they’d never win an award.

If they scaled, they’re in. If they didn’t, they’re not.

2. Define the zone visually, not verbally.

Saying “not too polished, not too rough” is useless.

Everyone nods. Everyone interprets it differently. Everyone makes something different.

What works: show them three actual ads. Give them visual references.

“This one is too polished. Looks like a brand made it.”

“This one is too rough. Looks broken.”

“This one in the middle? That’s us. That’s the zone.”

Now they can see what they need to do.

3. Give rules, not feedback.

Feedback doesn’t scale.

You can’t give notes on 200 ads a month. You’ll lose your mind. Your designers will start ignoring you anyway.

Rules scale.

“Hook lands in under one second.”

“Text on screen frame one.”

“No logo intros.”

“No slow builds.”

“If it looks like a brand made it, kill it before I see it.”

Now they’re not waiting for your opinion. They already know.

4. Give them the graveyard.

Everyone shares winners.

Nobody shares what’s already dead.

“Urgency hooks. Tested them four times. They tank with this audience. Stop trying to make urgency work.”

“Transformation messaging. ‘Become the best version of yourself.’ Tested it to death. Our audience doesn’t buy aspiration. They buy relief.”

Now they’re not wasting cycles on angles you’ve already burned through.

The graveyard is just as valuable as the swipe file. Maybe more.

5. Give them specifics, not your adjectives.

“Make it more relatable” is the kind of feedback that makes designers puke.

Here’s what actually works:

“She’s 34. Feels like everyone else figured it out and she’s still behind. That low-grade anxiety she can’t shake.”

Match her language, not yours.

---

Here’s the real point.

The bottleneck is never the model. It’s the quality of what you feed it.

This is why we grew our creative team more than 2x this year. Even as we did more with AI.

More AI means we need more people who are great at this stuff, not fewer.

If you spend $50k+/month and want to work with a team that’s obsessive about this, people who know how to brief and direct so the output actually converts, hit reply with ‘AI’ and let’s talk.

Or don’t.

I’m sure “make it more relatable” will start working eventually.

Later,

Shamanth
​
PS: some goodies from our socials below.
​
​The Rise of Quiz Funnels
​​
​Can your Meta ad account scale past $100k/month?​

​A millennial mom in San Francisco who just had a baby and wants to lose belly fat.​
​
​I just ROASTED Grammarly’s ads​
​
​Steal the exact decision framework I used at Zynga​

​In which I receive some management wisdom from Claude Code.​
​
​Tinder cracked a pricing model that didn't matter for most apps.​
​​
We went from $500 to $10,000/day in a few weeks.​

​The best-performing subscription app ads don't talk about features, benefits, or savings.​
​
​