# GitHub’s top: Copy any app, create gorgeous videos, 3d print anything & more

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=kqn_mSjSWMs
Duration: 27 min
Language: English
Words: 5566
Transcript page: https://viewrankai.com/tools/youtube-transcript/kqn_mSjSWMs

---

[0:00] I'm going [snorts] to show you how to use a tool that will copy any app for you. The NSA, the government agency, has a tool that will help you find passwords inside of apps. I'll demo that for you, too. I've got a way for AI to make beautiful videos for you, and a way for AI to make beautiful diagrams. All that and all the top repos of the week coming up. Presented by Zapier, the AI automation company. And then the first

[0:21] repo this week just absolutely took off. Look at this chart of how many stars it went. Look at that. Huge, right? Quarter million stars essentially in a week, and it's all because this is going to let people find an app online and then copy it and make it their own. This thing actually went into the popular press, tech press talking about it, people in the news are talking about it. Everyone

[0:43] is talking about it. What do you think of this? I mean, I bet it works. That's the first thing. It's It seems very cool technically. Mhm. Uh I haven't tried it, but I I've talked to uh you know, other people. I've seen the news just like everybody else has seen. Uh It's a little sketchy, right,

[1:01] Andrew? I don't I'm not sketched out by this. I think it's inevitable and I just have to accept it. Let's work with it. I'm not upset that this can copy everything. Yeah. Look at how uncomfortable you are. Yeah, [snorts] no. I No, I'm thinking about it. I I think it's uh you know, as I as I think about it. I actually went and looked like some of the examples we've seen are are people uh taking the Adobe products and then turning them back into other stuff. I actually went and looked, you know, when you download section 17 in the terms of service for Adobe is no modifications, reverse

[1:34] engineering, artificial intelligence, blah blah blah. Like has all this list of stuff. So, you can do it. You definitely can do it. It's maybe this is in a similar category as like downloading movies illegally. Mhm. You know, and and you could have your own kind of integrity. Like where's where's the line, right? That's sort of the question. When I first saw this, it's like, oh, is this like a black hat

[1:52] thing? Oh, is this a gray hat thing? Like then I went like, well, what are the white hat things? Like what are What are the reasons someone would use this which are very legitimate, very good?" Um yeah, hit me. Here, this is from our buddy Matt. He showed a bunch of use cases including

[2:10] this one. Iltar decided that what he wanted was to vibe code a status bar app for his uh air conditioning unit. In order to do this, he needed to take apart the app as it was and then make his own version of it. I think there are lots of use cases like this. He got out with 2.5 million

[2:26] views on X for this. It's amazing. I you know, I love that actually. I I'll give you my own now. Uh you know, I have a Kia and their app is let's call it sub par and I'd love to have my own integrations with with my car and I own the car. I bet I've agreed to some license that says that I won't uh try to access the API some other way,

[2:47] but I would love to build my own app. I'd love to have a button on my desk that could, you know, turn on my car as I'm walking towards it. Uh maybe maybe I would draw the line at Yeah, yeah, that would be okay. Do I think it would be okay to to reverse engineer Adobe Illustrator and then sell that or give it away to other people in exactly its same form? I

[3:07] think for me the line is is below that. I wouldn't I wouldn't do that. But would I take something that I have that doesn't quite give me all the features I want? I want to modify it, use it myself. Uh yeah, may maybe I'm on board with that. That that actually sounds okay. Okay, the biggest complaint that you all have had for us is that we are not showing actually how this thing is being used and I'm trying so much. So, I have video here that if you download this whole page or get access to the site, you'll get it. I think this is the best

[3:33] video that I've seen on how it works. I also have experimented with having my agent actually install the app like this. I want you to see it how it works. Basically, the agent needed to get um uh here, let me show. I think the best way to show it is here. The agent is saying the game 2048 is after it installed it, it said the agent 20 the sorry, the game 2048 is open in Chrome dev tools endpoint etc. Use REA tools to inspect how the page works then build a simple working copy of the game as one file. It went and did it then he tested

[4:04] to play it and here's the before and after like this and you can see that it works. Super easy to use. so exciting about REA here is that these tools have been around for a very long time. I've even used some for for more engineering and and other things past in my career, but they're so hard to use. It's so detail oriented and people make it their life's work to get good at reverse engineering and there's an art to it, there's a science to it and being able to have REA as this not just set of skills, but also MCP connections and tools that you can connect to an agent,

[4:43] I can see how this is going to bring this to I think anybody that can use an agent effectively could probably start to reverse engineer things today with not a whole lot of training. And that I mean that is like that is game changing. I will say this in Nick's video he had Claude copy an an app, the preview app that comes with every Mac and then he had Claude do it with this tool. It

