Taste is the new programming language (w/ Astrocade CEO Amir Sadeghian) | AI Basics

The TWiST AI Basics series is made possible by Google Cloud! Build context-aware AI workflows using DeepMind models and orchestration tools with Google for Startups' new Startup technical guide on generative media. Get the guide here: https://goo.gle/technical_guide Astrocade is building super-intelligence for the interactive world. Build the future of fun: https://www.astrocade.com/careers Today's show: *Making a video game once required a studio, an engine, and a team of five or more people working for six months. But with Astrocade, users just describe the game they'd like to play and share, and an AI assembles it from minutes, ready to publish to a TikTok-style feed. Still, founder and CEO Amir Sadeghian tell us AI can't actually build a great game all on its own. The magic isn't just in the automation, but the user's creativity and good taste. Amir demos how the app turns simple sentences and descriptions into full video games, explains why frontier models still regress to the mean rather than pushing boundaries, and unpacks how the game's human-in-the-loop workflow can turn even generic ideas into something original and fun. Guest: Amir Sadeghian on X: https://x.com/_amirabs Astrocade: https://www.astrocade.com/ Relevant Links Flappy Bird: https://flappybird.io/ Plants vs. Zombies: https://www.ea.com/ea-studios/popcap/plants-vs-zombies MrBeast on YouTube: https://www.youtube.com/@MrBeast Google's Nano Banana Pro: https://nanabananapro.com/ Timestamps: 0:00 Welcome to AI Basics, presented by Google Cloud 0:44 Amir Sadeghian, CEO of Astrocade joins 1:08 What is Astrocade and how does it work? 2:58 From describing to publishing: the TikTok for games 3:37 Amir's origin story: four continents, one vision 5:11 Why making games used to require years of training 6:46 The watermelon juice theory of product friction 7:35 Taste as the new programming language 14:18 Personalizing games with your own images 15:23 From consumer to prosumer: can you make a living? 16:22 Astrocade's creator fund and new creator economy 17:07 Open roles at Astrocade 18:10 Why play is the most universal human language 19:43 Google Cloud, Gemini, and wrapping up Subscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.com Check out the TWIST500: https://www.twist500.com Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Follow Lon: X: https://x.com/lons Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis Thank you to our partners! Build context-aware AI workflows using DeepMind models and orchestration tools with Google for Startups' new Startup technical guide on generative media. Get the guide here: https://goo.gle/technical_guide Astrocade is building super-intelligence for the interactive world. Build the future of fun: https://www.astrocade.com/careers Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland Check out Jason's suite of newsletters: https://substack.com/@calacanis Follow TWiST: Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups Substack: https://twistartups.substack.com

• 0:00Speaker 00

  All right, everybody. Welcome back to AI Basics. AI Basics, this is where we go over the fundamentals, the basic things that founders need to know. And we do this for legal. We do it for finance, and now we're doing it for AI. You can find all these episodes this week in startups.com/basics. And AI Basics is brought to you in partnership with our friends at Google Cloud. I'm really excited about today's guest because we're going to talk a little bit about how AI is impacting video games with us today. Amir Saad Egeen, he is the founder and CEO of Astrocade. This is an app that allows you to easily create original video games and share them with your friends. Welcome to the program, Amir.

• 0:44Speaker 01

  Thanks, Jason. Thank you for having me, man. I'm very excited about this.

• 0:47Speaker 00

  Well, you are in a category that I am seeing all over social media, which is games that were state of the art, I don't know, maybe 10 or 20 years ago. People are now able to fork, to build using AI, and you have a platform to do that. So tell us a little bit about what your product and service does and why you built it.

• 1:13Speaker 01

  I can tell a little bit about my story, but Astrocade, basically, you said it really well, is a platform that anyone can create any interactive experiences. Games is a large category of interactive experiences, but they're a lot more like education content, companions that we saw, a lot of different interactive content. But the core of it is any kind of content that is interactive. So anyone can create those content. The way they can create it, and the reason I say anyone can create it is through natural language, because you can just simply describe. You can just describe any kind of content, any kind of game, and the AI, our AI system actually generates them. And the other exciting part of Astrocade is, once you create those kinds of content, you can basically just press a publish button and publish it to a feed of content.

