1. Podcasts
  2. Denoised
  3. Is Software Dead? Inside the Vibe Coded Creative Suite, Office, and More
Denoised

16 Oct 2026 · 37m · 15 chapters

Episode artwork

Is Software Dead? Inside the Vibe Coded Creative Suite, Office, and More

In short

Whether “software is dead” in the sense that reverse engineering and “vibe coding” can clone or generate working apps, plus how AI model platforms defend themselves and accelerate creation (with examples from REA, Adobe-like clones, and agentic 3D generation in Blender).

Guests

Addy and Joey (co-hosts). No other guests mentioned.

Guest backgrounds

Not specified beyond their roles as podcast hosts discussing LA Tech Week/AI events, creative software workflows, and building/editing pipelines.

Key claims

Reverse engineering tools like REA can decompile binaries into human-readable code and recreate apps; clones may lag on performance (GPU/video real-time preview). SaaS may survive via reliability, maintenance, and proprietary data/workflows. AI aggregators defend with proprietary models or prompt/UX layers; agents + scripting-friendly tools (Blender/Resolve) accelerate production.

Notable examples

REA “reverse engineer anything”; “Crafting Apps” (PhotoCraft/FilmCraft/LightCraft) as Adobe Creative Suite clones; PhotoP and “cracks”/ISOs historically; Blender agent demos: Gemini-style pagoda built from a single prompt, cathedral from a single image, and a laundromat example (Opus 5.5). Tim Sweeney pivot to AI acceleration for Epic.

Written by AI. May contain mistakes.

Chapters

0:13

Tech Week Thoughts

Hosts discuss their experiences and opinions on LA Tech Week and its focus.

“I'm finishing up some projects, but it'll be out.”
1:11

Reverse Engineering Software

Discussion about REA and the implications of reverse engineering software.

“A couple of things dropped in the last week or so.”
2:36

Vibe Coding vs. Traditional Apps

Exploring the rise of vibe-coded applications and their effect on traditional software.

“And someone sort of built a free version of it.”
4:38

The Future of Software Development

The hosts speculate about the future of software development and the shifting landscape.

“Like they did reverse engineer current things.”
7:20

Building Custom Solutions

Discussion on the ease of building custom software for personal needs.

“duct tape thing together and support that and keep iterating and building on it.”
9:22

Challenges of Software Reliability

Hosts discuss the reliability of software and concerns over vibe-coded programs.

“We don't have computer science backgrounds and we don't program every day of our lives.”
12:53

Customization in Software Interfaces

Exploring the potential for customizable user interfaces in future software.

“But yeah, those are going to be questioned more.”
14:00

Generative Interfaces and User-Centric Design

Exploring the evolving landscape of software interfaces and customization.

“where it's like, okay, do you want it like light mode or dark mode?”
16:41

The Math Breakthroughs of AI

Discussion on AI's role in solving complex mathematical problems and implications.

“Yeah, I mean, also I feel like as humans, we will always find new unanswered questions for the next thing to figure out.”
19:50

The Business of AI Models

Analyzing the business strategies behind AI model development and licensing.

“they took the open weights and then they post-trained and made their own better version of it.”
Show all 15 chapters
22:24

Speculative AI Business Models

Speculating on potential business models for AI companies and products.

“But then somebody who's just getting started on AI will use it.”
25:02

Updates on New AI Models

Discussing the latest AI models and their features and applications.

“By the way, do not come to denoise for business ideas.”
28:05

Exploring AI Capabilities in Creative Processes

Learn about the advancements in AI that allow for the creation of complex designs from simple prompts.

“So it's building it from a single prompt.”
29:51

AI in Architectural Modeling and Game Development

Discover how AI is transforming architectural modeling and its implications for game development.

“But regardless, the fact that you can get to this level of quality and fidelity from a single prompt is just nuts.”
33:11

Tim Sweeney's Vision for AI in Gaming

Understand Tim Sweeney's perspective on integrating AI into game development for enhancing creativity.

“I mean, I think if you're any type of game or software developer, if you're not thinking of AI first, you're just going to be left in the dust.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On Denoised, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Addy: There's a level of people where it's just like, I just want to pay for something that works. You and I both have coffee machines at home, but sometimes we just go to Starbucks.

0:07Joey: It's just easier to just pay.

0:13Joey: Welcome back to Denoised. How's it going, Addy? Long time no see, Joey. Let's go. I know. I know. I'm finishing up some projects, but it'll be out. Actually, by the time this is out, the trailer for it will be out. Oh. I'll talk about it next week. Okay.

