Blue Shrimp Games

Read the story how we took our Indie Game from Cursor + Claude Code to Steam! 🔥 Yes, a 100% AI Coded Game on Steam 😲 ! No AI Slop, all art and music/SFX are by real artists 🫡 Article Below 👇

@CryptoGarga I would say @TheArtDude_ is your go to!

looking to hire a talented pixel artist for some character sprites, little side project 🙏

@SamSteffanina Thanks, if you want to read the story of the development of it, here it is :) web3dev1337.github.io/epic-survivors…

@BlueShrimpGames this looks super fun, nice trailer!

@MohamedWlied Appreciate that 🫡! We have a ton of changes and fixes ready to deploy very soon! We will keep going, thanks for the support 🙏

@BlueShrimpGames the game is actually not half bad i played it for a while and it's fun just needs a bit of tweaks nothing to chop your balls about and i don't get why that guy in the comments is pissed about just keep improving and keep the good work

Read the story how we took our Indie Game from Cursor + Claude Code to Steam! 🔥 Yes, a 100% AI Coded Game on Steam 😲 ! No AI Slop, all art and music/SFX are by real artists 🫡 Article Below 👇

@shawmakesmagic We launched our game on Steam :) 100% AI Code, started Oct 2024, no Engine. x.com/BlueShrimpGame…

There is a low key meta of devs just building games with AI Came across a guy building Sim City 2000 from original source code and assets on GitHub Everyone is doing this I’m doing this And if we just add a little crime we can sustain

Feel free to checkout the Gas Town GUI I made, still needs a bit of work, but can help you learn some of the key concepts github.com/web3dev1337/ga… Overall though, sub-agents have been pretty useful for a while now, for parallel tasks (research/documenting a code base etc, or applying changes that can happen at same time). Biggest benefit is the separate context window, and also speed. I do fear the sub agents might not read all the claude.md's up the chain (need to check).

Claude's task system using sub-agents is pretty interesting. I don't know how well it works yet but I like the idea. I suppose I'll have to get into GasTown pretty soon, unless a better options surfaces.

Yes, please let me know what you think after you read! Overall, I would start out with detailed prompts and plans, and go step by step. You can inject in any knowledge you know, OR do other research in another session (or chat GPT, deep research if needed), to help for context. Obviously just use git, and go on branches, so you can experiment, I think we must have hundreds of unmerged branches that were made.

Oh yes, raylib should be good! Anything that you can do 100% of the game with code (i.e no editor to get in the way...at least not forced like unity). You can read the full journey here: web3dev1337.github.io/epic-survivors… Let me know any questions, and yep you should get cursor or Claude code or codex, and try make something in Raylib for sure! Else yeah JavaScript ones are possible. We used Monogame for Epic Survivors, which is great if you want to get it on steam.

So you are doing game dev now? Are you using AI at all with it? What engine if so For online games, I recommend Puppeteer or playwrite, such as 3JS / Phaser / PxiJS. That is for the front-end testing, plus obviously unit tests. Key is to also get good logs, and if u can, a CPU bot that can play the game (even simply). Headless mode helps alot too. Overall, Claude Code is the best tool to get, or Codex from OpenAI. They can handle your tests, set them up, and run them! Main point is to build systems around it, command lines for start (i.e flags you can pass in that might run with a CPU bot, or load specific scenarios)...it's just like building blocks, the more tools you get, the better your agent can access them. Then once you built them, add them to a claude.md / agents.md, which your AI reads each time (or include a line like --Please read 'XYZ.md" if you need to do automated testing, and that will have all the info. Realistically though, you can get your AI to record screenshots and videos that you can verify yourself manually quickly, again it can save you time compared to having to keep playing it yourself! Add in various tools into the game like god mode, skipping levels, spawning enemies/pickups etc, so the AI has all the 'dev tools' it needs to test whatever...else ask it to make it's own tools based on the framework you setup :)!

Fair call, but this was the 'simplest' type of game genre I both knew/liked (I played all the different games in depth to understand it), and that would keep the scope manageable for a first game. Even John Romero recommends copying basic games to start when learning. Remember, this was made with NO engine, and by a first time devs, and coded 100% with AI. This was thought to be impossible in October 2024, and still by many today. As for the hitboxes etc, I already have fixed most of that locally and will deploy soon. Soul, well that is a tough one. Same with VFX, we are not artists...although we paid a great one for our base characters and some of the main enemies, rest are packs. Same with VFX, tough...but again we did pay a great guy for our level one music! Either way, I see it as a technical achievement...it proved that AI can produce working code, and performant code too. And it's far from slop, although I am sure far from optimal (although what game is???). We got a few god classes, that is for sure! The game is not the best, and it's in a burnt out genre, but that's also on us for being game dev noobs (they always say first game is a write off). In addition, ask how many Devs get stuck never shipping their first game for years...or even a decade? Shipping is the hardest part, and putting yourself out there for all to see. Too many are in discord discussing obscure programming techniques and code snippets than shipping their game. THE RESISTANCE is real, as Steven Pressfield calls it. We had the game 80% complete (in terms of what is there now, demo)...in Feb 2025! But we didn't get it live till Oct 2025 in Next Fest, after a 14 week sprint of 12-18 hour days. Plenty of fear leading up to shipping, and it sucks to get so much negative feedback. Are you working on anything currently you hope to ship public? If so, get the book "Do The Work", it might change your life. Are you a game dev btw?

