Subzero Labs

2026-11-21

Stop Building Copy Apps. Here's Where the Money Actually Is

AI makes apps fast, but copies flood the market. Learn what still pays: solving real problems, six moats AI can't copy, and three pricing models that work.

In short

  • Anyone can build basic apps with AI now, so habit trackers and dashboards are everywhere—the real money is in solving a specific problem clients already pay to fix.
  • The shift is from selling seats to selling results: per-use pricing, per-result pricing, or a human service on top of a free tool.
  • Six moats still protect you: proprietary data, network effects, your audience, physical assets, capital, and regulation—pick one and own it.

Why Everyone Builds the Same App

AI app builders like Lovable now see about a million new projects start every week [1]. That's a lot of people shipping fast. When speed is the only barrier, everyone lands on the same few ideas: habit trackers, note-takers, to-do lists, simple CRMs.

The speed advantage is real. In one controlled test, developers using an AI coding assistant finished the same task about fifty-six percent faster than those without [4]. That means more copies, faster. Wix bought Base44 last year, and five months after the deal, the app had two million users [3]. Most of them were building the same kinds of tools.

Even the people who already have software are doing it. Indie maker Pieter Levels cancelled almost all his SaaS subscriptions and started building his own versions with AI instead [5]. If your plan is to build app number ten thousand that does the same thing as the other nine thousand nine hundred ninety-nine, you're in the trap.

What Actually Pays: The Problem, Not the App

Here's the shift that matters. Nobody wakes up wanting an app. They want more bookings, fewer missed calls, or invoices that get paid. The app is just the window. The real work is solving a problem they already pay to fix.

Think of a calculator versus an accountant. A calculator is free. People still pay the accountant because the accountant gets the taxes done. For every dollar spent on software, about six are spent on services [10]. That bigger pile is where the money is.

This is what people mean by agentic. An AI agent takes steps on its own: it reads the message, decides what to do, and does it. The agent does the job. The screen is just the interface. Big players already price this way. Intercom's AI agent, Fin, costs ninety-nine cents per outcome, and it needs no seats [8]. Salesforce sells its Agentforce agents on credits too, which works out to about ten cents for each action an agent takes [9].

Six Moats That AI Can't Copy Overnight

A free tool with no moat is just a free tool. Someone gives theirs away too, and you're back to zero. A moat is whatever stops a copycat from taking your customers. Six moats are still left [6]: proprietary data or content only you have, network effects (the product gets better as more people use it), an audience you can reach whenever you want, something physical like hardware or a real location, capital to invest, and regulation like licenses and legal barriers.

For one person building alone, three are within reach. Your own data: every job your tool does should make the next one better. Templates, replies that worked, answers your users rated. That's the part a copycat can't download. Your own community: a free version, a group, a newsletter, people who help each other. That's an audience you own. And owning one niche workflow from start to finish, so deep that nobody else bothers.

Three Ways to Charge for the Result

Per use: charge per task, or per amount of AI used. Per result: charge only when it works. Intercom's ninety-nine cents per outcome is this model [8]. You only get paid when it works, which is exactly why customers trust it. Or a monthly fee for a human service on top, like a weekly coach or a personal meal plan made by an agent.

The per-result model is the scary one because you only get paid when it works. Which is exactly why customers trust it. Big companies still pay for security and someone to call, so per-seat pricing still works if you sell to large businesses. But for most solo builders, per-result or per-use beats per-seat.

Five Steps to Build Smart Software

Step one: don't build a copy of a basic tool. If you'd describe your idea as a simpler version of some tracker or dashboard, assume someone will give it away free. Step two: pick one niche and one job. Not a CRM. Something like follow up every new lead for a dog groomer, the same day. The agent does the job.

Step three: own the data. Every job your tool does should make the next one better. Step four: build the community around it. A free version, a group, a newsletter. That's an audience you own. Step five: price the result, not the login. Charge per task, per outcome, or per month for a human service on top.

Before you build, run your idea through six questions. One: could someone rebuild this in a weekend with AI? Two: does it do the job, or just store the job? Three: does it get better with every user? Four: do you own the data it creates? Five: is there a community or audience around it? Six: can you charge per result instead of per seat? Mostly yeses on two to six, and a no on the first one? You're building smart software.

Other Routes That Still Work

This isn't the only way. Three other routes still work. One: classic per-seat software. Big companies still pay for security and someone to call. If you sell to large businesses, seats can work for a while. Two: skip the software and sell the service. Do the work yourself with AI helping you, and turn it into a product later.

Three: a simple paid tool in a niche so small the big players ignore it. It can work. Just know the walls are low. The video above goes deeper into each of these paths and shows real examples. The key is knowing which moat you're building on, and which one fits the work you want to do.

The Plan We're Building

We're building a self-hosted AI dashboard: one small box at home that holds your AI services, sorted by project, tries free credits first, and shows what each job costs. The plan is to put the basic setup on GitHub for free, then offer a hosted web version on top for people who'd rather not run a box at home. Free tool, paid convenience. Same play.

It's a plan, not a proven result yet. But it shows how the shift works: solve a real problem (high AI bills), own the data (what each job costs), build the community (people who care about AI costs), and charge for convenience instead of the core tool. That's the whole game now.

Watch "Everyone Can Build Apps With AI Now. Here's What Still Pays." on YouTube

Sources

  1. Lovable, 1M new projects a week (TechCrunch)
  2. Wix buys Base44 (Globes)
  3. Base44 at 2M users (Calcalist)
  4. Copilot study, 55.8% faster (arXiv 2023)
  5. Pieter Levels: cancelled SaaS
  6. Pieter Levels: the moats left
  7. Motley Fool: software stocks
  8. Intercom pricing
  9. Agentforce pricing
  10. Sequoia: Services, The New Software
  11. Pi Pico W
  12. Raspberry Pi 5 8GB
  13. Raspberry Pi 5 Starter Kit - 16GB RAM
  14. NVMe Base for Raspberry Pi 5 M.2 HAT
  15. PiSugar Whisplay HAT for Raspberry Pi
  16. ELECROW 5 Inch Mini Touchscreen Monitor
  17. NIMO Powerful AI Mini PC, AMD Ryzen AI Max+ 395
  18. NIMO Mini PC, AMD Ryzen AI Max+ 395 128GB LPDDR5X
  19. AX9 Max Mini PC (vapor chamber), Ryzen AI 9 HX 470
  20. EVGA GeForce RTX 3090
  21. Raspberry Pi Zero 2W Kit
  22. Pi Zero Aluminum Case Kit with Pin Header
  23. Crucial MX500 2TB 3D NAND SATA 2.5 Inch
  24. Corsair Vengeance RGB RS DDR5 32GB (2 x 16GB)
  25. Hosyond 7 inch DSI touchscreen
  26. DA7212 I2S audio board
  27. Geekworm X1001 NVMe board for Pi 5
  28. ELECROW 10.1 inch ESP32-P4 display
  29. Waveshare ESP32-S3 2.06 inch AMOLED watch

More videos