All About AI

UPDATE: App Store AI Automation After 10 Days (Profit?)

By · · 4 min read

About two weeks ago I published my video on automating iOS app creation with AI in three days. The numbers in that one were small but promising — $33 in three days, two apps live. Today I want to share the 13-day update: revenue, downloads, what I learned, and a look at the latest app I just built using AI for the actual product (not just to write the code).

Watch the video:

The Numbers After 13 Days

Three apps live now: Needle Collector, Poke Machine, and the newest one Looks. Sales totals from App Store Connect:

Apple's developer fee is $70/year. So as of day 13 I am clearly in profit on the experiment, with the $20/day rate continuing. That is the part I care about: the pipeline produces stable, predictable revenue with no ongoing labor.

This is the same compound-tiny-loops thesis I covered in my Claude Code passive income post: each individual app does not have to be a hit. You stack them and the math gets interesting.

The Latest App: Looks

The first two apps were utility apps — static functionality, no API calls. Looks is different — it actually uses an AI model (OpenAI's GPT-4 Image API) as the core feature. The pricing structure has to be different to cover the API cost, so this was the first one where the unit economics matter.

What the app does: upload a photo of yourself, pick a "session" (one, three, or five), and the app generates four "old money lifestyle" images, six matching haircuts (separate male/female styles), and a wardrobe palette. Pure consumer fun.

The monetization is in-app purchase per session, not subscription. Each session is one API call's worth of generation. Which means the pricing has to cover the GPT-4 Image cost plus margin.

Pricing AI Apps: The Math That Matters

This is the part most "vibe-coded AI app" tutorials skip. GPT-4 Image API pricing (at the time of recording):

But that's per million tokens. For an actual image, you need to translate that into per-image cost based on quality (medium vs high) and aspect ratio (vertical, horizontal, square). For my "10 generations per session" feature, I had to calculate the worst-case session cost and price the IAP above that with margin to cover the App Store cut, refunds, and headroom.

If you skip this calculation, you can ship an app that loses money on every sale. You won't notice immediately because volume is low. By the time you do notice, the loss is meaningful. This is genuinely the most important step when AI is in the loop, and I gave it more thought than the actual app design.

The Build Flow Is Still Fully Automated

The build pipeline itself is unchanged from the original 3-day video. Clone the scaffolding repo, prompt Claude Code:

"Open Xcode with the app Looks, and the simulator."

Claude opens Xcode, launches the simulator, navigates the app, and tests flows autonomously. I can iterate on UI changes from inside Claude Code without touching Xcode directly. The same Surfagent-driven submission flow handles App Store Connect uploads — covered in detail in the Surfagent post.

What's Next: Trend-Surfing AI Apps

The plan for episode 3 of this series is what I am calling trend-surfing. The thesis: pick app ideas that don't need to be long-lived. Build for a current viral trend, ship fast, capture the demand wave, retire when the trend dies. Move on to the next.

The example I am chasing is "Umogle" — a viral image-based app I have been seeing on Reddit and X around looksmaxxing-style content. There is search demand, the category is hot, and there is no entrenched competitor yet. Perfect for a 3-day build.

If trend-surfing works, the value isn't in any single app — it's in the speed of the pipeline. Each app is a 3-day investment, makes whatever it makes, and the pipeline keeps shipping.

Why I'm Sharing the Numbers

Most "make money with AI" content avoids actual numbers because the actual numbers are usually disappointing. $275 in 13 days isn't going to change anyone's life. But it is also not zero, the per-day rate is stable, the margins make sense, and the pipeline is genuinely automated.

The honest takeaway from the 13-day mark: this works, slowly, predictably. The unit economics are positive. Scale is the only remaining question — and that's mostly a function of how many apps I ship, which is mostly a function of how reliable the AI pricing math is.

Same general lesson as my 504-hour autonomous agent experiment: small autonomous loops, run consistently, beat one-shot grand bets.

Resources

Next episode: building the trend-surfing AI app. Subscribe on YouTube if you want to follow this series — I'll cover the pricing calculations and trend-detection workflow in detail.

FAQ

How much can you make with AI-built iOS apps in 10 days?

After 13 days the three-app pipeline generated $275 in revenue and 130 sales, with daily revenue stabilizing around $20 — already in profit on the $99/year Apple developer fee.

What's the difference between utility AI apps and AI-powered AI apps?

Utility apps have static functionality (no API calls), while AI-powered apps make live LLM API calls and require careful unit-cost calculation so the in-app purchase price covers per-session API costs plus margin.

How do you price an AI-powered iOS app?

Calculate worst-case session cost from API token pricing (input + cached + output tokens at the right quality and aspect ratio for image models), add the App Store cut, then add margin — typically 3-5x the API cost minimum.

What's trend-surfing for iOS app pipelines?

Trend-surfing is picking app ideas tied to current viral trends (e.g. looksmaxxing, image-based meme apps) — short-lived demand windows where minimal marketing is needed because users are already searching for the category.