Making An Actually Fun 3D Game with AI

So I made a pretty incredible video game within a few hours, and the experience made me realize that the video game industry is anything but cooked. Yep, we are absolutely cooked, chat. GG. Now I'm no stranger to vibe coding games on this channel. In the past, I've put out a few versions of vibe coded games, and none of them I've been more proud of than The Librarian, which took me the better part of a day to come up with and is Actually a playable game. You can actually go over to this GitHub page and click on this link and test it yourself and see what a vibe-coded game looked like a year ago. Now, it's no secret that coding models and platforms have actually come a long way, but what really grinds my gears is that every time I'm over on X and I happen to see someone post their one-level game with very superficial elements, they're always suggesting that Game development is cooked and they couldn't be more wrong. And to try to prove that point and also show that it's not as easy as people make out, let's make The Librarian 2, which my producer Dave wants to call Look Ma, I Can Read Good Now. The title's a work in progress, but let's go ahead and build it and discover together how realistic it is for one person to pull off a AAA title. Now there's typically a lot that goes into game development, especially when you're talking about the bigger game studios, before the code actually starts getting written. When they're planning out a AAA title, there's a ton of roles involved, typically, like the game director, the creative director, the design director. Like, these big game studios have all of these roles that are all trying to do different things and pull different pieces together, oftentimes before the game is ever even started at all. And before they start to code, the art team is doing things like Mood boards and concept art and figuring out the color palette. They're going to look at competing games. What are the major themes, the character arcs, the story arcs? And most of these vibe-coded game demos that you're seeing, they don't do any of that. In fact, on this demo, I'm actually skipping over most of that. Now, me and Dave have had multiple game planning sessions where we figured out game loop and character styles and things like that, but we're also kind of skipping over a lot of the story and the narrative, things like that that would— typically go into a normal AAA game. Most games also figure out early on what are sort of the core pillars of the game. For us, we're basically taking the original Librarian concept and we're trying to uplevel it. So if you've never played the original Librarian game, you're in a library, you've got these students that come in and they knock books off the shelf. Your job is to go around and to take the books away from students or pick 'em off the floor, put 'em back on the shelf, The more books you pick up and put back on the shelf, you actually earn XP up here. You can level up, making it easier for you to chase down students, carry more books, and clean up the library. There's a chaos meter at the top, and the more cluttered and crazy the library gets, the more the chaos meter goes up. Your game is over when the chaos meter goes all the way to 100%. That's pretty much the entire game loop of the Librarian 1. So we're gonna build off of that. For the Librarian 2, we're gonna try to uplevel the graphics by quite a bit. We're gonna try to make a 3D version of it. The levels are gonna be procedurally generated, so every time you play it, it's gonna look slightly different. You're going to be able to collect new and different power-ups from the first version, and ideally you could sometimes create what's called a broken run. Meaning that your character gets so powerful that the game just feels effortless. Anybody who's played the roguelite genre of games kind of knows the idea there. There will also be meta upgrades in the game, so when you lose a round, you can still upgrade your character, and those upgrades are permanent. So your next round, the game is slightly easier than the last round. And we're introducing new boss systems and different types of disruptions in the library—things like sick kids—and we're going to introduce like natural disasters. type stuff. Like an earthquake can rock a whole bunch of books off the shelf and create chaos as well. So there's going to be some new systems introduced as well, but the main thing we're going for is better graphics, 3D environment, and the ability to create a super, super fun broken run occasionally. Now to build this game, I'm actually going to use 2 different coding models. I'm actually going to build out the initial game using Claude Code and the Opus 5 model because that's the model that's been the most impressive recently at creating really Good-looking games. Most of the demos that you've seen of people making things that look like Call of Duty and stuff like that, they've been made with the Opus 5 model. Once it's built with Opus 5, I'm actually going to switch over to Codex and use GPT-5.6 to do all the fine-tuning and dialing in all of the features and functionality of the game. So to start, I'm creating a new chat here. I created a new folder called The Librarian 2 where we're gonna build this game out, and I'm linking it over to my GitHub repository here of where The Librarian Part 1 is located. So it has a reference of like the initial game concept. So this is the repo for The Librarian 1. Here's the link to the repo. We're now building The Librarian 2, which is a complete overhaul and reimagining of the game. Now I'm just going to go ahead and use Whisper Flow and rattle off all of the ideas we want to be in this game Initially, this new version of the librarian should be 3D.

Look at the librarian one to understand the initial game loop. The library that the characters are in should not look the same every single time. Later.