0:28Addy: Have you been doing any LA Tech Week stuff?

0:32Joey: I went to the AI on the lot event, but I haven't. That's about it. What about you?

0:37Addy: Nah.

0:37Joey: I still cannot fully wrap my head around what exactly Tech Week is besides panels and parties that are loosely organized.

0:45Addy: You're right. To me, it feels very VC-driven, like A16Z. Yeah. That kind of thing where it's like get a bunch of founders together, figure out some new ideas, so we can go invest in the thing.

0:59Joey: Yeah, I don't know. Everything I've always gone to with these is like a zoo. It's just like over packed too. Yeah. Okay, speaking of building and whatever the future of software means. Oh God. A couple of things dropped in the last week or so. One of the big ones is REA, reverse engineer anything. So crazy, man. It's giving me the chills just looking at this. It's a FCP tool for basically kind of pointed at an app, but it can reverse, or I believe a website, and it can reverse engineer how it was built and then build, vibe code your own version of it.

1:36Addy: That's right, yeah. From my understanding of it, it gets into the post-compile binary, and then it can reverse engineer that to human readable code. So imagine all the ones and zeros turned back into C++ code. I don't know how it does it. It's crazy because decompiling is like incredibly difficult. And it does it well enough to then recreate the application where you can take over and rebuild it or, you know, modify it to your will.

2:06Joey: Yeah, the probably biggest example going around with this. I don't know if they use this or the person just vibe coded it, but someone built, it's called Crafting Apps. And there is a PhotoCraft, FilmCraft, LightCraft. It is basically a vibe coded version of Adobe's Creative Suite and other products and other Vibe-coded apps. This has been probably the biggest example going around of just like, oh, what are you going to do if you're Adobe or any other major software? And someone sort of built a free version of it. Now, apps and heavy-duty stuff like this, this feels like the last thing that I would jump ship or switch on and change my entire workflow to a vibe-coded app.

2:53Joey: Also, there's been free video editors and other tools out there forever. There's PhotoP, which has been online forever, which is like a web-based version of Photoshop that's ad-supported and free to use. If you're someone in high school, back in the day, you found the codes to unlock Creative Suite. I don't know what you're talking about.

3:14Addy: Actually, this was even before Creative Suite existed.

3:16Joey: This was just purely when you had to get the code for the software.

3:19Addy: And the floppies, even.

3:20Joey: Yeah, code's floating around the internet. If you're in middle school, high school, and you didn't have any money to pay for expensive software, if you're dabbling. But if you're in an enterprise company or a company where you need stuff to rely on. Yeah, yeah.

3:36Addy: Well, okay. Going back to the 90s, I think it was called a crack, right? So when you would download the ISO, the ISO is what this is. You have to download the ISO, which was the disk image. Which is the binary code, yeah. And then the ISO would mount to your hard drive like a disk. And then you would crack the license activation schema. And off you go. You have Photoshop on your Windows 95 box or whatnot.

4:09Joey: I was going for Premiere. Premiere 6.5.

4:13Addy: Oh, okay. Wow. All right. You were into video long before me, my friend.

4:17Joey: Yeah, I was doing video first.

4:19Addy: Yeah.

4:19Joey: The bigger existential thing is just like, okay, software dead, SaaS dead, if you can just sort of build. These are clones of other things, but if you could just build whatever you need, you need a video editor, you need it with whatever specific features you need, where does this go?

4:33Addy: Yeah, I am, okay. I have two trains of thought on this. So we could speak specifically on the Photocraft VectorCraft, like this thing. I believe this was REA devised. Like they did reverse engineer current things.

4:47Joey: It's got WordCraft kind of like. Yeah.

4:48Addy: And that's why they were able to nail the UI so well, like almost pixel to pixel, because it is just literally the ISO file, the binary file. But from what I am hearing, the performance is severely lacking. So like a lot of the magic in Premiere and even in Resolve, a lot of the NLEs is like how well it handles low level video computations on the GPU. right like when you're doing a time remap or when you're putting a transition between two shots that's like video calculations happening in real time for you to preview it and then when it renders it renders right like that stuff is not working correctly so you can load your timeline but the preview will be slow and choppy and then some of the transitions won't work and like I've seen a couple of videos on YouTube where you know they use video craft and it's like oof oof that's not quite there yet.

5:46Joey: Yeah, I mean, I'm not surprised that there's performance lagging and issues. It's not there yet, but we can see where this is going, obviously. Exactly, exactly. Where regardless if you have to point it at a working app, I mean, this is like the Higgs field genjitsu version of film where they take an existing film and restyle it. Oh, I hate that, I hate that.

