A developer shows how an open-source tool called REA lets any coding agent reverse engineer closed-source apps into working clones, then argues the only moat left is caring enough to polish what you ship.
Posted
2 days ago
Duration
Format
Demo
educational
Views
27.7K
580 likes
57 · 43
Big Idea
The argument in one line.
AI coding agents paired with the open-source tool REA can reverse engineer almost any compiled app into a working clone, so the only lasting edge left in software is how much the builder cares about polishing and maintaining what they ship.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You build software products and want to understand how AI reverse-engineering tools threaten closed-source moats.
You're a developer curious about using coding agents like Claude Code or Grok to rebuild features from other apps.
You want a plain-English explanation of how compiling and decompiling actually work.
SKIP IF…
You're looking for a step-by-step installation tutorial. This is a conceptual walkthrough, not a setup guide.
You have no interest in software development or coding agents.
TL;DR
The full version, fast.
AI coding agents can now reverse engineer almost any piece of software using REA, an open-source tool that connects an agent to established reverse-engineering programs like Hopper, Ghidra, and IDA. The agent repeatedly tastes the original app's behavior, guesses the underlying logic, builds its own version, and compares the two until they match, the same loop a human reverse engineer would run, but without fatigue or limits. The creator used this to rebuild a TradingView alternative, a Screen Studio alternative, and a Notion-style app in Rust, and even ported a competitor's response speed into one of his own production apps. His conclusion: once code itself can be cloned, the only thing that still separates winners is whether the builder actually cares enough to maintain and polish what they ship.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Viral clips of cloned games and apps set up the claim that any software can now be rebuilt by AI.
00:30 – 01:10
02 · Meet REA
His own Rust-built clones of a screen recorder, a trading terminal, and a notes app, then the REA GitHub repo that made them possible.
01:10 – 02:11
03 · How REA works
The prompt pattern for asking an agent to copy a feature, and the local MCP plus CLI that connects the agent to Hopper, Ghidra, and IDA.
02:11 – 04:05
04 · Compiling and decompiling, explained as cake
Compiling is baking the cake; decompiling is tasting it to guess the recipe. The reverse-engineering loop: taste, guess, build, compare.
04:05 – 05:31
05 · Why the rebuilt code never matches exactly
Source code versus machine code, and why an AI agent's decompiled version gets the same result through different logic (3+2 instead of 4+1).
05:31 – 06:29
06 · Sponsor: Scrimba
A plug for Scrimba's AI explainer tool, which the creator used to generate a video breaking down his own MCP server before a conference talk.
06:29 – 09:31
07 · Demos: TradingView, Screen Studio, and Paper clones
Walkthrough of terminal-rs, Glide and REC, and a Notion-style paper-rs, all built in Rust by Grokbot-orchestrated Claude Code sessions spread across his home network.
09:31 – 11:09
08 · Using REA in real products
REA applied to Pluto and Bezalel, two production apps with paying users; reverse-engineering a competitor's chat speed cut response latency from 50-117ms to 7-33ms.
11:09 – 12:57
09 · The real moat is caring
His closing argument: once any app's behavior can be cloned, the lasting advantage is polish and maintenance, not the underlying code.
Atomic Insights
Lines worth screenshotting.
REA pairs established reverse-engineering software, Hopper, Ghidra, and IDA, with a coding agent's MCP connection, letting AI rebuild any compiled app's features from scratch.
Compiling turns human-readable source code into machine code; decompiling works backward from that machine code, so a rebuilt app rarely matches the original's source line for line even when it behaves the same.
The reverse-engineering loop an AI agent runs is taste the original, guess the ingredients, build a version, and compare it back, repeated until the outputs align.
An AI agent can run that guess-build-compare loop indefinitely without getting tired, which is what makes AI-driven reverse engineering practical where manual reverse engineering wasn't.
Using REA, the creator rebuilt a TradingView alternative, a Screen Studio alternative, and a Notion-style app, all in Rust, just by pointing a coding agent at the repo.
He used REA to copy a competitor's chat response speed into a production app with real paying users, cutting response latency from 50-117 milliseconds down to 7-33 milliseconds.
For quick experiments he spins up one-off builds across idle machines on his home network with a lighter agent; for production work tied to paying customers, he switches to a more deliberate coding-agent harness.