So there's my long ramble of everything I want into this game, and this is gonna be V1. It's not gonna be perfect, but this could potentially run for like an hour or 2. So I'm gonna go ahead and submit it, and I'll let you know what we get back when it's Done. All right, so it took about an hour and a half, but it finally finished working through my initial prompt here. It's got the original game loop, figured that out. So let's see, let me just skim this real quick. We can see what it actually did here, and this is something that it did not do when I made Librarian 1, is it actually opens up the game in its own browser here inside of Claude Code, tests the game to make sure everything's working, and then you can see it sort of fix what it finds. It actually has a screenshot here of the game to see what it looks like. I mean, just looking at this screenshot now, I'm like, dang, this might actually be kind of impressive. The camera is punching through the ceiling. Let's fix that. Depth of field fixes. And we can see it just like keeps on running the scripts here, testing itself and then fixing what it finds. It also sort of tried to balance itself. Boss pressure is dominating the chaos meter and bosses never leave, adding patience timers and rebalancing their pressure. Now death is by by attrition from kid bumps, not chaos. So the game was testing itself and then rebalancing if it was way too hard or complicated. All right, let me just scroll down here, see what our final output is. It generated 10,400 lines of code across 33 modules, 0 asset files. Every texture, material, character, sound effect, and piece of music is generated in code. Items on the floor drive chaos. You vacuum them up by proximity and file them automatically beside a matching color bay with a compass always pointing at the Filing builds combos and XP. Each level you draft one of 3 upgrades. Survive 15 minutes. So there's procedural libraries, there's different bosses, we've got disasters— earthquakes, tornadoes. There's 4 branches which upgrade through lifetime XP. Bugs I found and fixed along the way: a soft lock where you can carry books no shelf would accept, kids that never stopped ransacking. What I want your eyes on. So balance is still the soft spot. I tuned it against a scripted bot plus design math. The book array currently fires carried books at range rather than being a physical returning projectile. All right, let's go ahead and run this and then we'll open up our localhost and see what our V1 looks like. All right. The Librarian 2. Look, Ma, I could read good. Begin a shift, choose a branch. Okay, so these are our levels: the Grand Library, Blockbuster's not open yet, Groove Merchant Records, Mega Mart Superstore. We got some background music. We need, we need this, uh, menu screen to look cooler, so we'll, we'll work on that. Staff development. So this is our meta progression here, which needs like a return to menu button or something. Let's see settings. So we've got graphics low, medium, high, and ultra. Awesome. Master volume, music, sound effects. Erase all progress. All right. Let's begin a shift. Oh my gosh. Okay. So the controls are reversed. This is way different than the last version. I need to get these controls dialed in though. Oh, that kid just took a book. Give me that book, kid.

Ah, okay.