[5:05] didn't really help to have this tool. We're basically there with the basic LLMs. All right, it's there folks. Give me feedback if you like this and keep complaining about stuff. God knows I love all these complaints. That's how that's how I improve. All right, this one is from last week. Any PS5 PS5 game files become PC programs, no emulator involved. This is the goal, this is the

[5:24] dream. It has absolutely been growing. I don't think we have that much to say about this. It hasn't advanced much more than than it has since last week. We're basically seeing that there's a game that's that's that's being copied this way. Right? Yeah, I I think it's it's very much in line with with REA which is hey, instead of taking the binary, taking the game and try to build an environment that can run the game. Let's just figure out what was the code that created the game. And if you figure out the code that created the game, you can change it, you could run it on any platform you want. Uh so,

[5:59] this is this is a reverse engineering effort as well. Now, the problem on PS5 is that all those discs, you know, you get the disc or you get the download, uh they're all encrypted. And so, the big blocker here is actually finding a way around the encryption so that you can even access the the game data itself. And that that's going to stop you, you know, you couldn't just go home and and reverse engineer all your PS5 games tonight. You're going to need to chain together a couple other exploits uh and

[6:26] and get around their copy protection. Number three for the week, constantly returning here. This is Matt Pocock's skills. Um he has a set of skills that he used to to use this to make him a better developer. People have been using this for a long time. It's one of these things that now when someone builds, they just install it and they start working with it. Um I had my agent try one of the skills. This is the uh TDD

[6:52] skill. What does TD stand for? Test-driven development. Okay, so it's that instead of coding first, you build a bunch of test first, and then you build code to try and make all of your tests succeed. You know, you're setting the you're setting the rules of the game before you actually play in the game. Most of us do that backwards. We play the game, we play the game, and then later we might add a couple test on the

[7:17] end. This this flips it on its head. So, I went back and forth with my agent to get this right, and you said, "Andrew, you missed the key part of this." So, let me show what I have, which is, "Okay, the agent went to Claude and said, 'Use TDD skill to build a tiny bill splitter in plain JavaScript. It returns how much

[7:35] each person pays, rounds to the cent.' Anyway, it continues on. It then did it, and then the agent said, "Okay, now test it." And it passed the five tests that the that the agent made for itself using the skill. You said, "Andrew, this is kind of backwards. Why Why did What did I miss here?" Or you missed You said you

[7:53] missed a step here in your visual. you know what what you would see this is a very small example. So so it's easy to kind of push through very quickly. If you had given this for a much larger application for a whole to-do app or or whatever app you wanted, uh what would happen is instead of the agent writing code, it would first write a lot of tests

[8:15] including, "What do I see on the screen? What does the interface look like?" The interface doesn't exist yet. And so when you first run all of the test, it's beautiful. In TDD, all the test fail because there's no application code. You've only written test and then all the test fail because there's no application code. Then the way that the agent would start writing the application code is, "Well, let me look at that first set of test. How would I make those test succeed? It looks like I need an app like a box with some design in it and there needs to be a button up here and that button needs to perform

[8:47] this task." And as the agent keeps working, you'll see more and more of the test begin to pass and the agent knows it's done when all of the test pass. That is TDD kind of beginning to end. I see. So in this visual, we skipped all those fail test fail test fail test and then what we did was we jumped to the end and we said, "Okay, now run this test." And of course it's going to pass

[9:08] at the end of TDD. Yeah. Um I thought the cool part was that it knew what test to make for this like uh it gives the it adds the tip before splitting it so on. I see the point. Here's what it looks like by the way in a visual so you can then split it to 20% tip and the whole thing gets tested uh the whole thing works. And it the agent did actually have a test run in the terminal which worked. All right, if you haven't tried it uh Matt Pocock skills are just battle-tested. People love them. I have a full interview with Matt where he walks through how he used it

[9:40] himself. Let's go on to Zapier. Now, Adam, here's the next bit of feedback that people gave us. They keep saying, "Andrew, I already know what Zapier is. Stop telling me." So, you know what I'm going to say, guys? They helped me get started with this. They basically funded a year of all the interviews and all the videos that we do. If you want to, I would love it if you thank them in the comments and let them know that you heard about them on here. And I just appreciate that they said, "Andrew, go run wild. Do what you can. Whatever helps people build with AI, we love." And so, I appreciate them

[10:05] for doing that. I'll move on to the next, which is HyperFrames. You write a webpage and then or you give it to HyperFrames via your agent and then you get a video back. I wonder how important this is going to be going forward, Adam, when honestly all the all the models are just giving

