Speaker A
Today we’re talking about something happening in the AI world right now that many people aren't aware of yet.
AI can now recreate entire software and games by decompiling and recompiling code, threatening traditional software models.
Key Takeaways
- AI can reverse-engineer software quickly, making proprietary code accessible.
- Software recreation enables free alternatives to expensive commercial programs.
- Video games are a leading example of AI-driven software decompilation and remixing.
- Traditional software protection methods are becoming obsolete due to AI.
- Companies must reconsider software investment and update strategies in light of AI advancements.
What the video covers
- AI models like Claude 3 Opus, GPT-4o, and Gemini 1.5 can decompile and recreate finished software without access to original source code.
- Recreations exist for major Adobe products and popular software, all available for free and installable via a unified installer.
- AI-driven decompilation allows games like Call of Duty to be rebuilt inside other games like Minecraft with original mechanics intact.
- The process of decompiling software, once extremely tedious and manual, is now accelerated by AI to take weeks instead of years.
- Fan communities use AI to preserve and modify old games, enabling them to run on new devices or browsers.
- Tools like Ghidra, combined with AI code agents, facilitate partial decompilation and reconstruction of software.
- Two main approaches to software recreation are AI reading the binary blueprint or reconstructing from observing program behavior (clean room approach).
- This technology threatens the traditional software licensing model, as software may effectively become open source regardless of company intentions.
- Many outdated enterprise programs could be replaced or automated more easily due to AI capabilities.
- The video discusses implications for software users, developers, and companies, highlighting a major shift in software accessibility and development.
Chapters
- 00:00Introduction to AI software recreation
- 00:39AI recreations of major software suites
- 01:26AI-driven game decompilation and modding
- 02:12AI models enabling rapid software reverse engineering
- 03:03How software is compiled and decompiled
- 04:36AI accelerating the decompilation process
- 06:13Combining game elements and software mashups
- 07:08Tools and methods for AI-assisted decompilation
- 08:38Clean room approach and open source implications
- 10:07Future impact on software industry and users
Full Transcript — Download SRT & Markdown
Speaker A
The models have become so good by now that they can recreate finished programs without knowing their code.
Speaker A
This here, for example, is not Photoshop. It’s a recreation that Claude wrote in about a week, and Claude can also control this entire instance. I’ve told it here that I want this image with its individual layers, and I can now click on all of
Speaker A
them here. I can move things, I can draw new things in, that's all no problem. I can edit all the effects here myself, just like you know it from Photoshop. And there are now also recreations of all other Adobe products
Speaker A
like Illustrator or Premiere, After Effects, Lightroom, InDesign, Acrobat—here PDF Craft—or even Word and Excel, Wordcraft, Gridcraft, even PowerPoint and also AutoCAD. And other programs are currently being worked on, and all of these can be used for free.
Speaker A
You can download them. There is now an installer where you can install them all. I'll also leave that in the video description if you want to take a look.
Speaker A
It started, especially in the last few weeks, with video games, because AI has become so good that it can break a finished game down into its individual parts—that is, decompile and reassemble it. This is currently resulting in things like this. This
Speaker A
here is Call of Duty, but in a Minecraft world with the weapons from Call of Duty and the way the soldier moves there. At first glance, it looks like a completely normal mod, but there’s much more to it, because with
Speaker A
a classic mod, the weapon would have at most only looked like it was from Call of Duty. It would have still fired according to Minecraft’s rules, though. So, Call of Duty itself is actually in here. The weapons behave
Speaker A
just like in the original, and harder blocks, for example, need more hits, depending on how much damage the weapon does in Call of Duty. So you’ve essentially taken parts of one game and packed them into the other game. And
Speaker A
there are even an extreme number of games now that simply run in the browser, even though they were actually developed for a console or as a real program on your PC, like Call of Duty here. And that works because the game
Speaker A
has been completely dismantled and exists as normal code again, which you can then recompile for any device, including the browser. Everywhere on the internet right now, you can see how all kinds of games are simply being combined, and this is happening on such
Speaker A
a scale because new models like Claude 3 Opus, GPT-4o, or Gemini 1.5 are extremely good at mapping all of that out, and in a very short time. Many people on X, such as the head of the programming platform Replit, therefore
Speaker A
believe that soon every piece of software will essentially be open source—meaning readable by everyone, whether the company wants it or not. In this video, we’ll take a look at how this works, what’s already possible today, and above all, what this means
Speaker A
for the programs you pay for every month, or even the completely outdated software still used in many companies, especially in Germany, that nobody really wants to use. So let's get started. To better understand this, let’s stay with video games for a
Speaker A
moment, because that’s where AI-driven recreations have made the biggest impact in recent weeks, mainly because there have been fan communities for decades that want to keep old games alive. They want to play them on new devices, fix bugs, and modify them, and
Speaker A
they have been trying to take these games apart for a long time to do so.