@BlueShrimpGames @Jonathan_Blow The 'Survivor-like' market is already drowning in clones, and the last thing we need is more GenAI-driven clutter. This Claude Code project is a prime example: zero soul, broken hitboxes, and messy VFX. It’s not 'game dev'; it's just more noise in an already saturated genre.

How do I short Claude Code?

@aguri_basutaado Other games we worked on were created with automatic tests from the start, with github actions on PR checkin. Again, still requires manual testing, but can pickup on most bugs before we ever see it!

First of all, plenty of manual testing is still required :) Second, later on we created a traditional 'AI' (CPU) that can play the game...then stringed it together with a headless game window later, and that finally opened up autonomous testing with launch commands. This enables the AI Agent to actually create real test scenarios that it can run in the real game, and monitor it with logs, or if needed video/image evidence. It has various helpers to spawn into specific levels, use weapons, whatever. x.com/BlueShrimpGame…

@AMA59620833 @Jonathan_Blow Appreciate the thoughtful comment, are you working on anything you would like to share? Any way we could improve the game to make it up to your standard? Any games you enjoy similar?

@BlueShrimpGames @Jonathan_Blow this is so insanely shit

No problem, and I 100% agree. AI can do the code perfectly if you use the right techniques, but that means nothing if the game is not 'fun'...and balance and game feel are something the AI struggles ALOT with. U can ask it to 'juice it up', and it sometimes can do cool things, but really you need a human to play test, and mould it. That is the bottleneck for sure...art, music, SFX, balancing, play testing and ensuring it's actually fun! AI can just help you do it 'faster'...our game is fully setup to be data driven, we can modify/create any level, enemy and all that without code, just modify/create YAMLs, and drop in assets...but yep we still struggle with ensuring it's actually fun and progression is good...plus balancing for people who already played these type of games (vamp survivors etc) vs a new person can be tough...considered dynamic difficulty adjustment (another tough point some like, some don't). Seems like games are getting more and more dopamine heavy...if u look at success of roblox, and the incremental games (feed blackhole etc), they literally are like boom boom boom...no waiting around. It's a tough one, as they are actually easier and faster to make, vs trying to make something really epic, good looking, which might just flop anyway, and GG!

@BlueShrimpGames @Jonathan_Blow Only being douchy intentionally about the AI sensationalism. You seem like a cool dude otherwise from our posts and I do wish you luck. I am just not undervaluing game feel which at this point really requires the human element IMHO.

Respect! That would have been a tough one bro, but happy you didn't let it get you down!!! Yeah, u are right, looks really don't equal success, all about the fun and game feel for sure. Are you working on any new games now btw? Perf as in performance u mean right? Performance does seem like a lost art for sure, amazing work on that :)! I have also spent a ton of time optimizing performance in my game, but a couple bottlenecks left that might impact low end devices.

@BlueShrimpGames @Jonathan_Blow I hope I handle it pretty well, I worked on the biggest entertainment failure of all time: Concord! Here is a hint: that game bombed epically but looked 100x cooler than your trailer. Looks != success. I’m proud of my 10/10 digital foundry perf review given I was perf lead.

Appreciate it! Yes, first time game dev on this game, but made multiple games in the last 16 months, Epic Survivors was just the first. Already got my agents onto fixing those clear bugs around hitboxes, and added the debugging stuff for abilities so I can viz it. Should be fixed by end of weekend :)

@BlueShrimpGames @Jonathan_Blow I am not sure how that addresses anything I said. I’m guessing you are first time game dev ( which is okay! ). Hey good luck bud!

@colin_k @arielsan Very nice! Any special partitioning at work?

100 boat performance check 🚢✅ @arielsan

@TheScoopy1 Thanks :D!

@BlueShrimpGames Crazy fun gameplay 🔥

@HerculesLegend Thanks , appreciate you playing 💙

@BlueShrimpGames Insane stuff! Actually one of my favorite games to grind and all Cursor + Claude!

Far from 'shat out by an LLM', a huge amount of work went into this game, just because an AI did the code, didn't mean it was some vibe coded 1 shot. I actually gave @ThePrimeagen full access to the code of this game and 3 other projects, but he never got around to reviewing it...wonder why? Happy to offer @Jonathan_Blow access to some of my code bases so he can do some on camera reviews and tear the 'AI Slop' to shreds on twitch/Youtube. And yeah it may have some of these now, but it didn't on Demo or launch, it added them over months and years after launch. The final sprint before the demo launch was insane I suggest you actually read about how it was made: web3dev1337.github.io/epic-survivors…

@BlueShrimpGames It has all of those things except the abilities and multi directional sprites.. What is the value add for this game over VS except the fact that it's shat out by an LLM