Table of Contents
Pop the Bubble!

Real talk, fam, we, as marketers, are drowning in AI hype right now.
The Kennan Davidson of AI (founder of icon.me), Sam Altman himself, admitted we’re in a bubble: yes, the tech is groundbreaking… and yes, everyone’s probably overexcited.
Meanwhile, companies are throwing $62 billion at AI this year, but McKinsey found 8 in 10 say it hasn’t moved the productivity needle.
The real question: did McKinsey use GPT5-research mode to figure that out? What do consultants do anyway? Are they just there for plausible deniability?
Even wilder? Abandoned pilots jumped from 17% to 42% in a single year.
BUT DON’T JUMP SHIP!
Translation: a lot of expensive science projects and not much that helps you hit ROAS.
My take on how this speculation affects performance marketing?
Here’s the reality: AI isn’t “push button, get magic.”
It’s more like: “prompt 27 times, curse, tweak, go to therapy, have a coffee, get back to together with your ex, have an anxiety attack, buy a dog and then maybe land something decent.”
There’s also a reason Icon.me has just become an Agency. Not sure how that happened with his myriad of credentials!!
The truth is, AI shines on the edges of performance creative, not in the middle. Think of it like a barbell (buns and thighs!):
Front end → analyzing reviews, clustering personas, finding motivators.
Back end → resizing, swapping headlines, cranking out variations.
Messy middle → strategy, creative leaps, and risk-taking? it u
And since AI has made statics cheap (and sloppy), motion stands out. Even just converting statics to GIFs gives you an edge. (See my Loom on that from last week here).
But enough talk, I am sure many of you want a tool to utilize in the weeks ahead:
Outlining the Flux Kontext + Kling to get amazing b-roll product shots and videos:
We’ve been using Flux + Kling ad nauseam. 40% of what we get back is garbage. But the 60% that worked, when paired with human direction, turned into motion ads that actually stood out in feeds clogged with sloppy AI statics.
Kind of like a recent drive back from New York. We got caught in a storm, pulled off the highway, and stumbled into this quirky B&B in Springfield, Massachusetts. Totally unplanned, but it ended up being the highlight of the trip. Vibes were impeccable and superhuman, man. James actually slept so well that night.
That’s the point: the best stuff happens in the detours. Same with creative. AI sets the stage, but your instincts, the accidents, the inconsistencies are what make the work memorable.
AI is inconsistent. I’m inconsistent. You’re probably inconsistent too. We’re human.
And yeah… I’m writing this late Thursday night, half-tired, fully caffeinated, trying to beat summer CPMs with fresh ideas. That’s the real work.
Here’s the step‑by‑step I actually use for Flux Kontext + Kling. No mystery. Just the workflow.
TLDW:
Get a clean product shot
Skip the messy PDP asset. Pull a flat, well‑lit image from DAM. No fake shadows. No reflections.Decide the idea before you touch AI
Persona, Angle, Motivation. If I skip this, Flux hands me nonsense. I write it out in one line so I can paste it later.Use GPT to structure the prompt
I upload the product image and ask GPT for a tight, JSON‑style prompt I can drop into Flux. I keep it anchored to my PAM line. One safe concept. One riskier concept.Run Flux Kontext
Open Flux. Load the clean product image. Paste the structured prompt. Generate a batch of statics.Cull hard
Most options are studies. I save the 1 or 2 that have a pulse. I crop for 9:16 or 1:1 and leave headroom for copy.Write motion beats for Kling
Back in GPT, I draft a short shot list. Example: subtle condensation, slow camera push, background element, simple can move. Ten words each. Verbs first.Run Kling on the winner
Import the chosen static. Paste the motion beats. Keep it short. Three to five seconds. Export draft.Expect weird
If it is off, I reprompt. I change one variable at a time. Verb, camera, timing, background. Two or three passes usually gets me there.Add copy and variants
Headline space at the top if vertical. Three headlines. Three CTAs. Product tag locked to frame.Export for testing
MP4 for Meta. GIF for cheap motion in feeds. Name files clearly so I can track. Ship the test and move on.
So here’s the play to help balance the scale:
Use AI to move faster on the edges.
Keep the human in the messy middle.
When in doubt? Add motion. Even GIFs beat lazy statics right now.
The bubble will eventually pop. But if you can balance AI’s speed with human creativity, you’ll be the one holding the needle when the bubble pops.
Catch you next week, and have a great weekend!
Best,
Will
