The Fermi Paradox of Generative Code: Why Aren't The Vibe Coded Games Here Yet?

Clearly, we can use generative AI to write computer code.

Given that, why aren't we drowning in high-quality indie games?

Now, one answer to that might be that we, in fact, are and I am simply missing them all because I am old and out of touch. This is always a potentially correct answer these days. Please, drop a comment and let me know if this is the case!

But, if my general sense of the vibe is correct and the last couple of years have been fairly quiet for indie game releases then something seems weird. Let me explain why I think that is and what I think that might mean.

First, let me explain why I want to focus on indie games, smaller projects made by smaller teams (sometimes even solo creators). Something more like Caves of Qud or Inscryption or Loop Hero than a big AAA or even AA title like a Doom or a Grand Theft Auto or a Civilization. This is both because those big games need big narrative and logical consistency over a long play time that, as many have pointed out, current models struggle with and because AAA game studios have had a variety of economic struggles of late that might impact publication schedules.

Small indie games though, seem like ideal candidates for AI assisted code production. While producing auto/visual assets or story scripts with AI might be widely condemned as slop, producing code this way has quickly become the norm, and outside of the very most stringent Butlerian Jihadis on Bluesky, few seem bothered by it.

Indie games tend to be passion projects, with a potentially huge number of would-be game developers with ideas they lack the code skills to implement efficiently. I can think of three of my own off the top of my head, and three friends with similar inclinations. So there ought to be large numbers of would-be developers who might give vibe-coding a game a try.

Indie games also tend not to need hugely high-production value audio visual assets. Most forgo things like voice acting, and many use deliberately retro aesthetics that make graphical and audio elements easier for a small team or single amateur creator to make.

Finally, the risks of vibe-coding a game seem very low! A game project is much lower stakes than enterprise software, so even if drastic mistakes are made while vibe coding, no one gets hurt or loses personal information or bank account access or anything.

So where are they?

Assuming, again, my vibe sense is correct and they aren't here yet, let me suggest three possible situations and what I think those might mean more broadly.

1.) Vibe coding is, and will remain, more of a skill than advertised

Maybe writing software requires a fair amount of skill, even with AI tools, and maybe that won't change as AI tools become more capable. I know that my friends in software talk about fairly involved methods for engaging with coding agents. Specification documents, testing procedures, etc. In my own experience, even writing a small static website theme using Claude Code required a little bit of expertise. At least the sort of experience and vocabulary required to describe the features I wanted clearly and understand when the tool made mistakes. "Align the text to the left margin of the image," simple stuff like that. That's not so hard to learn, but again, this was a very simple project. A game is much more complex.

Maybe this is just going to be the way it is. No matter how good the machines get at writing code, specifying the code we want, and debugging the code we get, will continue to require expertise. You can't just say "write me a killer FPS in a far future setting, Claude" you have to say "we're going to use the Nvidia engine for the main framework, here are the entities I need you to create in that frame and here's how they need to interact" or whatever.

If this is the case, then AI will be a very normal technology indeed. Basically just another step along the road we started on when the first assembly languages made it so programmers no longer had to hand code instructions in binary, and continued on when high-level languages allowed them to describe code in something like "natural language." The machine will be abstracted behind another layer of interpretation, allowing for more efficient code production, but expertise will still be required to create software. That expertise will change, and maybe even become more accessible in some ways, but it won't go away. Computer Science majors will just learn how to write a specification sheet for an agent, instead of a for loop or variable declaration.

In this case, we might get some new indie games, but not all that many more than we used to, since the time investment required to build them will still be substantial and keep many would be developers from building projects. Maybe we'll get a few more as agentic coding interfaces become more standardized and more widely understood, and maybe cheaper. But, good news, game dev will still be a fundamentally human task. Just a little different.

2.) Maybe the Machines Can Figure Out Programming But Not Game Design

Maybe programming either already is or will soon be something that can be completely automated. Almost anyone will be able to sit down and describe the software they want to an AI agent using "everyday" language (or something near enough to everyday language that it can be figured out through a brief tutorial) and it will be coded up for them.

But maybe Game Design isn't so easy to automate. Maybe the machine can generate code, but thinking up and refining a compelling game play loop requires a human touch. It can't be extrapolated as easily from the data.

This would mean we aren't seeing vibe coded games flood us yet because humans are still figuring out how to do design. Maybe there is someone out there, right now, finally sitting down to work on their long desired Naval Simulator and only discovering making it fun and playable is really hard. Maybe they will keep at it, and we'll get it some day, just not anytime soon.

In this case, great news, good games will remain human artifacts. But, bad news, we'll all be inundated in terrible slop games. The imprint of publishers and reviewers will be more important than ever, and they will have crushing amounts of work to do.

3.) Maybe Game Design Can Also Be Automated, It Just Hasn't Been Yet

Maybe the machines will learn to make a compelling and interesting game, they just haven't yet. After all, cultural production from games to novels follows genre rules. Why wouldn't a model eventually uncover these? Maybe a little more data, some RLHF, and they will get there. And when they do, the vibe coded games will arrive en masse.

This is the dismal one. Because then the entire game space (and presumably every other space) devolves into what the bros call "leverage." Creative capacity no longer matters. We'll all be spammed by an infinite fire hose of high-quality content. Some particularly savvy and ruthless operators will figure out which ones to buy low and sell high. Mostly they will just guess and some will get lucky. The world becomes even more a casino than it already is.

I'm not sure what happens after that. Maybe Benjamin will be belatedly proven right. Maybe a crisis of cultural over-production will collapse the entire superstructure and people will be at last compelled to face with sober senses the real conditions of their lives.

But honestly I doubt it.

I don't know why but I doubt the third one. I fall between one and two somewhere? Maybe just naive like that?

What do you think?

Oh hi. Have you been missing image of the day? I moved my photography to another webpage to try to promote it more broadly. Check it out: https://recycledlightphotography.com/