If an AI agent can clone any closed-source app's behavior, the code itself stops being a durable competitive advantage.
His conclusion is that the remaining moat is caring enough to maintain and polish a product, not the code that powers it.
He expects most of the open-source clones he built on camera to go unmaintained within weeks, because only the products he genuinely depends on get ongoing polish.
Takeaway
Why code alone no longer protects your product
WHAT TO LEARN
AI agents paired with a reverse-engineering tool can now clone almost any app's behavior, so the lasting advantage shifts from the code itself to how much the builder cares about polishing and maintaining it.
01Cold open: software is dead
Viral clips of people running Minecraft, GTA 5, and Super Mario Bros on hardware they were never built for are evidence that AI has made cloning existing software trivial.
02Meet REA
A single open-source GitHub repo, REA, is what turned 'ask AI to copy a feature' from a request into something an agent can actually execute.
03How REA works
The prompt pattern is simple: ask your coding agent how a specific feature works and tell it to build an equivalent, then let REA's local MCP server connect the agent to reverse-engineering tools.
REA wraps established tools, Hopper, Ghidra, and IDA, that have existed for years but were too slow for humans to use at scale; an AI agent removes that time cost.
04Compiling and decompiling, explained as cake
What you download when you install an app is the compiled machine code, the baked cake, not the source code recipe behind it.
The agent's reverse-engineering loop, taste the original, guess the ingredients, build a version, compare it back, is just automated trial and error run at a pace no human can sustain.
05Why the rebuilt code never matches exactly
A decompiled app won't share the original's exact source code; it'll reach the same result through different logic, so treat REA output as a functional clone, not a verbatim copy.
07Demos: TradingView, Screen Studio, and Paper clones
Full app clones, a trading terminal, a screen recorder, a notes app, were each built in Rust from a single prompt plus the REA repo, run across idle machines on a home network.
Treat one-off coding-agent experiments and production builds differently: lighter orchestration for throwaway prototypes, a more deliberate agent harness for anything tied to paying users.
08Using REA in real products
REA isn't just for novelty clones: reverse-engineering a competitor's response-handling logic and porting it into a production app cut real latency from 50-117ms to 7-33ms.
The biggest performance wins can come from studying how a competitor solved a narrow technical problem, not from building that feature from scratch.
09The real moat is caring
Once an app's behavior can be cloned by an AI agent, the code itself stops being what protects a product.
The builders who keep users are the ones who keep polishing and maintaining after launch, not the ones who shipped first.
Expect most AI-assisted clones, including the creator's own demo apps, to go unmaintained within weeks; only products someone genuinely depends on get sustained care.
Glossary
Terms worth knowing.
REA
Reverse Engineer Anything, an open-source tool that combines a local MCP server and CLI with reverse-engineering software, letting a coding agent inspect and rebuild the behavior of any compiled app.
MCP
Model Context Protocol, a standard way for an AI coding agent to connect to external tools and data sources during a session.
Decompiling
Working backward from a program's compiled machine code to reconstruct source code, or equivalent logic, that produces the same behavior.
Compiling
Converting human-readable source code into the binary machine code a computer actually runs.
Hopper, Ghidra, and IDA
Established reverse-engineering programs that disassemble and analyze compiled binaries; REA exposes them to an AI agent instead of a human analyst.
T3 Code
A coding-agent harness the creator reserves for serious, production-grade builds rather than quick one-off experiments.
“If software itself is no longer the moat, what is the moat? And I genuinely believe the moat is caring.”
clean thesis statement with a built-in question hook→ IG reel cold open↗ Tweet quote
12:03
“Care matters. Polish matters. How it looks and feels matters. And this is the moat.”
tight rule-of-three closer, works as a standalone pull-quote→ newsletter pull-quote↗ Tweet quote
12:32
“If you don't care, me and the boys and the girls are coming with the REA GitHub repo and our coding agent, and we're going to open source your app.”
funny, specific threat line that lands the whole argument→ TikTok hook↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
analogy
Software is dead. And there's no denying it. I shared this on Twitter and a lot of people were pissed off.
But just look at this here. People are putting Minecraft and GTA 5. People are running Super Mario Bros on PC.