I can't do this with the controls this way, so we'll need to go fix that. But I need to get out of this game here because I can't do it with everything going backwards. It's too messed up on my brain. Also, I need to fix this staff development page. There's no like back button. I'm going to go give that feedback real quick. I'm just going to keep on doing this one in Claude Code for now. When we start doing a lot more of the fine-tuning, I'm going to switch models though. And you know what, while we're at it, there's something else I've been wanting to do. So when I made my first version of the game, this was the thumbnail we used. I did some split testing on thumbnails. This was the one that won out. However, this character that you see here wasn't actually in the game, and I feel bad that I used a thumbnail and showed this character without this character actually existing. So I want to make this character that looks like me an option to actually be the librarian in the game as well. So I'm going to take that image and I'm going to drag it in and add it and I'm going to say, cool. Now I'm going to go ahead and give it this prompt and get those fixes made, but also bake in this new character. All right. And so after 15 minutes, it made the fixes. So the W and S are now swapped. The meta shop is now working better, supposedly, and it added a new character called Wolf. If I click on staff development, okay, the scrolling actually works now. It wasn't working before, and our back button is now in existence. All right, let's begin a shift, and we have our 2 characters. We might need to do a little optimizing on how these guys look, but yeah, you can be Marion or Wolf. Let's go ahead and use Wolf, and there we go. Now the actual movement is working. I want to make it so the camera angles change, but I'll come around to that one later. I can chase these kids around. Get out of here, kid! Yeah, I mean, it's looking good. The library is a little different looking. We actually have a map up in the top right, so I can actually go and see where the books are being dropped. Although the little arrow isn't like pointing the direction my character is actually moving, so that's something we should probably fix. Looks like there's a book on the floor here. Um, I can't seem to pick up this book here, so there's a little bug. Why can't I find any books on the floor? Chaos is at 3% and I'm not seeing where the books are. All right, so on the map it's showing some purple up here. Okay, so there's something on the ground. There we go. All right, so we got our first level up. Fire Drill Training: disasters last 15% less time. Overdue Fines: combos build 20% faster. Reading Glasses: +10% XP. Let's go ahead and make it so we upgrade XP faster. Look at the light coming through the window. I mean, it's even put like the shadow here. That's interesting. All right, so we're, we're making some good progress here, but it definitely needs some refinement. So let me go ahead and do that. And before I get too far into these refinements, I do want to start working with GPT-5.6 instead, just because I think that's actually a little bit better model to do some of the fine-tuning I'm trying to do. But before I do that, I want to do 2 quick things. One, I want to push this whole thing to GitHub and have it all sort of backed up there. And then over time I can actually like pull previous versions. If something goes off the rails and we go too far in one direction, I can sort of go back to a previous version. You don't need GitHub to do that, but I like to use GitHub because it allows me to like pick up where I left off and be working on this from a different computer and things like that. I also wanted to create a markdown file of what the game currently is and where it stands. That way, when I jump over to GPT-5.5, GPT can just read that markdown file and sort of understand the current state of the game and help me pick up where I left off. So I'm going to go ahead and do this. Create a markdown file for me that explains this entire game. Put as much detail as possible into this markdown file, and I'll go ahead and let it go off and do those things. All right. So a couple of minutes later, it made us a design.md file and it created our GitHub repository. Let's take a peek at the Design MD that it made, what the game is, quick start. So lots and lots and lots of detail here about exactly what the game is. So now when I jump over to Codex, it will know how to handle it. I'm going to go ahead and jump out of Claude Code, and we're gonna work in ChatGPT from here on out. I'm gonna switch into Codex mode, which is the area where we're gonna be doing most of our coding, and we're gonna create a new project called The Librarian. 2. And for the folder, we're going to select the folder that we created called The Librarian 2 when we did it with Codex. And just to sort of like warm up the context here, I'm going to tell it to read the design.md file, get a good understanding of the game, and tell me what improvements it suggests. Let's let Codex and the 5.6 SOL model actually suggest some improvements. It might have some good ideas. So I'll go ahead and create that prompt, let it go ahead and start working on that. And as I was playing the game, I also made a list of little improvements that need to be fixed. Then we'll actually move into balancing the game and making it feel like it's the right amount of difficulty. The game should be really, really hard when you start, but as you get upgrades, the game gets easier and easier, but you shouldn't be able to beat it your first time through. That's kind of how this type of roguelike roguelite game works. So it went ahead and it came up with about 12 different recommendations. You can see my highest priority recommendations: fix procedural reachability before doing more balance work, repair progression stacking before judging difficulty, add a short interactive first shift tutorial, and a bunch of little suggestions. Now I also gave it a prompt with some more ideas that I wanted to fix, and now that these recommendations have all been dialed in, I'm gonna go and test the game again, but this time I'm gonna jump on with Producer Dave, who helped me come up with this whole game concept, and him and I have been brainstorming what this game should be together, and I'm gonna review it for the first time since these changes with him.

Oh my God, this is awesome. That's a hype-ass menu. You even put the— look, Ma, I could read good.

Oh, now it's playing music. All right, so begin a shift. You could be Marion or Wolfe.

Who's Marion?

All right, so there's a tutorial. So pick up the cobalt book. Oh, and there's an arrow that tells me where it is. Okay, so I picked it up and put it on the shelf. Move with purpose. Sprint while moving. So I can sprint a little bit, and I have a stamina bar up here. Ooh. All right, so somehow I've already leveled up.

All right, so we got a Chromatic Shush. That's outstanding.

A shockwave that sends every nearby book of one color flying back To the shelf at work.

And it's an active— you like press F to do it? That's sick.

After-school rush. Small group has arrived. You little shit.

Like, they're just like, hey, they're going in, dude. Press your F. Go close to them and press F. See what happens.

Earthquake. Earthquake incoming. Oh, geez. Braden the bully.

Oh, you can see him on your minimap.

All right, let's go get him. What the hell? He's so fast. Braden, get your ass over here. How is he so fast?

You can't catch him. It's impossible. Oh, there he is. Look at his arms flailing. He's so fast.

I'm outside of the library.

There he is. He's— yeah, just follow him on the minimap. All you could do is look at the minimap. There you go. Yeah, yeah, yeah. Kill him.

It's not killing him, it's catching him.