• 1:56Speaker 01

  Very similar to TikTok and YouTube, there's a feed of content that all this content is generated with, from creators across the world. We have more than 100,000 creators from various countries, more than 100 countries, creating this kind of content and publishing it, which is super, super exciting.

• 2:13Speaker 00

  And yeah, it's astrocade.com, yeah?

• 2:18Speaker 01

  Yeah.

• 2:18Speaker 00

  And so when you look at the site, I mean, there's just all the greatest hits that you can imagine, from, you know, motorcycle games to person versus person fighter games, all of these different formats. But you do them in a format that looks like TikTok, where you can scroll through them. If you don't like this game, you just scroll down to the next one, and you can play a casual game, scroll down to the next one, play a military game, scroll down to the next one, play an archery game. I mean, it's just incredible, the creativity people have. And the idea of creating a video game, people would have to go to college, they would have to do internships.

• 3:01Speaker 00

  You would be five, six, seven years into training to actually make a game. And now you're telling me they can just do a prompt, make a game, and then publish it, and then play with their friends, yeah?

• 3:13Speaker 01

  Yes, exactly. It's hard. Making games is really, really hard. And I will tell you a little bit why this is so exciting, and it actually, I think I've been accidentally being trained to build Astrocade during my entire life. I can tell you a little bit, like, so the co-founder of me, it's one of my brothers, actually, my CTO and co-founder, and we've been raised in four different continents. We were born and raised in Spain for five years, five years in Zimbabwe, 10 years in Iran, and then we came 13 years in US. US has been the longest place we've lived in.

• 3:47Speaker 01

  And the interesting thing, I think, when you're raised this way, the interesting thing is you just see so many different cultures. Every country you go is completely different. The people are different, the culture is different.

• 3:59Speaker 00

  How the food, the traditions, the topography and the architecture, everything is unique.

• 4:05Speaker 01

  And it's so beautiful, because each of these places have been built by different people, with different cultures, with different mindsets. And I think one of the biggest things we had was from the beginning, we were all, by the way, gamers. We were like hardcore gamers, like I'm now more of a casual gamer. I'll call it even a player rather than gamer, because when you say game, like, it's a different visual. Yeah, and we were like, okay, like, why is this not happening with the kind of like the gaming and interactive content? And it boils down to what you just said, actually, because creating a kind of a game, it's so hard. You have to go get educated, and you need to have multiple set of skills or build a studio to do it.

• 4:48Speaker 01

  It needs coding, right? You need to know how to build assets, you need to build how to tune the game, how the game feels is very, very hard. And you need to also understand the psychology of player, right? To make something really fun, right? So basically, what we did is we removed all these barriers, right? We removed all these barriers, and thanks to generative AI, we started the company in 2022. And the reason, again, is another part of my life. I've been in AI 2013. I did my PhD at Stanford AI, when AI was actually not cool. Nobody was talking about AI at the time. And in 2022, we saw that the technology is actually getting to a place that you can actually generate this interactive world.

• 5:30Speaker 01

  It took us a while to basically improve the technology to be able to make fun content, but that's basically the reason. And we built this technology. It's a funny story. I think this is something I think you would enjoy. So the first office for us was in my in-laws, in my house, and we had a fridge in the middle. So, and one of my founding engineers, one of them said like, hey, let's just order watermelon juice. So we ordered a bunch of these, a small watermelon juice, we filled up the entire row of the fridge, and two days passed, the entire row got finished. It finished.

• 6:10Speaker 01

  So, and they wanted more, so we were like, okay, this is like, we need to be more economically viable. So we went and bought these beer, half a liter bottles of watermelon juice, and we put it in the fridge.

• 6:21Speaker 00

  It's tasty. It's a tasty addiction, yes.

• 6:24Speaker 01

  Great thing. Two weeks passed, nobody touched these bottles. From small bottles that like everyone finished, we made them just bigger and put them in the fridge. And the demand completely disappeared. So it's funny, my brother was like, okay, let's do experiment. So he then went and took these watermelon and put cups right next to them. Just this small change, now people started actually consuming watermelon. And we were watching this, and like, he was like, okay, this is exactly the story of Astrocade. The friction, we removed a big barrier from the users, and suddenly demand popped up. Like now everyone is making watermelon juice.

• 7:05Speaker 00

  Yes.

• 7:06Speaker 01

  Now everyone is creating games, sorry. Lon made a game.

