I Directed a Game Without Writing the Code: Making RUINS // dusk.exe with Claude

I have no formal background in game development. No Unity tutorials completed, no shader courses, no history shipping anything playable. What I had was a very specific image in my head—the low-poly, dithered dread of PS1-era horror games like King’s Field and early Silent Hill—and a conviction that I could get there anyway.

RUINS // dusk.exe is the result: a first-person, PSX-aesthetic browser horror game with procedural dungeons, custom shaders, synthesized audio, and a real itch.io release. I did not write a line of it myself. I directed all of it.

This is the story of what that actually looked like in practice—not “AI made me a game,” but what it means to sit in the lead seat when you are not the one implementing anything.

Vision first, technical vocabulary second

I could not have told you what “vertex snapping” meant before this project. What I could tell you, precisely, was what the game needed to feel like: the wobble of geometry that does not quite sit still, the crunch of a color palette that has been strangled down to almost nothing, the specific unease of a corridor rendered by hardware that is struggling.

That was the job I actually had—holding the aesthetic and emotional target with total clarity, and pushing back the moment an implementation drifted from it, even when I could not diagnose why it had drifted in technical terms.

That turned out to be enough. Claude supplied the vocabulary—vertex snapping, dithering, procedural generation, Web Audio synthesis—as solutions to problems I was describing in plain language. My job was recognizing, on sight, whether the solution actually matched the thing in my head.

My actual job: test, react, redirect

The entire project ran on a loop that had almost nothing to do with writing code:

1. Claude would implement a pass—a shader, a generation system, an NPC, or a sound.
2. I would load it and actually play it.
3. I would give short, specific feedback: “The wobble is too aggressive; it reads as broken instead of retro,” or “The corridors feel too uniform; I need dead ends that feel intentional.”
4. Claude would adjust, and we would go again.

I never had to say “use vertex snapping.” I said things like “the geometry needs to feel unstable, like it is fighting the hardware to exist.” Claude figured out that meant quantizing vertex positions and made it happen. That is the real shape of directing versus implementing: I owned the what and the why it is wrong; Claude owned the how.

The thing I underestimated going in was how much leverage there is in just being a good tester. I did not need to know shader math to know when a shader felt wrong. Taste and clear communication turned out to be the actual bottleneck-breaker, not technical skill.

Where I had to push hardest

The shader work was the part where my lack of technical background could have sunk the whole aesthetic, because “make it look PS1” is not a specification—it is a feeling, and feelings are easy to approximate badly. I rejected several passes that were technically “low-poly” but read as clean and modern rather than broken and old. Getting the vertex-snapping and dithering shaders to actually feel period-accurate took repeated rounds of me saying “closer, but it is still too smooth” until it clicked.

The procedural generation is where I had to think like a level designer for the first time in my life, despite never having built a level. I had to articulate what makes procedurally generated space feel intentional instead of random—pacing, the rhythm of tension and release, and when a dead end should feel like a mistake versus a deliberate, dread-inducing choice. I did not know these were “level design principles” going in. I just knew when a layout felt off, and I said so until it did not.

The decision to synthesize audio instead of sourcing sample libraries was mine. I wanted the whole thing to live in a single file with no external assets, and I pushed on that constraint even when it would have been easier to find horror ambience samples online. Claude built the Web Audio synthesis to make that constraint work, and the result ended up sounding more alien and more ours than stock horror drones would have.

Six NPCs carrying the entire narrative weight of the game was a deliberate call. I wanted story delivered by voices, not environmental text dumps, because that is what always stayed with me in the games that inspired this one. Writing that direction clearly enough for it to land in six distinct character voices was some of the most genuinely creative work I did on the project, even though I never touched a line of dialogue implementation myself.

Shipping, not just building

The part I am proudest of is not the shaders or the procedural generation—it is that this did not stop at “cool tech demo I can show three friends.” We took it all the way to a real itch.io release, store page assets and all.

That is the step most solo experiments, AI-assisted or not, quietly skip. I treated “is this actually released?” as part of the job description from day one, not an afterthought once the game was “done.” Having Claude as the implementer did not remove that responsibility from me. If anything, it clarified that shipping was the one part of this project that was unambiguously mine to drive.

What this actually proves

I do not think the lesson here is “AI can make games for anyone.” I went in with a specific, deeply felt reference point for what I wanted, and I was relentless about rejecting anything that did not match it—including things that were technically competent but aesthetically wrong. That taste and persistence were doing real work.

What changed is that I did not need ten years of shader programming or level-design experience to act on that taste. The gap between “I know exactly what this should feel like” and “this exists and you can play it” got a lot shorter, and I got to spend the project living entirely on the vision side of that gap.

By the end, we had already scoped a sequel—RUINS // descent.exe—with a style guide written specifically so we would not have to relearn the aesthetic from scratch. That project went on to add multi-dungeon architecture, a second full dungeon, individualized NPCs, and per-zone ambient audio, building directly on everything dusk.exe established. None of that happens if dusk.exe is just a demo I never push to a real release.

---

RUINS // dusk.exe is a first-person, PSX-aesthetic browser horror game with procedural world generation, custom vertex-snapping and dithering shaders, synthesized Web Audio ambience, six lore NPCs, and a full itch.io release. It was directed, tested, and shipped by someone who had never made a game before.

Play RUINS // dusk.exe on itch.io