The best one is someone was running GTA 5 on a freaking iPhone. And it's not just games, it's software. For example, I built my own screen studio called Glide.
I built my own trading view called Terminal. And I also built my own paper alternative called Papel. All of this built in Rust, all of this built UI, fully functional.
Even the software that I'm using to record the video is the software I built using AI. Now, it wasn't just AI that allowed me to build this. There's this one super powerful GitHub repo called REA.
Reverse Engineer Anything. If you take this GitHub repo and give it to your favorite coding agent, every single software out there essentially becomes open source to you. In today's video, I'm going to show you how REA works, how I've been using it so you can use it too.
And finally, if you're building software, what the true moat really is, because it's no longer code. Let's get into it. So REA stands for reverse engineer anything.
And on a high level, how this works is you are going to prompt your agent. How does feature X work? Build it for me.
An example that I'll show you later on a live example is I wanted to copy a feature that Grokbot had into my app Pluto. So I asked my agent with REA, how does feature X work? Build it for me.
My coding agent then used REA and REA is basically a local. MCP plus a CLI.
And this local MCP essentially exposes all these tools, Hopper, Ghidra, and IDA. These tools have existed for a couple of years and they've been used to reverse engineer software. But this is a very time -consuming thing.
But now with AI, people are using these tools via REA. It's incredibly easy for the AI to run loops and build out the software it's been told to reverse engineer. But I want us to understand this.
So I have a diagram that's going to... explain it to you as if you were five. Let's say you write some code, right?
And you can think of this as a recipe to bake a cake. The compiler takes this code, right? And bakes it.
The compiler turns this recipe basically and bakes the cake. But what really is happening is this code is being turned into binary. And this file that's generated is what the computer runs.
And this file just contains ones and zeros. And this is basically machine code that the computer is running. You and I, when we download an app, we're not getting the actual source code.
We're getting the ones and zeros. We're getting the machine code. We're getting the actual cake.
I am not getting the recipe. I am getting the recipe baked and ready to eat. So that's what you and I get.
So when you download an app, understand this. You're getting the machine code. You're getting the compiled code.
You're getting the cake. You're not getting the recipe. Basically, what REA with our coding agent, what it allows us to do, what it does is the following.
Imagine this. The cake's already made, right? You're taking a bite.
You're tasting the cake and you're trying to figure out what ingredients are in the cake, what the recipe is by tasting it. And then you're going to bake a cake with some guesses that you've made, some educated guesses. You're going to taste it and you're going to see if it aligns.
So here it says, REA hands your AI the tools. REA lets your AI agent use those tools. It follows one feature and writes down what it finds.
Basically, by testing the cake, guessing what ingredients it could be. building a version and testing and running this loop again and again, what eventually happens is you're going to have your own cake. The AI writes its own recipe that does the same thing, then tests it against the original.
And until it aligns with the original, it keeps on going. Now, for a human to do this is very time consuming. You can get tired.
You have to sleep. You got a family. You got to go get diapers.
But for an AI agent, there's no emotions. There's no tired, unlimited tokens. It's just going to keep on building.
kind of understand this compile decompile thing because i use these words multiple times source code is the code that humans write well used to right now we let our agents write code but source code is the code that humans write this is readable to us this is typescript this is rust this is all that that you know the good stuff and you can see here it says example if score greater than 100 you run the win function the compile stuff basically compiler coming and compiling this source code is turning it into binary this is the code that the computer runs and this is just ones and zeros.
So my code, my slop is then converted to ones and zeros. And then these ones and zeros are what the computer runs. Decompiling is taking the ones and zeros and then through iteration, figuring out what the source code is.
Now, here's the thing. You're not going to get the exact same source code. Let's say the source code is four plus one equals five.
And then you get, you know, your binary and zeros and ones. And then. By running REA, the coding agent might come to the conclusion that 3 plus 2 equals 5.
Now the result is the same. How it got to the result is different. So you're not getting the exact source code, but you will get similar results.
And this is what's so powerful about REA. And just to clear up confusion, REA is a local MCP plus CLI that connects to all these tools. You know what else is an awesome tool?
