On 17 October InfoQ wrote up a project called REA, short for Reverse Engineer Anything. Its tagline is plain: “Reverse engineer anything with agents, from app behavior down to native binaries.” It’s an MCP server. You register it with Claude Code, Cursor or another agent, and the agent gets a set of tools to inspect real software: JavaScript and Electron apps, .NET assemblies, Android apps, websites, and native binaries through Ghidra, Hopper or IDA.

The example in the README is the part that made me stop: “Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project.”

Thanks for reading! Subscribe for free to receive new posts and support my work.

I’ve only read about it so far. But I don’t need to run it to have an opinion on what it means, because I already had my own small version of this moment.

I didn’t need a decompiler

There was an indie game I used to play, Moon Chaser. Very simple, and it got removed from the store. One night I asked Claude Code if it knew the game. It did. A few prompts later I went to sleep, and by the next morning it was fully working: the mechanics, the levels, the art, the music, and the physics, which honestly amazed me. I built it for fun, for me.

No binaries, no disassembler. Just a description of how the game behaved and an agent that already knew enough to fill the gaps.

REA goes one level deeper. Where I described behaviour from memory, it reads the actual files and comes back with evidence. But the direction is the same. Rebuilding a feature you can see is getting close to free.

The part that used to beat me

For six or seven years I ran Qesheq, an Arabic writing platform, something like Medium. I built it by hand, before Cursor existed. It had a few hundred articles and freelance writers I recruited myself.

The one thing I never cracked was letting writers create their own accounts and write on their own. The profile and especially the editor were too much work to finish properly by hand, beside a job. So I kept posting everything myself through the CMS I’d built, and the platform never became what I pictured.

That editor was the wall between my idea and the product. Today my guess is an agent would build a decent version in an evening. And if I wanted to know exactly how Medium’s editor handles something, a tool like REA could go and look.

The wall is gone. That changes what the game is.

What’s left to compete on

I think the game is no longer who can build what. It’s whose idea is the brightest, who can make it approachable for real people, who can find the gap in the market, and who knows how to sell it.

A feature can be copied. Here is what I don’t think an agent can pull out of your binary:

Why you made each decision. Who you decided it’s for, and who you decided it isn’t for. What you chose to leave out. The trust you’ve built with the people using it. Where you show up so they find you in the first place.

A small example from my week. This week I changed who Relnk, the writing studio I’m building, is for. The site now speaks to consultants and freelancers who publish to grow their business, and onboarding asks if you write LinkedIn posts or write for clients. The code for that was a couple of pull requests. Anyone could copy the pages. Nobody could copy the decision, because the decision lives in my head, not in the repo.

That’s the part I spend my evenings on now. The agent does the typing.

The uncomfortable part

Not everyone reads REA as good news. The InfoQ piece mentions independent developers who worry it makes it easier to copy games and apps or get around paid features. I get it. If you spent a year on one clever feature, “an agent can study it and rebuild it” is a scary sentence.

REA’s own README says it’s for lawful research, education, interoperability, preservation, and software you own or are allowed to examine. I’m not a lawyer, so I won’t pretend to know where every line is. My own line is simple: learning how something works is fine, shipping someone else’s work as yours isn’t. My Moon Chaser stays on my machine.

What I’d actually point a tool like this at: an app I like, to understand why a feature feels good. A tool I already pay for, to make my own stuff work with it. And old code of mine where I’ve forgotten how a piece works.

What would change my mind

If copying turns out to be enough, I’m wrong. If clones that copy features start beating the originals on features alone, with no better idea, no better audience and no better selling, then building is still the game and I’ve misread it.

I’d bet against that. Everyone can build now. That’s exactly why building stopped being the thing that sets you apart.

Thanks for reading! Subscribe for free to receive new posts and support my work.