REA (Reverse Engineer Anything) brings reverse engineering to AI agents, but there are at least three obstacles between that and copying any software

"Every piece of software can be replicated now." (not true)

I’ve been reading this everywhere since REA came out. I downloaded the project and went through the code and the documentation.

What REA (Reverse Engineer Anything) is

REA (Reverse Engineer Anything) is an open source project that connects AI agents like Claude Code, Codex or Cursor to reverse engineering tools, meaning tools that let you understand how a program works without having its source code.

With REA, an agent can analyze:

- desktop apps built with Electron, like Notion, Slack or VS Code

- websites, by watching their scripts and network traffic

- Android apps and .NET programs

- compiled programs, through decompilers like Ghidra, Hopper or IDA

The list of supported targets and agents is in the project README on GitHub.

It’s serious, well-built work. But between this and "I can copy any software" there are at least three obstacles.

The three obstacles between understanding software and copying it

1. It can’t see the server.

With an app like Notion, you only see the part installed on your computer and the requests it sends over the network. The backend, the database and the logic running on their servers stay invisible. And that’s where most of the product lives.

2. It doesn’t unlock protected software.

REA has no deobfuscator and, by design, it doesn’t bypass protections, DRM or license checks. The authors say so themselves in the project’s disclaimer.

The REA disclaimer explicitly rules out bypassing DRM, license checks, anti-cheat and other technical protection measures.

3. Understanding isn’t copying.

Even if you understand exactly how a feature works, you still have to rewrite it, test it and maintain it. And a product’s value lies in its users, its data, its integrations and years of fixed bugs, not just in the code.

Rewriting, testing, integrations and maintenance are also the largest share of every custom business software development project.

The authors’ example: a 63-byte function

One detail says a lot: the most impressive example the authors have published is the reconstruction of a single function from an old video game. A 63-byte function that works out whether a sound goes left or right. Rebuilt to match the original exactly, which is a remarkable result. But it’s one function, not a piece of software.

The game is DX-Ball: in the showcase published by the authors (checkpoint of 7 October 2026) the 63-byte function converts a brick’s horizontal position into the stereo balance of the sound, and the rebuilt version is verified against 3,205 cases compared with the original x86 code.

What’s true about reverse engineering with AI agents

So what’s actually true?

Until recently, figuring out how an app implements a given feature took a reverse engineering expert and days of work. Today an agent with REA gets there in far less time, especially with Electron apps and websites.

Personally, I’m tired of all this weekly hype over things that turn out to be mostly smoke.

What do you think? Let me know in the comments 👇🏻