Today's sponsor. This product is literally a lifesaver. I got invited to speak at the MCP.
dev summit in toronto october 5th to 6th i'm going to be talking about how i'm tired of switching personal agents and i built an mcp server called bezala which is a product that over 700 people use and they wanted me to talk about it here's the thing and i'm kind of embarrassed to admit i ai slobcat in the entire thing but i don't really understand fully how it works but that's where scrimba's explain plugin on codex works with cloud code as well saved my life i prompted saying here can you use We're not going to watch the entire thing, but I just want you to know it broke down the entire project.
It broke down the architecture. It broke down how it works in a three -minute video, all AI, but it explained it so well, I'm basically ready to speak. Now, you could literally go to the Scrimba website, scrimba .com slash explain, and you can prompt it here.
either download their Codex plugin, you can either go to their website, or you can choose to suffer by reading code generated by AI. The choice is yours. Make sure to click the link in the description.
Check out Scrimba. Let's get back to the video. Now, I'm going to show you how I've been using GrokBot and CloudCode to spin up all these open source alternatives of apps I've been using.
For example, TradingView. I gave my coding agent the REA GitHub repo, which I will link in the description. It's actually number one right now in GitHub, so it'll be easy for you to find it.
And then I used Grokbot and I basically told Grokbot to spin up cloud code sessions on my network and to rebuild TradingView, ScreenStudio, OBS and Paper. And I'll just show you one example. This is the TradingView alternative.
Very nice. Fully built with Rust. All these apps have been fully built with Rust.
And it's just so fast. It's so snappy. It's so fluid.
Like, I love this, right? I would not pay for TradingView anymore because I have this. And this is investing in something I'm passionate about.
And this is something I... actually maintain another app is glide someone on twitter was like can you build a screen studio but with rust and i was like sure and literally with a couple prompts using grokbot and clot and i'll show you exactly how i do it i built screen studio alternative fully built with rust the ui is not as pretty but it's fully functional and it works just the same and one of the apps REC, REC, short for recording.
I'm actually using right now to record this video. And the last one is paper. And this is basically a paper alternative, fully built on Rust with an MCP server that my agent can connect to, to make designs.
So here's the exact prompt I used. I said, I want you to use REA and reverse engineer screen studio. I want there to be feature parity.
I want a fully open source, super fast built using Rust, using Rust version. Even the prompt was terrible. Use cloud code Opus Extra High and use slash cold to get it done.
Now, one cool thing about my setup, I'm going to do a video on this, is I have multiple machines running in my network. You can see I have a VPS running here. I have a MacBook Pro.
I have a TV. I have a framework laptop. I have a Linux machine here.
I have another MacBook. And I basically run Grokbot on my personal MacBook. And I told Grokbot to spin up a cloud code session.
Now, there's a skill that tells it that it has access to my network. use whatever idle machine or whatever machine has the most memory or is not doing any work and so what grokbot did is it spun up a cloud code session in one of my machines you can see here it says i'm setting up this on mac machine it's free and the build needs needs a mac and then it says cloud code will map out every screen studio feature first including auto zoom cursor smoothing backgrounds the camera overlay the timeline editor exports and then build it in rust on top of rec's fast recording engine i'll tell you when the session is up and running.
So when you combine your coding agents with the REA GitHub repo, it can essentially turn any software product, any desktop app, any website, any software that you use, you can build an open source alternative easily. And all these apps are open source on my GitHub. The link will be in the description down below.
Now, just to make it clear, I don't use Grokwalk and Clog to build all software that I'm building. I generally use this to spin up slash goals for one -off products that I want to build and test. For anything that's actually serious that I need to be heads down focused on what needs to get built, I'm using T3 code.
And I used REA recently in two products that both have customers. The first one being Pluto, the second one being Bezos. Both of these products have users.
Bezalel is actually making money, which is fantastic. And I built features from other software products into a production product that people are using. And it was all thanks to REA.
For the first one, it was Grokbot. Grokbot has beautiful, just the way it handles chat, the streaming and all that stuff. And you're going to see, I'm going to send hi, how are you?
And you're going to see how quick the agent responds. And it's not even going to stream the message. It's just so quick, right?
And I wanted that for Pluto. So... I used REA to reverse engineer how Grokbot works.