• 7:10Speaker 00

  Well, in under 10 minutes, yeah. And I have one that's being made as we speak in there. I have a spaceship game where I go eat donuts. But what's really nice is today, a lot of the concepts are already baked into the language model. So if you say, I want to create a spaceship game, it has the history of video games, it has reviews of every game ever made, it has people writing blog posts or podcasts about video game culture, what they like about games. So how does, I just put it in one sentence, I want a spaceship that goes and collects donuts. How does it, how do you build the harness and the knowledge around it to then decide how complex the game is?

• 7:55Speaker 00

  The levels of the game, all that stuff, you know, and is that something where you want the users to front load that, or you want them to just make a simple game and then refine on the back end? Because refining AI-generated content has always been a bit difficult, but here you're doing coding, so maybe it's not as difficult. You tell me.

• 8:14Speaker 01

  That's a very good point. We actually tried to find a really good balance here. You could let AI to do everything, and you can put the user in the middle and like try to ask them to refine the experience. What we learned actually through the past two and a half years of like really working closely with our creators is AI is still not there yet. AI cannot make a full game. It can copy the games that they have been trained on, like Flappy Bird. You say, I want a Flappy Bird, it can generate, because there's been so many Flappy Bird quote that AI has been trained on and it can generate. But if you could, the moment you start making actually a new experience or some modification to an experience, actually AI struggles a lot.

• 8:55Speaker 00

  Yeah.

• 8:55Speaker 01

  That's where our creators come. Actually, most of our like cool games, they've been through many iterations and many, like, literally wishes. We call it wish prompt, we call the prompts wishes. They go through multiple iteration of wishes to become really, really good. I can show you. Do you want me to show you a demo of one of these things?

• 9:15Speaker 00

  Absolutely, absolutely. And it's really interesting, because you have thought about, I'm in my interface right now, hey, is this a simple game or an advanced game? What's the acceleration, the rotation speed, you know, just what the obvious things are I should change? But here you're making Arrowstorm, so let's see how you do this.

• 9:33Speaker 01

  This is actually, this is from one of our creators.

• 9:36Speaker 00

  Oh, one of your creators.

• 9:37Speaker 01

  Our good creators, yeah. I just copied for the sake of the demo. So this is like a game that you have to basically go kill enemies, you can upgrade your arrow and stuff. Okay, actually level up. Look at the.

• 9:48Speaker 00

  Oh, you leveled up. Don't want to, I don't want to mess up your streak here.

• 9:53Speaker 00

  These games are so addicting, too. So does it understand gamification natively? And can you tell it, hey, I want to do gamification here, I want to have these levels and, you know, equipment upgrades, et cetera? Or does it just guess that you want that?

• 10:09Speaker 01

  AI can give you suggestions, but I think the best way that works is like your own taste and choosing it. AI can guide you. You can say, hey, add gamification to my game, but like, most of the time it's not a good gamification. It gives you ideas, but I think you have to choose. So this creator actually chose this kind of gamification. And even these numbers needs to be tuned. So one of the hard parts, even with AI in the loop, is every number and every tuning matters. If you change this 25% to 50%, suddenly game becomes so boring, because you can just go and win it. So here it still needs to play it, like, you need to create it, playtest it, go back and iterate on it. So I'm going to choose this, like, and I'm going to continue.

• 10:52Speaker 01

  So this is basically the game. Look at the clouds on top, look at every, all of these.

• 10:56Speaker 00

  I mean, the fidelity is amazing. This is like if I was on Nintendo Switch or my iPhone playing a game. It's, looks like a modern, you know, casual game on an iPhone or Android phone, yeah.

• 11:06Speaker 01

  This would take six, two years ago, three years ago, this would take like five people, six months to bake this game with a lot of budget. So let me show you actually how this game was made.

• 11:16Speaker 00

  Okay, so Amir, I see on the left-hand side you have the chat window, just like you would have a conversation with any large language model. And then in the middle. We have the game and on the right, we have game settings. Yeah.

• 11:31Speaker 01

  Exactly, exactly. So, and I'm gonna explain. So, the main way is to natural language. So, you describe, hey, in the menu, push up button, other play button, whatever you can say, resize it, make the shop button blue. So, do you see actually Creator is like in tracking with AI? But a lot of things that I mentioned needs tuning and natural language sometimes is not the best way to do it, right? For example, if you wanna fine tune the flower occurrences, you don't wanna go like back and forth on like, say, point twenty-five, then go to tell you a point three. So, sometimes it's easier to just like tune it with like.