6:07Addy: Yeah, like this is the same thing

6:09Joey: where you're pointing it at an application that has had a lot of development and works and then copying it.

6:15Addy: Right.

6:15Joey: But put that aside and it's just like, I need an editor, I need a text editor, I need something, let me start building it.

6:22Addy: Yeah, we tend to forget. And optimize it. How crazy all of this is because we're just inundated by one development, one progress, one thing after another, right? Like, wait, this was like NSA technology like five years ago. Only like the highest levels of the government do they have the technology to de-reverse engineer a piece of software that China built or whatnot. This was used for geopolitical chess. And now anybody has it. Think of an application that you use every day. Of course, we use these applications every day, but also stuff like, I don't know, like Mac OS or something crazy like that.

7:06Joey: And you can reverse engineer that probably. I mentioned about the enterprise thing before. I think there's still a level. Well, it's like, okay, so maybe just software development evolves or the companies evolve where, like, because you can move so fast and you build things so fast. Obviously, not everyone's going to want to vibe code their own solutions and have some duct tape thing together and support that and keep iterating and building on it. Even as good as agents get, like, you still need something to guide it or just make sure it's all working. And it's like, maybe that's what software companies kind of turn into as the just - The maintenance.

7:37Joey: They can run leaner and faster and push stuff out. But yeah, I'm also thinking of like Linux, right? Linux has been around. Linux is free, but they, I believe it's Linux. I mean, they make a bunch of money off of supporting and running the servers and like the professional services. I see, I see.

7:59Addy: Oh, I see. Interesting. Like you think that's the business. That is no business model then. Like it's very slim pickings at that point.

8:07Joey: Well, I mean, the other argument, or like, okay, if the software is not a moot anymore, it is what you build. Right. Something else has to be the secret sauce, which is proprietary data or something else that can't be replicated.

8:22Addy: Yeah. The one thing I will say is we're forgetting to mention that this makes theft really easy. Like, that's no bueno in my book and your book, too, I'm sure. like shouldn't be stealing software just like you shouldn't be stealing uh data for training

8:38Joey: models and stuff like that yeah i mean look obviously if you're pointed at like i don't want to keep picking on if you pointed at resolve you pointed at final cut you're like copy that and make me a version of that dude it's like okay yeah that's like ripping it off but if you're just like i need a video editor yeah that does x y and z i know it is you know completely capable of starting to build stuff like that, which arguably could be better because it is like, okay, you're building a custom piece of software built exactly for what you need to do and your specific needs.

9:10Addy: Exactly. Like for this podcast, we don't need the entire might of Resolve or Premiere or Final Cut. Like we just need a couple of things to make this podcast edited, right? Yeah.

9:21Joey: I mean, honestly, there have been gaps in editing this podcast because even with these faster podcast editors, they don't really support, like I so many times have a conversation where it's like okay we have your feed we have my feed I need to put them like split side by side right and I need that to be treated as an angle to cut to and a lot of these auto cutting things like don't understand that they don't create a third

9:41Addy: third camera that is the two yeah it's like I need a fake third camera that is a two shot of us

9:45Joey: and I have yeah on the verge of just being like oh let's just I'll just build it just build something

9:51Addy: Well, that's also the incredible moment in humanity that we're going through is that the average Joe, like you and me, can just build something. We don't have computer science backgrounds and we don't program every day of our lives. But we're going to build something that requires thousands and hundreds of thousands of lines of code and we can do it. Crazy, right?

10:14Joey: Yeah. Yeah. I mean, it's been a crazy unlock. 2026 has been an insane year. 2026, right? Yeah.

10:19Addy: And we have more stuff to cover as we'll go through. Like it's just getting started. Like the entire self sort of like just being able to build an entire tech stack yourself plus make content plus, you know, do all these other things is insane to me. Like the power we have at our fingertips.

10:42Joey: Yeah. You know, I mean, we're tinkers. Tinklers. Tinkerers. Tinkerer is like I was running through five different versions of that word in mind I'm like that's not the right word Yeah your NyQuil is hitting real hard right now You know obviously there's a level of people where it's just like I just want to pay for something that works I don't want to deal with all of this even if as easy as it gets

11:06Addy: That's the thing like you know you and I both have coffee machines at home but sometimes we just go to Starbucks it's just like

11:12Joey: it's just easier to just pay for expensive coffee I thought we were going with that. We have coffee machines at home, but sometimes we just want to have a nice pour over and grind the beans ourselves. But you went the other direction.