[10:24] me video now. What do you think? know. That's a good question, Andrew. I think as we look at these tools, this is a good example of a tool that that Claude and Codex and others are are going to eat. You know, they they want to build this stuff, too. I I I think what I like about having it broken out in a toolkit like HyperFrames is that you have a lot more control. A lot of times when we see that ChatGPT or Claude or whoever, they

[10:53] they they'll build in a feature set. Like, let's take Claude design. Claude design has a lot of really interesting skills, but it's also very opinionated and it's difficult to customize its way of thinking. And so, breaking it out tools allows for a lot more creative control. Now, there's a difference between a tool that helps beginners produce something that's kind of good and a difference, you know, the other flip is take someone that's kind of good at it and I want to make you great. And I think HyperFrames is probably in the second category where they can take people that know what they're already doing. They've already got the frame,

[11:27] but they want that control. They want to be able to tell it what to do on every step and they're going to be able to create great things, whereas everybody else, all of the rest of us beginners, we we get to, you know, go go to go to Claude and and make a cool video. Again, I've got a video here in the uh download or the clickable, and this is us using it. Basically, we installed

[11:46] Hyperframe. Um Uh we needed Node 20 Node.js 22 to upgrade to that, and then we said, "Okay, let's The first step was cre creating the HTML. Here it is, filling it out with information. And so, it said, uh the agent said, "Using Hyperframes, edit this index file in this folder into a 10-second vertical YouTube Shorts promo for Andrew Warner's show, GitHub uh hot repos episode for October 9th, which

[12:18] happened, of course. Here's what I want. The Beats Kinetic titled GitHub uh hot repos quick countdown." I won't go you take it through the whole thing. And then, you say to Hyperframes, "Now render this, turn this into a video." And here's what that looks like. Um There you go. Nice setup. And um I feel like maybe going through all of these different terminal commands

[12:43] is slowing things down. I'd like to hear what the audience thinks, but if you're feeling that it is, I think we should just move faster through this. I won't take everyone through all of these. Let's [snorts] go through number five. Five. Katherine Lavery, another person who had been on the podcast here doing a show with me. She's got a system here that creates beautiful diagrams that are made for your brand. 55,000 stars. It

[13:06] continues to grow to to grow. She's a designer. She's not even a coder, but she's just been loving through AI all this uh all the world of code that's been opened up to her. What I like about this is how fast it makes or how well it makes designs. Uh here, you can see a bunch of examples like the state machine here, IT current state flowchart. And

[13:28] you can customize it to your own needs. What do you think? How would you use this? Yeah, it's it's the same way. You know, I I look at a lot of these designs here and I think, "Wow, like they all look so simple. Why can't AI just do this without a package?" Yeah. go and ask AI to create charts like this, oh my god, they're so complicated and ugly and there's so much flash and it takes a lot of work to talk to the AI to get it to help find what is the simple way to communicate this. So, I I I see this is actually a way to

[14:02] constrain the AI into a set of charts and graphs here that that just kind of work and then they're customizable. And instead of starting from something that's really ugly and big and spending all your AI credits trying to get it to look nice, you start with something so simple and then she's got the tools in here to customize it. Oh, you want to layer on your brand, you want to make this flashier, you want it to be pencil drawn instead. There's all these different tools to help template and and customize the style. So, this is just a a better building blocks kind of kind of

[14:34] way to get there. Really great tool. I'm glad to see that she's doing well with this and that it continues to grow. Next, Ghidra. AI agents now drive the NSA's decompiler. This one I actually do want to show. So, it comes from the NSA. I do want to show how it works. So, what we've got here is we put Java on the system, then the agent created this app called the vault, and then it tested it with a with the wrong password to open up the vault and it said, "Nope,

[15:08] you cannot come through." Then, it said, "Okay, I want you to decompile this." And it did. And one of the lines in the in the code is this. This is a hex. Is that what it's called, Adam? Yeah. That's And then it said, "Okay, print what that is and here it is the code. It is 4271 and then it used that code to open up the vault that it made and sure enough it worked. And that's the thing that you get. You take this and it decompiles it into this. Are you What do you think about this? Are you

[15:39] against this too? Well, so this is one of the This is one of the three decompilers that RII uses. So, this really This and RII are hand-in-hand. RII drives this tool. This tool is difficult to use. I mean, you we saw the commands there as you were just demoing. There's a lot of detail in how you tell it exactly what you're trying to analyze and where you want the code to go. And that's the That's the hard work I've told you has existed for years and RII takes it and says, "Well, here's how you use this particular reverse engineering framework." And the And the agent can

[16:14] help you through it. So, the reason this is popular is because RII is popular and people are downloading this one because it's publicly available. It's open. It's free. And then you can use RII to drive it. But I But I don't think anyone is like learning this tool from scratch today and actually typing the commands.