Oh, sorry. He's causing some chaos here. Look, you're at 36% and climbing. This just became hard.

It did, but he's way too freaking fast.

Look at him go.

Oh, he just did it. He just died. He disappeared into nothing.

Oh, no. Yeah, this is problematic.

A tornado. Where's the tornado?

It's inside the library. What happens if I walk into the tornado? Your health is dropping. Your health is actually dropping.

Oh, I have a health bar. This is not doing nearly enough damage that a tornado seems like it should have done.

No, it knocked like 4 books off the shelf. That doesn't— that's not tornado worthy. Dude, chaos is skyrocketing. 65%. You might actually lose this run.

I probably will.

Uh, 95%. You're toast.

I don't know what— I don't know how to like keep the chaos down at all.

Just spam F.

Oh no.

The building has been surrendered to the children.

This is, this is where we're at right now.

I'm impressed.

I mean, visually, it's a step up.

You know what I think the biggest takeaway is for me? Is that like to make this a really enjoyable game with lots of great features, great balancing, really clearly defined progression. And replayable game loop, it'll take a long time.

Yeah. Yeah.

And a lot of work.

Yeah. So the CEO of Take-Two once said, like, you know, when it comes to gaming, coding was never the bottleneck for us. It's all of the little decisions and the balancing and the little like elements that you decide to add and the elements you decide not to add. And the, you know, all of that stuff is what makes a good game, not Being able to code.

So I guess then it's fairly safe to say that game developers aren't cooked.

Yeah, no, I definitely—

We still, we still need them. Like, I don't feel like there's any part of me that wants to like stop buying games. No, just vibe code them instead. I have zero desire.

It's funny that you say that. This is a guy that I follow on Twitter and he's, he's taken some flak for his recent take. He says, I'm pretty sure gaming as a category is dead because everyone is just going to build their own games.

Nope.

Games don't have ongoing maintenance costs like companies replacing SaaS vendors. Perhaps an argument for wanting someone else to build an experience for you. Easier to watch a movie than create one. But video games and vibe coding are so close. The difference between creating the form and playing in a form is getting closer and closer to zero. So why play someone else's game? Just pay ChatGPT $20 and make your own. And he showed this screenshot of like a vibe-coded Smash Brothers, right? And literally nobody in the comments agrees with him.

Easier to watch a movie than to create one. Yeah, that's the point. It also takes the original idea to like be something inspiring and interesting that people want to watch and want to experience for the first time. If you could just come up with the idea yourself, then nothing is exciting, nothing is interesting. Looking at this Super Smash Brothers, I'm sorry, but that doesn't look good.

No, it doesn't.

I don't want to play this. I'd rather play Super Smash Brothers. All due respect to whoever this guy is, but take this off the screen right now.

It's just making me mad. All right. Now that Dave and I have had a chance to sort of play around with it and find out what's working well and what's not, I'm going to give it one more round of fixes and that should do it. So a few of the fixes I'm making: change the camera angle. Fix some of the lighting in the windows. The difficulty scaling needs to start off a little bit harder, but not get so hard so early in the game. So sort of trying to balance that a little bit. I want the natural disasters to do a little bit more damage than they actually do. I want to slow down the bully a little bit and just a few other balancing things. I also want to see if I can optimize the Wolf character a little bit more to look a little bit closer to the thumbnail. We'll see what they give us here. And I'm also setting it to Soul Ultra. The best model right now from OpenAI to see if it can really dial all these things in.

A few moments later.

All right. So it took about 30 minutes, not too bad. And it fixed all of my stuff here. Let's go ahead and give it a test. All right. So it didn't really change my character too much, even though I asked it to try to make it look closer to the character of the image. It changed the colors. It got a little further away. Whatever.

It's fine.