11:25Addy: The option that you're getting now is you don't just have the pour over and the grinding, but you can grow the coffee bean. That's how crazy things are getting. You can really go back to the source. But there's nothing wrong with going to Starbucks and just paying$6 for giant venti, whatever. Yeah, I need coffee. And, you know, I don't want to.

11:47Joey: Yeah.

11:47Addy: And I think like a lot of the software remote in that sense, I think will be undisrupted is because they're low priced enough to where it's negligible for the everyday person. You know, if I'm not going to maybe speak on like VectorCraft and this stuff, but like Resolve is$300 for the lifetime of the license. Yeah.

12:09Joey: It's like, yeah, that's a no brainer. And all of this dialogue of the software, if I didn't see Resolve come up at all.

12:16Addy: Yeah, they're like, no, no, leave that alone. Yeah.

12:20Joey: Also, just basic Resolve is free, is literally free and download,

12:24Addy: and you just know it works. Yeah, I think the companies that are in trouble is maybe not even Adobe, but if you go up higher in the food chain, AutoCAD licenses are really expensive, as they should be. It's a highly capable tool. You look at Maya. Maya is, I think, something like$10 ,000 a year on an enterprise seat, if I'm not mistaken. But yeah, those are going to be questioned more.

12:58Joey: Right. But something like AutoCAD is like, okay, are you going to rely on designing a multimillion-dollar, hundreds of millions-dollar building infrastructure? Oh, my God.

13:09Addy: Oh, God, can you imagine?

13:11Joey: Vibe-coded software.

13:13Addy: Like building the Burj Khalifa in Dubai with Vibe-coded software. Oh, I was off by 10 millimeters.

13:19Joey: You're absolutely right. I think for some of these, it's more a conversation of how does AutoCAD or Autodesk improve their internal process, where it's like, okay, you can just ship and move stuff faster, but still have the integrity of reliable software.

13:35Addy: The other thing the SaaS companies do that is maybe not public knowledge is that they're willing to build stuff into those tool sets for you for a cost, right? Like if a studio needs some custom animation tool in Maya, Auditor will probably build that for you. I'm just giving an example here. Like in the past, but that would have been like a enterprise deal. That would have been like, yeah.

13:58Joey: I mean, I've been wondering too, even just on a user level where it's like, okay, you build the software, you have user preferences. where it's like, okay, do you want it like light mode or dark mode? Do you want your sidebar? You know, what little widgets do you want on your sidebar? In this future of just whatever generative interfaces look like or something, it's like maybe the backend and everything is still the same of how it operates and works, so it's multi-user, but just maybe the entire interface is built around whatever the specific user needs and how they like to operate and how they like to work.

14:30Joey: And it's just a whole different paradigm of like what we've been used to of like, okay, you got to figure out how the software works and like you're stuck figuring out where the buttons are. And now it's like, you don't want the buttons there? Delete it. Yeah, move it. We'll move it.

14:43Addy: Build the buttons you need. In defense of traditional SaaS, I will also say like, I just thought of side effects and Houdini, right? Like Houdini is such a well-respected tool in the industry. It's really magical what they've done over the years, right? Like pre-real-time rendering, that was the only way to get like fluid sims and smoke and particles and collisions. And even outside of that, you can do animation rigs. And you can really do everything in one package. And the reason Houdini is successful is because SideFX, who makes it, has scientists and researchers and PhDs working away, chipping away for years on this stuff.

15:21Addy: So are you going to vibe code something like Houdini? You're not. Because that company is specifically built to deliver on something that the normal person can't.

15:32Joey: well I've like sort of wondered about that too of like okay where are these things where they took like really complicated like math breakthroughs or other things and and then I didn't read the whole full I didn't like dig into it but I kept seeing the headlines of like the there's like mathematicians at universities that math problems that have been unsolved for ages and now it's just like Fable or Astra it's just like solving them

15:57Addy: and I've seen

15:59Joey: it's like what X

16:00Addy: They're just like, what the fuck? Right? The thing that is even more perplexing is that they're leaving you with the, the AI models are leaving the mathematicians with the answers and sort of how they deduced it. But then they have to go back and prove it. And so like it's homework on them to then try to prove this impossible math. So it. Homework on the human or the model?

16:24Joey: On the human.

16:25Addy: The model's like, yes, X plus Y is this. So then Z plus Y is that. and that human has to be like, well.

16:33Joey: The meaning of life is 42. Yeah, exactly. Don't worry about it.

16:35Addy: Don't worry about it. Just trust me. Yeah, so there's a lot of funny stuff happening in the math world right now for sure.