Speaker A
And on the site decomp.me, you can see over 200 such fan projects, and these are exactly the people who are now using AI for it. And now you might be asking yourself how one even takes a game like that apart. Put simply, the
Speaker A
process looks like this. You download the game, and a program file ends up on your computer, for example, a .exe file on Windows. You can then start it and play the game. However, you can’t look into the source code while doing
Speaker A
so. The file essentially only contains zeros and ones, so-called binary data, and these are instructions for the processor that no human can make any sense of. And the developers, of course, wrote the game in a programming language that humans can read. That is
Speaker A
the source code, which is basically the blueprint of the game. It states, for example, that the character ducks when you press a certain key, and that the magazine refills when you reload. And before the game is sold, a program
Speaker A
translates this blueprint into those exact zeros and ones. This is also called compiling. This means you are essentially buying the finished house, but never the blueprint for the house.
Speaker A
For decades, that was also how software was protected. Retrieving the blueprint from the finished file is called decompiling. And that has always been possible, but it was incredibly tedious because almost everything that helps humans read it is lost during the
Speaker A
translation. And in the source code, one function might be called "player crouch" and another "player reload." But in the finished file, those names are gone. All that remains is a memory address and a sequence of computational steps. The developers' notes and the
Speaker A
organization of which piece of code belongs where have also vanished. So, if you want the blueprint back, you have to work through hundreds of thousands of such steps and figure out what each one does. And then you have
Speaker A
to write code that, when compiled, results in the exact same zeros and ones as the original. And you have to imagine that dozens of people sit there to decompile this game, and it really takes a very long time. Take the game
Speaker A
Super Smash Bros. Melee, for example; it took people 6 years to decompile it, and for most of it, they didn't use AI, but at the very end in 2026, they finished the whole thing with AI. For an AI, this decompilation work is ideal
Speaker A
because it can check for itself whether its result is correct. It essentially writes a piece of code, translates it back into numbers, and places the result next to the original. If both are the same, it takes the piece; if
Speaker A
not, it tries again. Like a puzzle where you can immediately see if a piece fits. Only thanks to AI, it happens a thousand times in a row. All through the night with several AI agents simultaneously. This means retrieving the blueprint from a
Speaker A
finished program used to take years, but with AI models like Claude Opus 5 or GPT-6 Astra, it sometimes takes only weeks. So when you have the game's blueprint, you can do exactly these kinds of things here, where you
Speaker A
simply install parts from one game into another, just as you would install a normal mod. But it's not just the look that is changed; you are bringing in entire functions, like building and blocks from Minecraft into another game,
Speaker A
because you know how to translate one game so that it works in the other.
Speaker A
This means the rules of several games run simultaneously and interlock. The game therefore becomes like Lego. You simply take parts out, put them into another, and build something new from them. It is important to mention here that these are not yet finished games.
Speaker A
Many of these mashups still have numerous bugs. But that will probably change soon. And by the way, once you've decompiled a game like that and have its blueprint, you can recompile it for any other device—for PC, mobile, or even to run directly in a
Speaker A
browser, just like the Call of Duty game I mentioned at the start. If you want to try this out yourself, there are two tools currently being used by many people. The first one is called Ghidra. It was developed by the NSA, a
Speaker A
US intelligence agency, and they released it for free in 2019. You open a program with it, and Ghidra turns the zeros and ones back into code. So, it partially decompiles the program, though without all the original function names and notes from the
Speaker A
developers. But you don't have to struggle through all that code yourself. You can connect Cloud Code or Codex to Ghidra vi
Speaker A
part, explains it to you, and gives everything understandable names that you can use to continue working. The second tool is called Rea, short for Reverse Engineer Anything, which was recently at the top of the GitHub trending list for a long time. You also
Speaker A
connect Rea to Cloud Code or Codex and then basically ask in plain language how a specific function in a program works. See a feature in an app that you want in your own product. Ask your agent to investigate it with Rea. It
Speaker A
can inspect the app without its source code, explain how the feature works, show the evidence and build a version for your project. This means the agent finds the part of the program you're interested in, shows you why it
Speaker A
identified that specific section, explains how the function works, and can even build something similar for your product if you wish. With many modern desktop apps, this even works without additional tools, but for classic program files, Rea often needs
Speaker A
tools like Ghidra running in the background. So, feel free to try it out , but it's best to only use it on your own programs or open-source software, as things can quickly get legally tricky with purchased software. More on
Speaker A
that in a moment. You can find links to both tools in the video description.
Speaker A
And games are just the beginning, because it goes a step further now. With games, the AI reconstructed the blueprint from the program file. But by now, the models are so good that they often don't even need that anymore.