• 12:03Speaker 00

  Yeah, put in a number, give a, make it very explicit because large language models are non-deterministic. They're taking a guess. That's the power of them as to what you want because they can fill in the gaps for you. But sometimes, you know, in a game, you want it to be deterministic. You get three lives. This, you know, boss has a thousand hit points and your sword does ten hit points of damage each time, whatever it happens to be.

• 12:25Speaker 01

  Exactly. And you wanna tune it very quickly. So, that's basically what we generate. Also, not that AI that is generating their game, but it also generates the editor for you. So, it's actually like generating two things for you. And then you can basically go and tune it. And that's basically the, what we learned is the optimal way of creating interactive content through natural language and through editor. And this is, you don't, literally, the only barrier right now is taste. And I call, actually, I say taste is the new programming language.

• 12:57Speaker 00

  Yes.

• 12:57Speaker 01

  For interactive experience.

• 12:58Speaker 00

  Because large language models have been trained on everything, there is a reversion to the meme sometimes, people, to the mean, to the average sometimes. So, people look at web design and if you build something with a large language model, sometimes it goes to what is the most common design, which is not what you want. So, then you've gotta get a little bit, you know, more granular and you gotta give it a better definition of that. And that could happen here. Hey, it looks, it looks like every other game. Sometimes that's good because it's familiar. Other times, if you're trying to do something unique and you want plants versus zombies, you actually want a different theme. You want a different art style. So, you, you gotta get more specific in that case.

• 13:40Speaker 01

  And eventually, as you get more games and all those things, you can retrain. We have a plan to retrain these language models on our data and make them better and better and better over time. So, it's gonna be, it's gonna be a self-evolving kind of agent or model that just keeps getting better.

• 13:57Speaker 00

  Yeah, recursive learning based upon what these builders want. Because these builders are different. Like, I, the first thing I thought of was, I wanna make a game about our family vacation and then play it with my daughters. So, I was like, well, we have to be the characters and our Bulldogs have to be the characters. So, do I have the ability to upload assets like pictures of my Bulldogs and have a Bulldog video game based on my three Bulldogs?

• 14:18Speaker 01

  Yeah, you can do that. You can drive asset, you can upload any images. We have safety, so you can upload anything.

• 14:25Speaker 00

  Yes.

• 14:25Speaker 01

  Yeah, we have like, you can basically generate and personalize all these things to whatever scenario. And that's the beauty of the world, right? That's the beauty of the world. That's the beauty of games and interactive experiences. You can leave it, you can act it, and you can personalize it to your taste for the first time.

• 14:40Speaker 00

  This is like the, one of the great things of technology. I remember when I first got my printer in the eighties, the idea that you could print a birthday card for somebody, they were like, literally, one of the hottest couple of years, the top software selling program was a greeting card maker.

• 14:55Speaker 01

  Yeah.

• 14:55Speaker 00

  And you would take a eight and a half by eleven and you fold it twice. So, you could, you know, make it into like a birthday card. And it was incredible. Everybody was making personalized birthday cards with photos on them, et cetera. And that creativity is like sort of what you're bringing here. Will you be able to eventually publish not just to your website and, but maybe even publish to an app store? And then you, you go from being, hey, this is for consumers to maybe prosumers. Maybe I could make a living off of this video game and have a business model inside of it. Maybe people will pay ninety-nine cents for it.

• 15:29Speaker 01

  Funny you say that. So, so the two, two things I wanna, I'll add a little bit to the, the, the birthday card. It's funny you say that because we actually saw one of the application, one of the things that people make is interactive birthday cards. And they put actually their friend's image and send a happy birthday as the interactive experience to their friends, which is amazing, right? And that's something like I didn't know is gonna happen when we start building this technology. But yeah, the second part is like what you mentioned on earning money. Actually, one of the things that I'm just generally very, very, very excited about AI and recent advances of AI

• 16:05Speaker 01

  is, one, it's just, it's just making everything more efficient and like amazing. But I think it's the, the part that is the most exciting for me is how much new opportunities this new technology is actually creating. We, we have now a lot of people making money from Astrocade. We have a Creator Fund. We pay people based on number of plays that they get on Astrocade. And, and there are people that actually they're full-time building games and interactive experiences on Astrocade and making thousands of dollars a month on Astrocade. So, we are kind of like building a new creator economy. Similarly, like you know how YouTube made Mr. Beast?