16:41Joey: Yeah, I mean, also I feel like as humans, we will always find new unanswered questions for the next thing to figure out. And so it's like, okay, we'll figure out these math problems, but I'm sure that they will be a new level of things that are just like, okay, we haven't figured out yet and the AI agents' models can't figure it out.

16:57Addy: This is a rumor. It's just fun. I think you'll get a laugh out of this. And so with our viewers, there's a couple of people inside Palantir who apparently started smoking now because they think we're just going to cure cancer real soon anyway. Like that's the level of trust in AI that they have.

17:20Joey: That's a bit of a AI derangement syndrome. But also, all right, I noticed the smokers, But the thing holding me back from smoking was not the risk of cancer. I was just like, I don't want to smoke.

17:35Addy: It just smells bad. Right. Yeah.

17:38Joey: Next up, actually, I had a lead into this next model update that also riffs on what we just talked about. Because going back to SaaS, but also the aggregator platform, the AI aggregator platforms that don't necessarily have the original models, but are kind of the big tool set. So like Magnific, Hicksfield, OpenArt.

17:59Addy: Runway. Well, they have their own model. Oh, sorry. Yeah, the ones that don't. Sorry.

18:03Joey: Yeah, everyone that has a model, like they are also trying to be the platform, but the ones that like don't have it, like they don't have like an AI lab original model, but what is their defense? And then in one level we saw Fall, which, you know, was an API aggregator, but then they post-trained Minimax, created their own version of Minimax that's better. And now it's like, okay, that's a Fall original. And also they have something that's original and defensible.

18:27Addy: Magnific came out with their own image model.

18:31Joey: Magnifique One. Ish. Magnifique One. I say ish because actually I did some research and we can talk about this. But this definitely seems focused on, I'm guessing, what their core user is of like brands. You can do a brand guide. Yeah, like Instagram commercials. Images with, yeah, images with art direction and that just gives you better output in the images. Now, I assume this was like a post-train model on something else, but it seems like in this VentureBeat interview, it's -

19:03Addy: Brand new from scratch, like foundational model?

19:06Joey: No, other direction. It is just a harness around GPT image too. Ah, that makes sense.

19:12Addy: Okay. Okay.

19:13Joey: So that makes me, I'm like, no shade to that. Maybe if it works, it works. But, actually, those are hard to defend because it's like, if OpenAI just adds their own design tool set feature in the app that's already based off GPT Image 2.5.

19:30Addy: I could see from a business perspective why they would have their own model. Like, it would totally make sense from a money standpoint because when you don't have your own model, right, You're basically the middleman. So, you know, the token. Who's already talking about?

19:47Joey: Yeah, but it's not, I mean, but it's not a, like it's not what Fall did where they took Minimax, they took the open weights and then they post-trained and made their own better version of it. It's like, this is just kind of a wrapper and hardest. Here, it subsequently described the systems distinguishing technology as prompt-based and consisting of textual instructions shaped by the company's creative specialists.

20:08Addy: specialists so they're doing like prompt engineering on top of the image box it's prompt engineering

20:14Joey: under the hood which is you know something we've seen in the past it works for now until the models

20:20Addy: just get better and you don't need it anymore also gpt 2.5 is so much better than 2.0 like it's a

20:26Joey: no-brainer yeah and then it's like that then you're also stuck in this position because if you built all of these prompts and like really like worked around to figure out how to tweak the model for this one model and then the new one comes out and it probably doesn't need as much prompt support as you had before but also now you're delayed and getting out something new and updated using the latest model secret prompts under the hood i feel like is a tough model because the new especially the new lms they just keep getting so much better that they don't need all the skills and instructions

20:57Addy: and stuff that you had to use before anyway so like when you prompt into gpt2.5 i can bet money that there is some kind of LLM inference right as you prompt that turns it into an expanded prompt and then that goes into the model. And I know you can turn that stuff off in certain aggregates, but who wants to? It does a really good job.

21:18Joey: Yeah, you don't need so much of the skill stuff that you built before.

21:21Addy: Yeah, well, what I was going to say was a couple of speculative things.

21:25Joey: And man, remember last time Joaquin corrected us when we said some speculative things? Yeah, you thought they got funding from somewhere else and he was like, no, nothing very easy.

21:34Addy: Yeah, no, yeah. I'm already saying I could be wrong here. But like, I think training your own model is a huge expense that could be deducted for taxes.

21:45Joey: Come here for tax advice from Maddie. For the business? Yeah. This is not tax advice.