[16:31] It's just agents. Ah, [sighs] next. We just had a meeting before this where we said, "Hey, let's start selling ads for 2027." And we are and if you're interested in in getting an ad here, talk to us. That's the email address, the same one you've seen me give up throughout. Love to hear from you if you think you want to reach this audience. And we We've got a growing audience. Some of these videos, Adam, are getting hundreds of

[16:53] thousands of views. Can you imagine it? That's awesome. I know. I feel so much love and it's exciting, too. Yeah. Thank you all. All right. Next, billion contexts keep one AI out in one AI coding session going for weeks, not hours. Maybe you could explain the problem that it's solving and then tell me what your opinion is cuz I know

[17:14] you've got a strong opinion on this one. Yeah, the problem is pretty simple, which is all these agents we're talking to today, they can only look at two, 300,000 tokens at a time. You know, at the tops you see maybe a million tokens. And a token it's not quite accurate, but a token is more or less a word, right? And so if you can only give the agent 200,000 words, well, what can you ask it? You can't ask it about everything. It can only look at that 200,000 words. That's a couple books in a lot of cases. So, we're all out there and we've been out

[17:49] there saying, well, how can I feed more context into my agent? I have more data, I have more context, I want my agent to understand it all. And there are a lot of tools like this that say, well, I can take your data, I'll compress it down so that we can use fewer tokens to represent your data, and now the AI can look at everything all at once and give you the answer you're looking for, the summary, the result,

[18:13] the math, whatever. And so, why don't you think this is incredible? Well, the problem is compression isn't, you know, I'll use a tech word here, compression isn't lossless. And lossless means all of the data is is, you know, restorable. If I compress, if I take the image of you here and I compress that down and and then I take a screenshot while you're only this big right in front of my face and then I try and stretch that out to fill my screen again, you and I both know what we're going to see. It's just going to be a mess of pixels and colors and I won't be

[18:51] able to see you at all. Well, that's compression. You took it, you made it really small, so well now I can fit thousands of photos on my screen. When you make them big again, they're not there. So, if you give the AI compressed data and then you ask it detail-oriented questions, it simply doesn't have the data to answer the detail-oriented question. And they even call it out at at the beginning here. The The way the paper works or the summary of it is they compress that data by using very intelligent summaries of of putting it together. Here, stop right there. The model decides when and what to compress

[19:26] into high-fidelity summaries, not a hard truncation limit. That just means they're they're trying to do basically compaction better. That's the compression that most of us are familiar with. You get to the end of a of a chat session, it says, "All right, I need to compact so that we can keep talking." And what that compaction does is it looks it's the AI looking at the whole conversation history, saying, "Ooh, how would I summarize all of this so we could start with a fresh window?" And and, you know, take all of, you know, all 100,000 tokens and compress them down into the 10. Uh but it's not

[19:56] perfect. We after you compact, the AI forgets things, it loses track. Uh there there are like very clear big downsides to that. So, I'd be careful using this thinking, "Great, now I can just feed in a billion tokens, everything will be perfect." Uh that's not the case. I like the way you analyze this. By the way, we should have a link to Adam's LinkedIn so you can follow him there where you can see what else he's working on. Um plus whatever he's doing Gateway X, they've been investing in really good companies, and I'm excited to see I I just heard from Jesse, your partner in the business, about what you all are

[20:25] doing and what the companies that you're investing in are doing, and it's so exciting. I can't wait for you to be able to talk more publicly about it. Okay, till then, we're going to talk next about text to CAD 20,000 stars only, but man, is it impressive. Basically, you talk about a part that you want, and then you get to send it to your 3D printer and get it. I do want to show this one because we tried it, and you said to me before we got started, "Show

[20:48] the the input output." And here it is. Use the CAD skill to make a simple phone stand I can 3D print a base plate with a slanted back rest and a lip at the front to hold the phone. Export as STL and save a snapshot image of it so I can see it. And here it is, it did it, and it looks good. Obviously, you could adjust it beyond that, but it was a very simple prompt. The thing that I thought was interesting was Oh, first step is it took telemetry off and I said, "Why did you do that?" It said, "The readme recommended it and I didn't want to have

[21:15] our information going out, so I took telemetry off." And then it said, "Okay, here's the the uh the command that it gave it. And I was wondering, how did it know how much of a cord tolerance to give, what to what to do?" And essentially, that's what this skill does based on so much experience. And um And there's so much information about how how thick a cord is for a for a phone out there. And then it ran these tests afterwards and it worked out great. I love this. I'm glad this is doing well. I want more of it. What do

