All About AI

How I Farm Money On The App Store Using AI

By · · 6 min read

Two apps. Four hours of total work. $789 in sales across 319 units. That's about $197 per active hour, and the trend is up, not down. This post is the full playbook — how I find the topic, build the app, automate the App Store submission, and decide which ones are worth iterating on. The unlock is treating each app as a one-hour experiment, not a project.

Watch the video:

The headline number

Last 90 days, two paid apps live on the App Store, two test apps that didn't go anywhere. I've spent roughly four hours of focused work across all of them. Result so far: $789 revenue, 319 units sold, current run rate around $50–70/day. That's a $197/hour effective rate before Apple's cut, and it's compounding because the apps keep selling once they're up.

This is the same shape as the 10-day update post from a few weeks back, just with more data behind it. The pattern works. The question is how to keep it working without falling into sunk-cost on any one app.

Step 1 — Find the rising topic

The single biggest leverage point in this whole flow is topic selection. I don't pay for any marketing on these apps, so they have to be discovery-friendly out of the gate. Three sources I check, in order:

Google Trends (US, last 7 days, by category)

Filter to US + past 7 days, then sort each category by search volume. Gaming, entertainment, shopping are usually the best veins. You're looking for a thing that exists, is rising fast, and doesn't have an existing dedicated app — or has one that's clearly bad.

SubReef — Reddit's growth tracker

"Discover top growing communities on Reddit", weekly sort, all sizes. This is where I found the non-toxic-products subreddit example from the video. SubReef gives you a list of communities accelerating right now, with daily and weekly growth. If a subreddit doubled in a week, that's a niche with hungry attention.

Google "is there an app for X"

Literal Google query. Find topics where people are asking if an app exists and the autocomplete or top results don't surface a good one. The phrasing alone tells you intent — someone typed the words "is there an app for".

Across all three, the rule is the same: rising signal, no clean existing solution, can be built without a backend. The last constraint matters — backends mean ongoing cost and rejection risk, and I want apps I can ship in an hour and forget about.

Step 2 — Agentic research, not vibes

Once I pick a topic, I drop into Claude Code and ask it to produce a research package via sub-agents. Same pattern as the 3-part agent system: a research skill with explicit scope.

Prompt shape:

These are sources for a new iOS app called [name]. The app should be very beginner-friendly for people to understand [topic]. Your task is to do research for the app and create a research package we can use to build it. Good luck. Use sub-agents if needed.

Sources I paste in: the rising subreddit's top posts, a couple of Reddit threads where people ask questions in that domain, anything Wikipedia-level on the underlying topic. The agent spins out sub-agents, comes back with a structured package — categories, definitions, common pitfalls, examples, glossary terms. That package becomes the seed for the app's content.

This step is non-negotiable. The reason these apps stand a chance is because the content is actually good — the agent has done domain research and produced a real reference. The neo-brutalist UI is a wrapper.

Step 3 — Build in one prompt, fix in two

Move the research package to the MacBook (which is where Xcode lives), drop it into a fresh Claude Code session, switch to plan mode, and describe the app:

Next step is to build the app. Super easy to use. Goal: help users understand [X]. Design: neo-brutalism look. Colorful, interesting icons. Non-complex. On-device only, no APIs. Create a detailed plan.

"On-device only, no APIs" is the magic line. Removes server cost, removes review-rejection risk, removes the need for a privacy policy that mentions data processors. Everything ships in the binary.

The first prompt produces a build that's 80% there. Run it in the iOS Simulator, screenshot the obvious bugs (overlapping text, dead links, weird spacing), paste those screenshots back into Claude Code with one-liner fix requests. Usually two or three rounds and the app feels good enough to ship.

Claude Code drives Xcode directly through the small automation rig I built earlier — same pattern as the automating iOS apps post walks through. Without that, you're hand-managing build/run cycles and the time creeps up fast.

Step 4 — Automate the App Store submission

This is where most people would lose the time advantage. App Store Connect's new-app form has roughly 40 fields — name, subtitle, description, keywords, categories, age rating, privacy details, screenshots, build, compliance. Doing it by hand is 30+ minutes of dropdown clicking.

So I don't. I use Surfagent to open a CDP-controlled Chrome already signed in to App Store Connect, then tell Claude Code: "we are logged in on App Store Connect in CDP Chrome — start work on the upload of the new app [name]."

Because Claude Code already has the research package and the built app's metadata in context, it fills the form correctly without me re-typing anything. Screenshots come from a quick Xcode capture pass. Compliance flags get answered ("uses encryption? no"). Add for review. Submit.

End to end from "I picked a topic" to "submitted for review", with research + build + submission, is about 60 minutes if I'm focused.

Step 5 — Don't iterate on a dud

This is the rule that keeps the per-app time bounded. Once submitted, the app gets ~24 hours of Apple review, then sits live. After that:

The sunk-cost trap is real. If I let myself spend a week polishing an app that wasn't selling, my $/hour collapses to zero. The whole model works because each shot is cheap enough that misses don't matter and hits compound.

Honest costs and limits

What I'm shipping next

The non-toxic-products app I built in the video is in review now. I'll know in a week whether it sticks. The shortlist for next builds, all surfaced from the trends/SubReef sweep this week:

If you want to follow which ones live and which ones die, the updates land in the AI_automata Discord first.

Resources