21:50Addy: Yeah, like the infrastructure cost and like just people and R &D and everything to build your own model, that's a nice little write-off for a company that's really profitable, you know, that needs to show that. The other thing is if GPT-2 weights are really cheap, like if you can just buy them from OpenAI pretty cheap, then why not just make a little bit extra money from that by just like putting a new branding on it, putting a new property writer on it. And of course, your power users like you are going to go to GPT-2.5. You're not going to touch that. But then somebody who's just getting started on AI will use it.

22:28Addy: And a lot of that token usage will be in-house. It'll be like Magnifique's token usage. They don't have to pass the profit on to OpenAI.

22:36Joey: So you're saying they should license the GP2 weights and host it on their own compute to save money?

22:44Addy: That's right. Yeah, I think there is a business model where if you have your own weights and you have figured out your own infrastructure, then you can charge the user, let's say, 10 % less for an equivalent model, but then keep a much bigger portion of the margin because it's your in-house cost. You've vertically integrated, right? Like it's your own model, your own infrastructure. So you get to keep a bigger chunk of that token usage.

23:13Joey: Do you know if that's even an option where you could buy or license GPT-2 weights or other closed sorts models?

23:19Addy: I don't know about GPT-2 specifically, but yeah, you can buy weights from other companies like, you know, I don't know, NEMA.

23:27Joey: Well, I know there's a lot of companies where you can license or they give you licenses if you do a certain amount like LTX or Flux, Black Forest Labs.

23:36Addy: Yeah, LTX is like kind of open source, but then you have to license it. I don't know what the deal is with LTX.

23:42Joey: It's a license. It was like free to use, but if you like, it's kind of like the Unreal license thing where it's like free to use. If your company makes more than 10 million, then there's like, you got to talk to them about licensing.

23:52Addy: Just imagine this. I'm just totally making this up. Okay, so Nano Banana 2 is a closed model, right the weights are proprietary to google but if a giant like uh amazon or netflix came to google and said we want we want a licensed nano banana for 50 million dollars would they say no i mean

24:13Joey: it's just like a matter of business business i mean google's google's a tricky one because google would be like we already have our own compute i could see if they went to you know some maybe not Anthropoc or OpenAI. Well, Anthropoc doesn't have this, but if they went to OpenAI, but someone else a little bit lower that needs compute, then maybe that kind of thing makes sense. But look, I mean, if you were like, it is possible to buy or to pay a bunch of money and get the weights and run it on your own stuff, I would not.

24:44Addy: I mean, I don't have definitive proof on this stuff, but to me, it's a product, it's a thing that you can download and then you can license. You can attach like a license thing on top of it. And why not make money, extra money on the side as much as you're doing on Gemini? Yeah. Anyway. All right.

25:02Joey: Sure. We'll take Addy's tinfoil hat theory. Addy's tinfoil business plan. By the way, do not come to denoise for business ideas.

25:13Addy: This is the word that you were going to go broke, my friends.

25:16Joey: Yeah, just get a bunch of money and license these weights. And then when they come out with the new version, they got to give it more money.

25:23Addy: Yes, exactly. Oh, isn't that how the AI hype cycle is working right now anyway? It's all funny money.

25:29Joey: All right. Other new models. Muse from Meta came out with a Muse Wonder, which is...

25:38Addy: Yeah, can you walk me through what Wonder is?

25:41Joey: No, I can't because all their names are Muse. Muse, Muse, Muse. As far as I can tell, Muse Wonder is like, and according to the graph they have, is like their version of Fable and Astra, their state-of-the-art model. That is apparently pretty good. If you ask me, how would I access Muse, wonder, I don't know. Like, I don't know what the equivalent of Claude desktop is. I'm sure there's a Muse app. I just have not gone down that rabbit hole yet.

26:10Addy: Did you see the little, like, puck thing that they debuted with the little cute, like, animal on it?

26:16Joey: Yeah, everyone, yeah, their little little movie.

26:18Addy: So like you can speak to it, I guess. What does it do?

26:22Joey: That's what everyone, I mean, I don't know. OpenAI's got one too. They got the dot thing. Everyone's got little tchotchke AI things. I don't know. Do we need tchotchke AI things?

26:30Addy: I don't think we need more listening devices in our lives.

26:33Joey: Yeah. Yeah. I will say the consumer stuff I have, I don't pay as much attention.

26:39Addy: Sure enough.