Okay, so I can actually change the camera and rotate. Oh, that is actually smooth. I think it's inverted though. I think I want it to be the other way around. All right, let's go over and see how the lights look. Look at that. That looks a lot better than it did. Can I still go outside the walls with the camera? No, the camera won't let me go outside the walls anymore. Perfect. Perfect. Oh, changed my mind. Yes, it can. Oh no, there's a Karen. Not a Karen. What is she complaining about? I'll be leaving a review. This is unacceptable. Okay, Karen. I got the book-a-rag now so I can go boom and just throw them back to the shelf. That's cool. Oh, but it's got a cooldown, so I can't just like immediately do it. Oh no, a volcano! There's a volcano grew out of the center of the library. Yay! Okay, so I beat this level. New branch unlocked, Blockbuster Night. All right, so let's test some of these other levels. We've tried the Grand Library. Let's see how Blockbuster looks different. Let's do Marion this time. Okay, so Blockbuster doesn't look a whole lot different other than a different color. Okay, let's abandon the shift. Let's try the Record Store. Okay, so the Record Store looks identical to the Library. Let's see if the Grocery Store looks any different. Okay, so not Not really. There were some extra things we added to the grocery store, like if a banana falls off the shelf you can slip on it, and if like a mushroom falls off the shelf it makes you grow like Mario. Definitely need to optimize so each of these levels look different other than just the color though. Safety bonus: shaken soda can contained— ooh, that was a mushroom! Okay, so I just picked up a mushroom. You couldn't tell that it was a mushroom, but it just made me a lot bigger. I will make a link where it's playable using GPT Sites. So anybody who wants to play it can play it. I'll also put it up on GitHub so you can download it and fork it and make your own version of it. I'm going to keep on optimizing this and fixing a few of the things that I want to fix, but I do have some final thoughts. So let's, let's get into those. All right. So my final thoughts on this whole game development process. First of all, it is really, really fun. to get AI to help you make a game. But do I think game developers are actually cooked? Not even close. And AAA, like, even further away from being cooked. AI can make a fairly decent, pretty fun, and actually surprisingly polished-looking game, and it can do it pretty quickly. But I don't actually think that's what makes a good game. I actually agree with the take that development was probably never really the bottleneck to creating good games. There's probably really, really good developers out there that have just made absolutely crappy games, while there's also probably people out there that don't know how to develop at all that have come up with really, really good games and then hired developers to help them make them. I think the real hard part about making a good game is all of the other little elements, making judgment calls about what should be in the game and what shouldn't be in the game. All of these little decisions need to come together to make a really good game, and all of those little decisions are the thing that make making a really good game difficult, not the physical typing of the code. So again, game development is nowhere near cooked. Now, will game development be sped up a little bit? Probably. I think game developers will be a lot faster at getting games out because of AI, but I don't think AI is going to do a very good job at finding all those little nuanced items and figuring out how to make the game fun and balanced. Like, yeah, I made the librarian too, you know, decent. It's actually a pretty fun game, but is this gonna turn into a game studio? No. Am I gonna put this on Steam? Absolutely not. I mean, I might throw it up on GitHub and let you play with it and download it and fork it and do whatever you want with it, but I have zero intention to sell this. This is nowhere near sellable. So again, I think AI is gonna make game development faster, but this does not mean people are gonna stop wanting professionally developed games or games created by like really good game designers. Because Because the hard creative parts and all the little nuances haven't disappeared as a result of AI. I also don't buy the narrative that people are going to want to make games instead of buying games. I think most people would rather let other people make the decisions of what goes into a good game and then buy other people's games and play them. So I think AI is a fun toy if you're like me and not a hardcore developer. And want to try playing around with making some good games. But I mean, even most of the games I've made for myself, I've never gone back and played very much other than to make demo videos of them. But I'm still a big gamer. I still go and buy and play a lot of games that other people made, and I don't think that's gonna change. I think some people will have fun making their own little games and play with them and share them with their friends, but they're still going to the big studios. And I still think game development has a lot of life left in it, and I'm still excited to go buy and play other people's games, even though yes, I can make games myself if I want to. I just kind of choose not to most of the time, and I think most gamers will agree with me. Anyway, that's what I got for you today. I know this was a little bit different. I love testing tools in this way and trying to challenge myself to build other things. We're going to be doing this more from time to time on the channel. The Librarian one came out almost a year ago exactly. It came out in August of 2025. It's now August 2026 when I'm shooting this one. So I thought it would be fun to see how far these tools have come in one year when it comes to game development. Back then, it was a lot of back and forth, a lot of me creating the assets and sound effects and music and using all sorts of tools to tie them together. Now it's like one or two tools, give it a few prompts, and I have what I'm looking for. Like, it's kind of crazy how far along this stuff has come. But again, I'm still looking forward to the next games out of the best game developers. If you like videos like this, And you want to stay looped in on the latest of what's going on in the world of AI, whether you love AI, you hate AI, you fall somewhere in the middle. I'm just trying to do my best to keep you informed with what's out there. And if you like that kind of stuff, consider liking this video and subscribing to this channel. It'll make sure more stuff like that shows up in your YouTube feed. Thank you so much for hanging out and nerding out with me. I really, really appreciate you. And, uh, hopefully I'll see you in the next one. Bye-bye.