Speaker A
They simply observe how a program works , for example by controlling the computer and taking screenshots, and then just rebuild it themselves. You can best see that in these Adobe clones right here. The developer essentially used several Claude code agents, all
Speaker A
based on Claude 3.5 Opus. He recreated Adobe programs by simply telling them to look at how the program works. You can find it on my computer, so go ahead and build it. And the whole thing only took about a week. That is totally
Speaker A
crazy, and it’s not even illegal because he didn't decompile these programs. This means the AI didn't touch a single byte of that Adobe code.
Speaker A
It just looked at what Photoshop looks like and what it does with every function. It also read Adobe's public documentation of the file format and then rewrote everything from scratch.
Speaker A
This is also known as a "clean room" approach. It’s basically a reconstruction from the outside. And as I said, all these programs are open source; you can find them here on GitHub. You can take a close look at
Speaker A
everything that's in there. And the cool thing is, all seven of these Adobe clones also have a built-in interface for AI agents. So Claude can operate them itself. If you're wondering now if you can cancel your Adobe subscription,
Speaker A
the answer is partially yes—it depends on what you want to do with it, as some features are still missing as of now. Some keyboard shortcuts don't quite work yet, and features like Generative Fill, which is an extra function that runs via Adobe's servers,
Speaker A
don't work here, of course, but most basic functions are there and are already enough for many things. And the project will continue to develop. This is just the state after one week, during which over 300,000 lines of code
Speaker A
have already been created. I can easily imagine that in a year, you won't even need the regular Photoshop anymore. So that means there are two ways to recreate a finished program. Either the AI reads the blueprint back from the
Speaker A
program file on your computer, like with games, or it watches the program while it works and recreates it from the outside. Just like with Photoshop.
Speaker A
In the end, both methods result in a program that behaves like the original and that you can modify however you like. Legally speaking, the one way involving decompilation is, of course, forbidden, but the other way—simply observing how a program works and
Speaker A
recreating the features yourself—is, of course, permitted. Only the code is protected, not what a program can do.
Speaker A
However, you are, of course, not allowed to take the name and the logo. What does this mean for the future now?
Speaker A
I believe we'll see significantly more open-source projects and replicas of paid software, because any program running on your computer can now basically be recreated if someone puts in the effort with AI. It's quite possible there will be a
Speaker A
counter-movement to this. Companies could move their programs more and more to the server so that no program file ends up on your computer that could be taken apart. But that still won't stop the replicating, because you can still
Speaker A
see what the program does on your screen. And that is exactly what is enough for AI nowadays to recreate the functions. And of course, we are still at the very beginning, but you also have to imagine that all of this is
Speaker A
happening with the worst models we will ever use again. If a single developer can manage a usable Photoshop replica in a week today, what will be possible in two years? So, what does this actually mean for you and perhaps for
Speaker A
your company? First of all, you will become significantly more independent from software vendors in the future, because assuming a small company pays several hundred euros a month for a program where they only use a part and are still missing features, creating
Speaker A
their own version or a stripped-down one can now pay off, because the effort to build it is no longer so high thanks to AI. But in my view, the biggest change will be with outdated software, because in a study by the market
Speaker A
researcher Lünendonk, 62%of the companies surveyed say that parts of their business-critical software are so outdated that they need to be replaced.
Speaker A
Many of these programs no longer receive any updates at all, which is extremely obstructive for companies because they can't address things like automation; their only option is then to control the program via computer use . But there are no APIs, no MCP
Speaker A
connections, or any other interfaces that agents can use to access the program. But AI can now largely take over this exact process of understanding and recreating outdated software, specifically in the way we just saw with Photoshop, right? Claude
Speaker A
simply operates the old program and records what it does, what inputs it receives, what outputs it provides, and what rules lie in between, and no one has to look at the code for that.
Speaker A
That's why this is also the legally clean way to do it. And just to make it clear, because I know it will probably come up in the comments, I'm not saying you can just easily rebuild every program with AI, not that. But the
Speaker A
effort required to develop your own software, even using other software as an example, has become extremely low thanks to AI and it will get much easier in the future. That is why companies can now ask themselves whether they should invest the time to
Speaker A
tackle such a project, which was completely out of the question before. I would be extremely interested to hear what you think. So feel free to write your opinion in the comments, and if you want to stay up to date on current
Speaker A
AI developments and also learn how to apply AI practically, beyond just prompting in ChatGPT, then please subscribe. It is completely free of charge for you. And with that, I would like to thank you very much for watching. I would say we will see each
Speaker A
other again in the next video. See you then. Yeah.
Topics:AI software recreationsoftware decompilationAI automationClaude AIGPT-4oGemini 1.5game modding AIopen source softwaresoftware reverse engineeringAI in gaming
![RESIDENT EVIL 2 REMAKE [VOL. 3] — Transcript](https://i.ytimg.com/vi/0NkrSvFQVtg/maxresdefault.jpg)