26:39Joey: I mean, I also don't, yeah. I don't know if it's really going to take off. Says the guy with Ray-Ban glasses. those are for filming for documentary purposes gotcha man

26:50Addy: I gotcha good you're gonna crap on my business advice we're gonna go there

26:57Joey: well okay also probably part of the reason we didn't see so much about Muse is this post from Alexander Wang their head of AI who is also an amazing shit poster on Twitter they were basically like yeah I didn't see a bunch of stuff from us because we both were coding and we didn't train on 3D stuff. We trained on coding and it's a really good coding agent.

27:18Addy: Whole team feels dumb now, but it's true. We did not explicitly train on 3D demo games. Obviously, in retrospect, we probably should have done a bit of that, but no regrets. Please try Muse Wonder on your actual coding work.

27:29Joey: Oh, so it's really built to be a software writer.

27:32Addy: Yeah, for now.

27:33Joey: Fair enough. Yeah, here's some of his excellent posts of the Muse Labubu thing. I know it's not Labubu. I'm sorry I'm infringing on Labubu's trademark, but it's not.

27:45Addy: it's the it's a nice looking labubu i thought that muse was really targeted at marketers is

27:51Joey: that not the case i am sure under the hood for facebook ad manager it will be all muse all the time i mean where is this all going i'm sure it is to turbocharge making ads between intelligently running them writing the copy making the images making the video once they build all that i'm

28:09Addy: just going to run reverse engineer anything on it and then i'll have it just kidding guys

28:15Joey: just kidding see but they got a moat because they got they have network effect so that's one of the other moats where it's like they got the platform you can't copy the platform literally everybody

28:25Addy: on the planet is on their platform which is crazy yeah yeah all right other new one we talked about

28:31Joey: this but now it's actually out i'm an i4 oh yeah argon and they now they did trade on 3d dude this

28:38Addy: is so good so i ran across this video yesterday um joe you probably already seen it by the time i sent it to you i saw it pop up i didn't watch the whole thing so yeah what's what's what impressed you the gemini team every year the whenever they launch a new gemini they have this like japanese pagoda you know peaceful zen garden thing that they build and then they see the quality over time improve and that's like a visual indication so this is their culvert theater this is their cover tier. Yes. So it's building it from a single prompt. But what's really neat here is that single prompt then goes into an agent that is a blueprint agent, like an architectural blueprint agent, and it'll build a blueprint of this pagoda.

29:23Addy: And then it'll go into a nano banana agent, which is generating images off of that in many different angles for reference. And then it goes into an adversarial model, or rather an agent that is quality controlling the entire build process as multiple agents are building it in Blender.

29:41Joey: And it's doing this on its own. You didn't have to give it the instructions to like, hey, do a multi-agent, review your work, and blah, blah, blah.

29:48Addy: I don't know if you did. So that's the thing. He said it's off of a single prompt, but how he set up the system may have been manual. But regardless, the fact that you can get to this level of quality and fidelity from a single prompt is just nuts. It's bonkers. This is like hundreds, if not thousands of hours of man hours of somebody doing this hand by hand.

30:14Joey: Yeah. I mean, that's why it has been impressive because the amount of detail work in the single prompt and not having to tell it and refine it. And the interior.

30:21Addy: It looks like it built the interior as well. It's so crazy. And then, yeah, the shaders came in, and they rendered the whole thing in Blender. Blender seems to be like the proving ground, the LM arena of things nowadays. So, yeah.

30:34Joey: Well, it's such a good thing to – because it's just also so agent-friendly and easy to control because it's so scriptable. Right, right. So it's just an easy 3D thing to build out and demo. Yeah. Very cool. You want to get another Florida take? Yeah, please. This looks like the Japan land in Epcot. This exact angle right here.

30:57Addy: I've never been. I've never been. I'll take your word for it. I'm sure you've been many times. But that just kind of brought the whole thing down a notch. Because I've been to Japan, and I've seen these pagodas in person. And yeah, it's like, yeah, that's in Vegas. It's not a big deal. Like, the Eiffel Tower's in Vegas.

31:17Joey: I have been to Japan. Epcot. Check around the world. Can we go to the next Blender example? All right.

31:26Addy: So I guess we don't know what model they use, but this is also done through Blender connected to an LLM. And just incredible, the amount of like knickknacks and the real life randomness that this desk has. And it's all done through an agent.

31:46Joey: Yeah, the detail here. I saw another one. I think a lot of the ones I've been seeing lately, they've been saying Opus 5.5 has been doing it. I mean, also, Argon just came out, so I haven't quite seen a showdown yet between Opus 5.5 and Argon. Yeah, here, using Ashtra and Opus.

32:07Addy: Ashtra and Opus, okay, that makes sense.

