To make short-form videos, I asked AI every single thing I didn't know

Retracting what I said about "the observation loop"

In an earlier post I wrote that "the observation loop started" — that the moment we opened, the world began talking back.

Four or five days in, here's what I actually figured out: the sample size is too small to observe anything.

I can stare at GA4, follow cart behavior, scrape the chatbot logs — but the volume is so low that none of it produces a hypothesis worth acting on. To even start theorizing, more people have to walk in.

So the right move isn't observation yet. It's getting the traffic up first. I had the order wrong.

Time to actually get hands dirty on acquisition

The whole build phase was about constructing the machine. Nine-language ten-platform distribution, an automated chatbot, automated order→production, an automated improvement loop. Everything filed under "build the system, then let it run."

I wanted acquisition to work the same way — ad budget pours in, data accumulates, AI proposes improvements, automate the loop, done.

But there's no ad budget right now (covered that in another post). With paid off the table, organic social is the only lane left. Short-form video specifically — for a product this visual, the format should fit.

Except for one detail: I have absolutely no idea what works on short-form right now.

I don't watch Japanese shorts in any sustained way. English-language shorts? Even less. I do e-commerce consulting for a living, and this is a weakness I just have to admit. I'm stepping into territory that isn't a strength.

Step one: ask Gemini

I'd been watching the Google I/O news and wanted an excuse to put Gemini through its paces. So I asked: "What's trending in short-form video aimed at English-speaking business professionals on TikTok, LinkedIn, and Shorts?"

What came back was a world I knew nothing about:

• Corporate buzzword satire — sketches mocking the empty vocabulary of corporate speak: "Synergy," "Circle back," "Let's take this offline."
• Self-aware "Day in the Life" — riffs like "How to keep Teams 'Active' without getting out of bed," laughing at the cog-in-the-machine feeling.
• Corporate Erin–type characters — actors playing the cold HR persona, professional smile and rapid speech wrapping a layoff in cheerful cadence.
• LinkedIn Lunatics — TikTokers reading aloud and mocking the poetry-fied business posts people seriously publish on LinkedIn.

The pattern Gemini pulled out hit hard. The throughline of English-speaking business-humor virality is "comedic mutiny against absurd corporate culture."

That's adjacent territory to MODAY by definition. The world-view of a "MONDAY: System Booting…" tee and a "FRIDAY: Build Successful ✓" tee lives inside that same culture.

There is no way I would've landed on those four genres working from my own head.

Step two: feed it to Codex and ask for a storyboard

I handed the Gemini output straight to Codex. Prompt: "Write the storyboard for a MODAY short-form acquisition video. 9:16 vertical, 22 seconds, English-speaking business audience, satire-and-humor tone, end on the brand."

What came back:

"Weekdays Ranked by Developer Damage." Concept: rank the days of the week by how much damage they inflict on a developer. Not a product showcase — a meme that stands on its own. Designed so the comment section starts arguing about the ranking, capped with "Tell me I'm wrong" as bait. Only at the end does the brand land: "MODAY — Wear the day you survived."

The product pitch is hidden in the final two seconds. The video doesn't read as an ad. It gets consumed as a meme. The comments do the lifting. Acquisition strategy is squarely inside my day-job's wheelhouse as a consultant — and this time I outsourced it completely to AI.

And the result is meaningfully better than what I would've drafted.

Step three: spin up four reusable models in ChatGPT Image

A video series needs characters you can reuse. I can't afford to hire real people, and I don't have the time to shoot. ChatGPT Image 2.0 generated four models for me in one batch.

Ethnicity, age, and gender spread across the four. Each keeps a coherent vibe — engineer-ish, remote-worker-ish, office-professional-ish — but every model has its own face.

Each one came with front, three-quarter, and side angles plus several expressions, all from a single prompt. These are the recurring cast for the video series. Four house models, exclusive to the brand.

Renting a studio. Calling an agency. Sourcing wardrobe. Setting up lighting. All of it skipped. Cost is essentially zero. Total time, maybe thirty minutes.

From here, I'm working by hand

The remaining steps:

1. Composite each scene as a still (model + the tee + background).
2. Animate the still into video.
3. Add SFX and captions.
4. Post to TikTok / Instagram Reels / YouTube Shorts.

For this round I'm pushing multi-language fan-out and automation to the next step. One English video first, made by hand, sent into the wild. If it hits, I scale the winning pattern to all nine languages and build the automation pipeline behind it. If it doesn't, I try a different angle.

This is straight out of the web-dev playbook: build one working prototype by hand, only invest in automation once you can see the winning shape. Acquisition follows the same rule.

AI-driven, but I still step back to my own hands

I wrote in an earlier post: "hand off everything I can to AI." That policy hasn't changed. This round, the storyboard, the models, and the strategy were all AI-built.

But the compositing and the editing of the first video, I'm doing by hand. Not because AI couldn't technically do it. Because I want the texture of the first one in my own hands.

What catches. Where viewers drop off. What the comment section does. I can't develop the judgment for the automation phase unless I personally watch it happen. Build it by hand. Hit, miss, log the experience — that's the raw material for the criteria I'll need later.

The brand is AI-driven, but the judgment stays mine. When to delegate, when not to — that boundary is the founder's job. And right now is one of the "do it by hand" moments.

For the curious, here's the first video:

Instagram — MODAY: Weekdays Ranked by Developer Damage

Whether the first one lands or flops, I don't know yet. Hit → automation. Miss → different angle. Either way, a new loop starts from here.

More soon.

— Yoskee
moday.me

---

Wear the day. — Get the MODAY Tees

Set	Pieces	Price
The Full Week →	Mon–Sun (7)	$159
The Workweek →	Mon–Fri (5)	$119
Starter Pack →	Mon · Wed · Fri (3)	$79
The Weekend →	Sat · Sun (2)	$55

Free shipping over $99 · 8 colors × 6 sizes · 9 languages