• 16:43Speaker 00

  Absolutely, yeah.

• 16:44Speaker 01

  We are seeing the same exact kind of like market dynamics.

• 16:48Speaker 00

  Yeah, subscribe, donations. I mean, there's different modalities to do these kind of things. So, this is an incredible startup. Where can people find out more? And are you hiring for the startup? People always come on the program and they need a couple of positions filled. So, we always love to, there's somebody out there who would love to, I think, join your company and build this.

• 17:07Speaker 01

  We would love that. We have a lot of positions from product side, from recommendation system. It's very important. It's a very hard problem now that you have hundreds of thousands of like content, how to surface them. That's a big thing we need to, we are hiring for AI agents post-training. A lot of people now, AI startups want post-training people. That's something like we want for training models on interactive experiences. We want to have the best models in the world for generating interactive experiences. So, we are hiring for that. There are a ton of position. And I think the beautiful thing about Astrocade is that the mission is really fun. You know, Jason, one of the, there were actually two big reasons we started Astrocade. And it's very interesting. Number one, I really care about impact. And I think

• 17:51Speaker 01

  when we started this, we wanted to build consumer. You can see, not a lot of people are building consumer companies in today's world, like most of it is B two B. We wanted to build a product that touches a billion people. And when you wanna build a product that touches a billion people, you have to find something that people need. It's so innate to humans, right? Like, a lot of companies are based on communication. But I think the most innate thing to humans, and I have a two-year-old and I have a one-month-old, and I can see it like literally in them, is play. Play is so universal, is a universal human language. You learn via play, you kind of like communicate and socialize via play. So, play is a big thing.

• 18:32Speaker 01

  So, we are actually building a product that actually makes you create play and actual play.

• 18:37Speaker 00

  Yeah.

• 18:38Speaker 01

  Again, the second thing is about our product is this is very interesting and our mission is actually really fun. You're actually sharing fun with the world. It's mentally games that like, there's been a lot of research on how much helpful is for your cognitive.

• 18:52Speaker 00

  Oh, yeah, this is a, a big part of it. What do they call it? Plasticity, like neuroplasticity. I always like to take up a new sport every year or play a new game. And when you learn, even a variation of a poker game as an example, you learn how to play PLO. You have rules you learn from No Limit Hold'em, but then you've given four cards and the whole game changes. And you have to reconceptualize all the heuristics.

• 19:17Speaker 01

  Amazing.

• 19:17Speaker 00

  You know, chess doesn't change. It would kind of be interesting to make new versions of chess where the game piece has changed in some way or who knows. But it is for neuroplasticity. Neuroplasticity, a great thing to do. Great job on this. And thanks you so much for sharing it. Everybody go to astrocade.com to find out more. And thanks to our partners over at Google. They are doing a great job with Gemini. I have to say, I'm using Gemini all the time. You know, they have code review and all this great stuff. The API is fantastic. I have to say, there was like two or three things that I had on my vacation that worked really well.

• 19:56Speaker 00

  Google Local and Google Maps and Gemini have all been tied together now. So, I was in France. I was like, I'm here right now. I need the best croissant, the best pan au chocolat, the best hot chocolate, whatever I was doing with my daughters. And can you put it on a map for me? And then I was asking about very specific dishes. And it was like, oh, my God, I have this concierge who is working for me twenty-four hours a day and just put the best pan au chocolat within walking distance of me and the time it opened. Google's crushing it.

• 20:31Speaker 00

  If you, if you wanna find a great pan au chocolat when you're in France or wanna listen to more of these basics, obviously, go check out Gemini. Great suite of products. The, the images and the video is also great. That's what we default to when we're making fun memes is Nano Banana. It's incredible. But for the basic series, thisweekinstartups.com/basics. If you're a founder, get all these basics. And you gotta get Nano Banana Pro. Yeah, I pay the extra ten bucks or whatever it is a month. It's incredible product, Amir. All right, we'll see you all next time. Bye-bye.