32:09Joey: I'm curious how they handed it off between the two or if they found one is better at something like research or blocking versus textures or if it's just purely maxed out the usage of one time to switch to the other.

32:27Addy: Also, did you see that it was all from a single image? Like, is that the way then just to give it image references? Because that's how the Gemini guy did it too. Like, Nano Banana Agent generated images and then those images went into the Blender Builder.

32:42Joey: I have found, yeah, images definitely help. So if you give it more stuff, the better. But also if you give it just some direction, it will also just start building reference images and making it. Or if it's also based off something real or historical, it'll spin up like research agents and then find images online or maps or diagrams or something and then start building off of online research.

33:06Addy: Do you think this is why Tim Sweeney, CEO of Epic, is all of a sudden pro AI? Like he's seeing stuff like this. He's like, wait a minute.

33:13Joey: I mean, I think if you're any type of game or software developer, if you're not thinking of AI first, you're just going to be left in the dust. Yeah.

33:23Addy: So he's just basically... Where the wind is blowing? It's that.

33:27Joey: This is an Argon example. This cathedral built. From an image.

33:31Addy: From a single image, and then it built and lit. It's doing render passes, yeah. Oh my God. It has volumetric fog and everything.

33:41Joey: The other one I was looking for was this laundromat video. This is Opus 5.5.

33:47Addy: Which is one of my faves. I love Opus 5.5, but I have never tried to do this with Blender in it. You know what? I'm inspired today. I'm going to try something like this. Okay, cool. Not a laundromat. Maybe something more visually appealing.

34:06Joey: Were you about to say something?

34:07Addy: No, just Epic had the lead. you know, like five, six years ago, pre-generative AI boom. Like they were, and I think they still are, they have AI talent in-house because like a lot of the game engine stuff needs to be agentic. Like if you look at NPCs and things like that. And I don't know why, but they totally missed the boat on the first wave of AI. Like, you know, the way that sort of Blender has kind of caught that, Unreal didn't.

34:38Joey: Like in just making...

34:40Addy: Just being a platform for where AI progress happens and testing happens, Unreal could have been that platform.

34:48Joey: I think the ones that were successful in that was accidental just because they happened to build a platform that was very scripting friendly. So Blender is very scripting friendly, Resolve is scripting friendly, and that just works really well with an agent.

35:01Addy: Yeah. As far as I know, Unreal is also script friendly. It's all C++, Python supported. but maybe yeah for building

35:10Joey: or for blueprints or both

35:11Addy: oh you're right

35:12Joey: yeah for blueprints yeah like I'd expect blueprints yeah for sure you could that one should be scriptable but like the actual build model out of the world I don't

35:20Addy: I don't know yeah you're right I mean I guess recently they now have an MCV server but yeah it's just such a bummer to me and I know so many people in the the Unreal Engine world you know that could have really benefited from a lot of the AI progress It would have been just like rising tides, floats all boats kind of a thing, where modeling becomes easier, shading becomes easier. You can build games faster in Unreal, and that would in turn help the Fortnite ecosystem and all of the stuff that makes money for Epic just be more profitable. It would have been a win-win. But nevertheless, they're getting into it now.

36:00Addy: I sent you the tweet from Tim Sweeney the other day.

36:03Joey: We have a huge pivot underway here at Epic. moving developers around to retool for maximum AI acceleration.

36:09Addy: The aim isn't to do the same stuff with fear people. AI or SI. I mean, I think you got to get on board to accelerate your work.

36:18Joey: Not on board to use AI to replace everything, but use AI to accelerate the human directed work.

36:26Addy: Cool. Look, he replied to somebody who was dropping some hate. Your inability to read the room is really astounding, Tim. and then he responded, much of the room wants to be in 2016 again, but it's not. Our job as the provider of tools and technology to much of the industry is to design for the present and the future that will soon be. Personally, I'm optimistic about it and see it as empowering for creators. Dude, that's how we feel. This guy knows. Tim Sweeney knows. No, I met him once in person. He's awesome.

37:01Joey: All right.

37:02Addy: Good place to wrap it up. For those of you who've been listening on Apple Podcasts, give us a like and a five-star review would be really nice. You know, it helps with the algos.

37:11Joey: You know how it goes. It does. Yep. Links to everything we talked about at denoisepodcast.com Thanks for watching. We'll catch you in the next episode.

From the publisher

From reverse-engineering compiled software to building complex 3D environments from single prompts, AI tooling is moving past simple text generation. Joey and Addy break down the performance limits of "vibe-coded" apps, Gemini 4 Argon and Opus 5.5's Bl...

More from Denoised

All episodes