[21:47] you think? Yeah, I think I think I'm like waiting for the the next heyday for 3D printers. Uh the biggest problem with 3D printing is it's very difficult to create these designs. And I I have a 3D printer. I print stuff constant I have a print running right now in my house. And I'm

[22:02] using AI for all of my design now. Because you can take measurements, you can give feedback. Even that one stand, if you just went back and said, "Oh, this is great, but I want all the corners to be rounded." Yes, I was thinking that, too. to be even more softly rounded. It'll do that, too. It can take all of here's what I want, here's what I think it should be. And then think of AI or 3D printing is is like pairing very well with AI, which is maybe you want to make 30 of these. Maybe you want to sell this as a product, but you don't want to just

[22:32] have to create the whole thing. How is it cheap to test it quickly? Well, you have AI build your design, you 3D print it, you look at it, you touch it, you decide you want to change it, you throw that one away, you go back to your AI. I'll even take pictures of the product with my with my phone and send it to AI and say, "Hey, here's what it actually looked like in practice. Here's the cord not fitting. What do you think we should do?" And it'll come back to Oh, I see that. Yes. Uh take a little measurement for me. It's amazing. So, I I think

[23:00] everybody needs to go out and buy a 3D printer, uh and you should all be using AI to CAD up some cool designs. I'm excited about this one. Next, Swift UI agent skill. You laughed as soon as you saw this. How would you explain what this is? Yeah, I I didn't even click into it, and

[23:18] I was immediately I I in love. Anybody that's writing Swift, and that has been for a while, knows that the documentation that Apple puts out is just it's out of date, it's terrible, it's lacking details. There is no love for the documentation for Swift. And unfortunately, AI has been trained on the documentation for Swift, for Rust, for JavaScript. And when the docs are bad, the AI is kind of bad at writing it. And it's out of date, you say, "Oh, like iOS just or yeah, iOS just updated, and

[23:51] they updated Swift along with it." They're constantly turning out updates. How does the agent know what should actually work today? And I've watched the agent work on this. It'll try and fail, and try and fail, and it's because it's got old knowledge. So, this is taking a better set of documentation along with all the changes that are happening kind of in real time, and helping you get that information to the agent. So, the agent doesn't spend nearly as much time just struggling to

[24:16] get the thing working. All right, we tried it. It's pretty straightforward trial. I'm going to skip this one and let people download it. Uh finally, number 10 for the week. It is back 12.3 thousand stars, gym tracker you own with 5,600 exercises built into it. I I'm now thinking, especially having gotten feedback from the audience, I don't know how important this is. Why is

[24:41] it still doing this well? What do you think? Yeah, honestly, I think there's a lot of people out there that are working out. That's the biggest thing. It's It's a very It's a game a lot of people play, and they want to be able to win the game. And a lot of the apps that you would download to help track your exercises, that app actually is designed for a different game, and that's to make that company money. Right? If I download somebody's app, they're going to show me ads, they're going to recommend products, they're going to try and push me towards their partners. But no, I

[25:13] want to see how much I can lift. I want to see how far I can run. I want to see if I can lose weight. And something like this that's open source, that doesn't have all that other crap mixed in, that is a is a refreshing experience if you genuinely want to play the game and you're not there to buy a product. So I I I think these are these are going to be pretty good competition for uh for all these apps we've seen for years that are a couple bucks to download or that are constantly showing you ads and trying to get you to download something

[25:39] else. Okay, fair point. I do like it, and geeks working out are my favorite because I am a geek and I don't relate to the people at the gym, but I do relate to people who are number crunchers. All right, it is there, and uh I will say my goal, Adam, for the year is to get to 100,000 subscribers before the end of the year. I think we're on track to do it by the end of January,

[26:01] February. I'm really eager to push it. I've been talking to my family about it, and someone said, "Why aren't Are you talking about it publicly?" And I said, "No, I don't want to be one of those people who keep saying subscribe, subscribe, subscribe." Every time I do, I feel guilty or weird or just pushy.

[26:16] I said, "Well, why are you telling me? Why are you telling your kids? They can't do anything about it." So I said, "All right, I'm going to try it." And the more we can get people to subscribe here, the more I can get Adam creators who've been on here we featured to come back on and do videos where they really go deep dive into what we do. And we we've gotten quite a bit of these um I want to get more of them. So if you if you're into it, subscribe and you'll get more of these videos. And if you like this one, I got another one for you to

[26:40] watch right here below. See you in that one.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt