Did AI Really "Make" a Steam Game? An Honest Look at Cook The Dungeon — and How to Follow the Same Path

The 30-second version

A Steam page appeared for a game called Cook The Dungeon, and social media promptly crowned it "the first fully AI-made game." In reality it's a first-person roguelike deckbuilder about cooking the monsters you kill, the author is solo developer Danny Limanseta, the game hasn't shipped (release is slated for Q2 2027), and AI served as an assistant for code and visual assets. Every final asset, by the author's own account, was hand-picked, refined, and placed by him.

What does that mean if you're a beginner dreaming of your own game? Not that an AI will do everything in one evening. And not that no-code games are a myth. The truth sits in between: AI dramatically lowers the barrier to game development, but final decisions, taste, balance, and release polish stay with the human. Below we'll look at what's actually known about Cook The Dungeon, why the "first AI game" headline doesn't survive scrutiny, and how to walk the author's path without the hype.

What Cook The Dungeon is

The Steam page: store.steampowered.com/app/4364310/Cook_The_Dungeon.

Per the Steam API, the game is titled Cook The Dungeon and its appid is 4364310. The developer and publisher listed is Danny Limanseta — a one-person project. The genre says Strategy, and the description paints a more specific picture: a first-person roguelike deckbuilder where you roam a dungeon, kill monsters, gather ingredients, and cook dishes at a campfire.

The core mechanic is Precise Cut: monsters have to be finished off at exactly 0 HP. Overshooting or undershooting affects ingredient quality — and therefore the dish. That's a step beyond "run and shoot": the game demands calculation, deck planning, and an understanding of which combos work. The roguelike side means every run plays differently, and the deckbuilder layer adds depth — you assemble a deck of skills or recipes and test different strategies.

One more detail the noisy headlines drop: the game hasn't launched. Its Steam status is coming soon, with release targeted for Q2 2027. No price yet, no reviews, no metrics. This isn't a finished product — it's a page for a future release. When people write "AI made a Steam game," they skip a simple fact: a store page is not a game in players' hands.

The author has a profile on X: x.com/DannyLimanseta. As of July 21, 2026, no detailed thread or devlog documenting the making of Cook The Dungeon could be found. So everything about tools and timelines below is presented cautiously — as the author's social-media claims, not established facts.

"The first fully AI-made game"? Let's be honest

The loudest claim around Cook The Dungeon is that it's "the first fully AI-made game on Steam." Let's see what it's standing on.

Steam's AI disclosure (a dedicated section on the game's page) says otherwise. It states the author used generative AI during development, including help with code and creation of visual assets. But the key caveat: "each final asset was personally selected, refined, and placed by me." That's not "an AI made the game" — it's "an AI helped, and a human controlled the output."

So why does "the first fully AI-made game" keep flying around social media anyway? It's a classic marketing play. There's already precedent: CODEX MORTIS by GROLAF / Crunchfest billed itself as the "world's first 100% AI-developed game" back in December 2025 (as a demo), then launched into Early Access on March 26, 2026. CODEX MORTIS's Steam disclosure says outright: "AI is used for entire code of the game, text, arts, sounds and music". The page: store.steampowered.com/app/4084120/CODEX_MORTIS.

In other words, even taking the claims at face value, Cook The Dungeon is definitely not "first" — CODEX MORTIS came earlier and explicitly declared 100% AI. And Cook The Dungeon is a case of partial AI assistance, with the human remaining the author of curation and polish.

There are even earlier examples: the text-adventure AI Dungeon, plenty of indie games built with AI help in Pygame, Unity, and Godot, experiments inside ChatGPT. The idea of an "AI game" isn't new. What's new about Cook The Dungeon is that it's a visible Steam case that's convenient to spread as "AI made a game."

Our framing is more honest and more useful: the right way to see Cook The Dungeon is "a solo developer making heavy use of AI." The models sped up the work substantially but didn't replace the author. Which is good news for beginners: you don't have to wait for AI to write a whole game — you need to learn to work with AI as a tool and then make the final calls yourself.

Which tools and models the author used

Here it's important to separate the confirmed from the claimed. Confirmed by the Steam page: generative AI was used for code and visual assets. Everything else comes from a Russian-language repost of @neuraldvig / @xor_journal messages dated July 21, 2026, which cites the author's own words. No primary-source thread or devlog could be found, so these details are labeled "per the author / reportedly."

According to those posts, the author claimed that:

• he had no prior experience in game development or programming;
• he assembled the game in roughly 3 weeks;
• the engine is Godot;
• his main editor is Cursor;
• the code was written by GPT-5.5, Opus 4.8, Grok 4.5;
• the art was generated with ChatGPT.

These details are interesting but not independently verified. "3 weeks" could mean a prototype, the Steam page setup, or one specific iteration — not a full release. "No programming experience" is routinely exaggerated in viral stories, because readers love the arc of "I knew nothing, then I built a game." We're not saying the author is lying — we're saying these are his words, and they deserve healthy caution.

If you're a beginner, here's what those names actually mean:

Godot is a free, open-source game engine. It's approachable, supports 2D and 3D, and uses GDScript (Python-like) along with C# and C++. Godot is a frequent pick for a first game thanks to its simplicity and active community. For more on choosing an engine, see our (Russian-language) article "A game engine for beginners".

Cursor is a code editor built on VS Code and designed around AI. It lets you write prompts right in the editor, generate and edit files, and ask questions about the code. Effectively it's an interface where the model becomes the developer's extra pair of hands. If you want to see how AI writes code in practice, check out "When AI writes the code".

Why might the author have picked this exact combo? Godot is free and easy on hardware, Cursor lowers the barrier to code, and several models (GPT, Opus, Grok) let you compare outputs and keep the best one. It's a solid stack — but far from the only viable one.

Can you really make a game without programming

The honest answer: yes, with caveats. AI covers a lot of the technical work, but not all of it.

What AI can genuinely do:

• Write code from a text description. Say "make the character move left and right and jump" — and you'll get a working script.
• Generate sprites, textures, concept art, icons. Describe the style, get options.
• Come up with ideas: mechanics, plots, dialogue, names, descriptions.
• Help with balance, tables, damage formulas, progression.
• Create sounds and music via audio generators.
• Explain how to configure the engine, export a build, publish a store page.

What stays with the human:

• Game design. You decide what's fun and what isn't. AI doesn't feel fun.
• Curation and polish. The model will produce 10 options — picking the right one, refining it, and fitting it into the game is on you.
• Product vision. What atmosphere the game has, what pacing, how hard it should be.
• Playtesting. AI won't notice that a level is too hard or the controls feel awkward. Humans check that.
• Release and marketing. The Steam page, screenshots, description, talking to your audience — all human work.

The big illusion worth killing: "the AI will make the game for me." No. It will make individual pieces, but stitching them into a coherent product, making it fun, and carrying it to publication is the author's job. It's like building a house: AI can cut boards and hand you tools, but the blueprint, quality control, and the decision of where the front door goes remain yours.

If you want to see what this looks like in practice for a very first app, read "Demo: launching a game in 5 minutes".

Step by step: how a beginner approaches their first AI-built game

Here's a grounded plan that doesn't promise a Steam hit in a week but actually leads to your first game.

Step 1. Pick a tiny game. Not an MMORPG, not an open world, not an RPG with 50 quests. Go for something like Snake, Tetris, a clicker, a runner with jumps, a simple top-down shooter. Constraints are your friend. Fewer mechanics means faster results and less confusion.

Step 2. Choose a simple engine or builder. For 2D, Godot is a great fit. If you want zero code, try GDevelop, Construct, or even Roblox Studio. We covered engine choice for beginners in "A game engine for beginners".

Step 3. Generate code in small chunks. Don't ask the AI to "write the whole game" up front. Break it down: character movement, collisions, score, game-over screen, main menu. One chunk per prompt. Confirm each piece works before adding the next.

Step 4. Make the art with AI. Be specific about style: "16×16 pixel art, green slime, white background", "low-poly tree for a 3D scene", "fantasy-style sword icon". Generate several options, pick the best, touch it up by hand or ask the AI to fix it.

Step 5. Assemble the game. Glue together the mechanics, UI, levels, sounds. This is where the grind starts: bugs, mismatches, clunky controls. AI will help hunt down errors, but the patience has to be yours.

Step 6. Test it on friends. Hand it to a few people and stay quiet. No hints. Watch where they get stuck, what confuses them, what bores them. That's the most valuable feedback you'll get.

Step 7. Publish. For a first game it's usually easiest to start with itch.io — no barriers, no fees. Once you feel confident, you can start preparing a Steam page.

If you want to see how we walked this path ourselves on a concrete example, read "How to build a Snake game with AI". It's the real process, from idea to a working game.

This whole path rests on one skill: giving an AI a well-defined task and accepting its work step by step. That's exactly what we teach at skillmake — from the first prompt to a finished product.

Which AI to use for what

We won't pad this into a "Top 15 AI tools for games" listicle. Here are the specific categories and tools that actually deliver.