And this is the merged PR. You can see that this is merged. And you can see here how much time was saved by the before and after tests, right?
From 50 to 117 milliseconds to 70 to 33. Like this reduction is insane. And that's because REA reverse engineer how Grokbot does it.
And I implemented it in Pluto. Now to do this, I used my favorite meta harness called T3 code. This is actually built by Theon Julius.
And this is fantastic. Fantastic. I have a whole video coming on T3 code, so make sure you're subscribed.
But the reason why I'm sharing this is REA is not just for fun little apps and games and putting Minecraft and Elden Ring. You see a feature you like from another app and you want to re -implement it or you want to figure out how it works and how they did it. The REA GitHub repo is a must.
And the final thing I copied is there's a feature I built on Pluto that was really cool and I just copied over to Bezalel. I probably could have just given my agent access to both repos to do it, but it was fun and it worked. Now, what is the moat?
If software itself is no longer the moat, what is the moat? And I genuinely believe the moat. is caring.
Because here's the thing. I built all these apps. I'm going to keep it above 50 with you.
I'm probably going to get over them in the next two to three weeks, right? I'm not going to maintain them. I'm not going to polish them.
If an app is really super important in my life, then yes, I will do that. But I'm not going to open source every single piece of software that I use. So when I say software is dead, I'm not saying software as a whole, and we're all going to use our own vibe coded versions.
That's not what I'm saying. But what I'm saying is the days are gone of you launching your to -do list app and you actually getting users because apps like that are done most apps can be open source now who stands out the people that actually care the people that actually spend time polishing the people that actually spend time with the ui and the ux and making it such an immersive and beautiful experience that i just want to pay you my money care matters polish matters how it looks and feels matters and this is the moat so if you're building something i hope you actually care if not if you don't care me and the boys and the girls are coming with the rea github repo and our coding agent and we're going to open source your app it is a great time to be alive i know for some people especially developers it's a stressful time it's an identity shock but if you look at the glass half empty versus half full you'll start to see all the opportunities in front of you i for one i'm super excited you've been awesome my name is ross thank you so much for watching this video i'll see you in the next one peace
The Hook
The bait, then the rug-pull.
Ras Mic opens with a blunt claim: software is dead. He backs it up with viral clips of strangers running Minecraft, GTA 5, and Super Mario Bros on hardware they were never built for, then pivots to his own apps, Rust-built clones of TradingView, Screen Studio, and Notion, all reverse-engineered by an AI coding agent using a tool called REA.
Frameworks
Named ideas worth stealing.
03:37model
The Reverse-Engineering Loop
Taste the original
Guess the ingredients
Build a version
Compare to the original
The four-step loop an AI agent runs to reconstruct a closed-source app's behavior: observe the real thing, hypothesize the underlying logic, build an attempt, then test it against the original and repeat until they match.
Steal forany workflow where an agent needs to replicate an existing system's behavior without access to its source
CTA Breakdown
How they asked for the click.
VERBAL ASK
05:31product
“Check out Scrimba, link in the description.”
Mid-roll sponsor read demonstrating Scrimba's AI explainer by generating a 3-minute video that breaks down his own codebase.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A creator who pays for every major AI coding subscription ranks Codex, Claude Code, Cursor, and Devin on models, subsidies, and harness quality — then reveals how to get your employer to cover the bill.
How one developer chained a computer-use verification loop and an automated code-review loop to ship 75+ pull requests with an AI cloud agent, without reading most of the code.
00:11
00:20
00:43
00:51
00:59
01:22
01:31
01:39
01:53
02:01
02:19
02:30
02:37
02:50
03:00
03:05
03:23
03:32
03:35
03:44
03:59
04:20
04:27
04:32
04:51
04:56
05:07
05:17
05:24
05:44
05:54
06:06
06:13
06:21
06:41
06:48
07:02
07:13
07:17
07:29
07:42
07:49
07:58
08:10
08:16
08:25
08:40
08:49
08:56
09:08
09:15
09:30
09:51
09:55
10:03
10:12
10:28
10:36
10:44
10:56
11:05
11:18
11:32
11:44
11:53
12:03
12:13
12:23
12:32
12:42
12:52
17:36
19:22
Video of the Day20:40
21:24
17:06
Video of the Day17:29