AI can copy any app. Now what?
What an app does just got easy to copy. What still protects the business is the data, the infrastructure and the relationships around it.
AI is already able to replicate what any app you install on your computer does. The question left for anyone who builds software is now what?
Video games showed the way
Gamers were the first to show this. One group took a classic Nintendo 64 game and built a version that runs directly on a PC, with better graphics and mod support. They never had access to the company’s code, only to the game itself, which anyone had at home.
The same thing happened with an Xbox 360 game, which got an unofficial PC version made by the community. In both cases, a tool translates the game from one machine to another, like translating an entire book from one language into another.
The next target is the PlayStation 5. The open source project AnyPS5 wants console games to run on a PC without emulating the whole device. According to Tom’s Hardware, it already covers all the graphics work it set out to cover, and shortly after a PS5 exclusive showed up running on a PC.
Games are the most complex programs most people use. If they can be rebuilt this way, the app on your computer can too. The community has already noticed that AI is speeding up this process.
AI became a program reader
The tool that spread the subject is called REA, at rea.tools. It is an open source package that connects an AI agent to the tools that examine an installed program. You ask how something works, and the agent opens it, inspects it and explains each part.
The example on its own site is for people who are not specialists. The agent figures out why the Windows calculator gives 220 when you type “200 + 10%”, then rebuilds the percent button with the same logic. Understanding what a program does became a conversation.
The scale is striking. According to Decrypt, the project passed 100,000 GitHub stars in about a week. Part of that number looks inflated, but the interest is real.
Practical examples have already shown up. Someone tired of waiting 30 seconds for the TV video app to open studied how it worked and wrote a new app from scratch with AI’s help. Others show how to ask the agent to take apart a feature from another app and build their own version.
And it does not stop at the computer. People are already getting a printer to accept another brand’s ink. The devices you buy are part of this too.
There is still no proven case of a whole program rebuilt by AI alone, with no people involved. What exists is AI making the people who do this work much faster. The most expensive step, which was understanding how something works, just got cheap.
The oldest protection has fallen
Decrypt put it very well: software’s oldest protection was how hard it was to read the program. And the agent ended that. A copy used to take months of specialist work, and now a good prompt is enough.
This has happened before, on a smaller scale. In the 1980s, Compaq recreated how IBM’s BIOS worked, the basic program that starts the PC. One team studied what it did, another wrote everything from scratch without ever seeing the original, and that is how the market for compatible PCs was born.
The difference is that now it no longer takes a whole company. If understanding a program got cheap and writing a new one with AI did too, what the software does loses value. The logic of the processes, the screens and the business rules become easy to replicate.
If you sell an installed app, you have to take this into account. Any competitor can study the program and launch something similar in a short time, taking your place. The advantage of having done it first lasts less and less.
On the web you can copy the screen, not the engine
On the internet the situation is different. Anyone can copy how a site looks in minutes, but the part that makes the service work stays on the company’s servers and never reaches your computer.
Your bank is the best example. You can take the bank’s app apart, but not the bank, because the balance, the rules and the checks stay on the other side of the line. You see the dashboard, but you do not open the hood.
The price of that protection is infrastructure. Keeping everything in the cloud means paying every month for servers, bandwidth and maintenance. Computing and bandwidth are still expensive, and the data a service builds up with use never shows up inside any program.
The market is already reacting
There are reports of accounts banned by AI providers after the agent was used with REA. The providers do not want to be seen as the tool that helps copy other people’s products.
Serious projects that recreate famous apps, like Adobe’s, follow the Compaq recipe. One team studies the behavior, while the other writes the code without ever looking at the original, to stay away from lawsuits.
The rule is very clear. Copying how it works is accepted, copying the code is not. In the United States, the digital copyright law lets you study a program to make it communicate with others, but legal protection covers the text, not the idea.
Now what?
What the app does, on its own, just got easy to replicate. What really makes something hard to imitate is the data, the infrastructure and the relationships around it, like contracts and the use that builds up over time.
For your product to hold up, build on data that only you have and that grows with use. Keep the engine and that data on the server, where someone who copies the program cannot reach. Compete on what cannot be taken out of the app.
On the other side, some people are going further. A post that spread this week combines REA, a frontier AI model and a notes app into an agent that takes a system apart, explains each part and builds its own version. According to its author, this comes out up to 70% cheaper and 4 times faster. The message is to stop paying for tools you can build yourself.
One example makes this very clear. Clone Spotify and you walk away with just the player, without the music and without the label deals that let you charge for it. The program itself is the easiest part to clone. What holds the business together lives outside it.
Value has left the code. It sits in the data and in what exists around it, and those who understand this early build a business that is harder to copy.