Code. The best options are Claude Code, Cursor, GitHub Copilot, plus the AI assistants built into editors. For a beginner, Cursor is often the most comfortable: it understands project context, can generate and edit whole files, and explains errors. Claude Code shines for complex architecture and long sessions.

Art. Midjourney, DALL-E, Flux, Stable Diffusion, plus ChatGPT's built-in generation. The choice depends on style: Midjourney for beautiful concepts, Stable Diffusion when you need control and local runs, DALL-E/ChatGPT for quick ideas.

Sound and music. Suno, Udio, ElevenLabs for voice, specialized sound generators like ElevenLabs Sound Effects, or asset libraries with AI search.

Ideas, writing, dialogue. ChatGPT, Claude, Gemini. The trick isn't "invent a plot" — it's providing context: the tone, the characters, the constraints. The more specific the prompt, the better the output.

We have deep dives on individual tools in other articles — here we're keeping the focus on the big picture.

How much time and money it really takes

Social media loves numbers like "3 weeks." They grab attention, but they rarely reflect reality.

Why one person's "3 weeks" doesn't equal your "3 weeks":

• The author may have had adjacent experience (design, production, tooling) he doesn't advertise.
• "3 weeks" may describe a prototype, not a finished product.
• Behind those weeks may sit months of learning and trial and error that didn't make it into the viral story.
• Your project, your motivation level, and your circumstances are different.

What's free: Godot, Cursor (has a free tier), some models on free plans, itch.io for publishing.

What people typically pay for: subscriptions to Cursor Pro, Claude Pro, ChatGPT Plus, Midjourney, Suno, GitHub Copilot, plus Steam's developer registration fee.

But the main resource isn't money. The main resources are taste and persistence. Taste helps you recognize what looks good and what's fun to play. Persistence carries the project through when the AI ships yet another bug and your friends say the game is "fine, but kind of boring." Without those two, every AI in the world becomes an endless draft generator.

How to publish on Steam or itch.io

For a first game, itch.io is usually the better start. It's an indie-game platform with no publishing fee and no heavy moderation, where you can upload a build fast and get feedback. itch.io is great for experiments, prototypes, and first steps.

Steam is more serious. To publish a game you need to:

• pay the developer registration fee;
• get the store page through moderation;
• prepare quality screenshots, a trailer, and a description;
• follow the platform's rules.

Since 2024, Steam has required an AI content disclosure. That's exactly what Cook The Dungeon's author did: he stated honestly where AI was used and added that final control stayed with him. Far from blocking publication, this approach builds audience trust.

If you plan to sell an AI-assisted game on Steam — you can. The essentials: honestly disclose where AI was used and make sure you hold the rights to every asset, generated ones included.

FAQ

Is it true the game was made entirely by AI?

No. According to the official AI disclosure on Steam, AI helped with code and visual assets, but the author personally selected, refined, and placed every final asset. The phrase "first fully AI-made game" is marketing hype, not an established fact.

Can a beginner with no coding skills make their own game?

Yes, but not with a single button press. AI models write code from a description, generate art, and help with ideas. Game design, asset curation, balance, testing, and polishing for release still fall to the human. A first game is entirely doable with a basic grasp of logic and some patience.

How much does it cost to make a game with AI?

You can start for free: Godot is free, Cursor and ChatGPT have free tiers, and itch.io charges nothing for publishing. For more serious work you'll pay for subscriptions: Cursor Pro, Claude Pro, ChatGPT Plus, Midjourney, Suno, and others. Publishing on Steam requires a registration fee. But the main costs are time and attention, not money.

Which AI is best for making games?

There's no universal answer. For code: Cursor, Claude Code, GitHub Copilot. For art: Midjourney, DALL-E, Flux, Stable Diffusion. For music and sound: Suno, Udio, ElevenLabs. For writing and ideas: ChatGPT, Claude, Gemini. The best approach is to use several tools for different jobs.

Can I sell an AI-assisted game on Steam?

Yes. The key is to honestly disclose the use of AI content in Steam's dedicated section and to make sure you have the rights to every asset. Steam has required this disclosure since 2024.

How long will a first game take?

It depends on the project's complexity, your experience, and how much free time you have. A simple game like Snake or a clicker can take days or weeks. A more complex project — months. The "3 weeks" figure from viral stories is more of an outlier or a prototype than the norm for a beginner.

This article is part of skillmake.ru — a Russian-language project about building real products with AI agents. The article you're reading is a translated case study from our production experience.