I am thrilled to share that I started a company, Clementine. Our goal is to make open source AI technology for multi-agent games.

The recent breakthroughs in generative AI completely change what is possible in gaming, just like 3D and broadband did in the past.

Some games will be completely reinvented. RPGs will have life-like NPCs and emergent story lines. Simulation games will offer deep relationships to the game characters. Strategy games will field opponents and allies that will mimic the strategic and diplomatic qualities of humans, while never stepping out of the game lore. These games can become infinitely deep.

An even bigger opportunity lies in the new genres that will arise. Game mechanics that currently live in our isolated imaginations will enter the market, and spur a new generation of games.

Right now there is a lot of enthusiasm on the changes that generative AI brings, but no clear path on how to land the technology, much less how to iterate quickly.

I am among the enthusiasts.

When Dalle 2 came out, I was drowned in ideas of how generative AI would change, well, everything. I started a short list of things that were now possible, some of which I prototyped.

But one idea was always stuck at the top: how could this be used for making more exciting games?

I am a sucker for stories. The games I love the most have either rich storytelling aspects (e.g. FF7) or are, as Tynan Sylvester says, story engines (like the Civilization series). When I play these games, I let my imagination run wild. I read every book in Morrowind, and imagine side plots that don’t exist. I think about the day-to-day life of my citizens in Civilization VI. What would they be reading in their newspapers?

The recent breakthroughs of AI give us new super powers for storytelling and for story engines. In fact, the first things I created with generative AI were illustrated short stories with Dalle and GPT-3, and when ChatGPT came out I created Morrowind fanfiction.

Since then, I have been puzzled, why have we not seen an explosion of games that really take advantage of generative AI? It is an obvious, if not the clearest, use case and yet we see relatively few games out there trying it out.

I left my job last September, and spent some time building Zaranova. As I wrote then, I wanted to get a better understanding of how to use generative AI during playtime, what were the challenges and perhaps get a sense of why there had been no explosion of AI games.

I found multiple factors that hamper the speed of adoption.

First, while putting simple proof of concepts is easy, tuning an experience that relies on AI is a hard, new discipline. Game development is already insanely multidisciplinary, adding yet another one may be too big a burden on already over encumbered teams.

Second, as this is a new discipline we don’t yet know what are the right abstractions. Game developers want full control over their creations, but we don’t know what the right knobs and levers should be.

Third, there are no obvious standards or entry points. Most attempts at bringing generative AI to a game is ad-hoc, even if they repeat the same structure as many do. Neither do AI engineers have an obvious way to their creations in a game, nor do game developers have an obvious way to pull a recent AI workflow to their game engine.

Finally, using these models during playtime is expensive, giving non-trivial costs to the marginal minute of gameplay. This means that, for now, only games within platforms with recurrent revenue streams can deploy this tech without serious worry about costs.

This situation places big studios as the most likely to deploy generative AI within playtime. But they are very slow: their development pipelines take years since they are aiming for fully polished, AAA experiences. And that is with understood mechanics and tools. Generative AI presents as much an opportunity for them as a PR and brand challenge. This will likely prolong their timelines.

But I want AI games now! And I want to see a ton of them. In fact, what I really want is to see a proliferation of AI games, the same way that Flash spurred a bonanza of browser based games made and enjoyed by the community.

I founded Clementine to make this a reality. To bridge the gap that indie and small developers face to be able to quickly ideate, create, and deploy AI games. I’ve been lucky enough to be part of a16z’s Speedrun and very grateful for their investment.

We will create open source tools that allow game developers to develop and deploy AI games with ease. Open source allows us to find the right tools and their abstractions as a community, and gives developers the ultimate control over their experience.

I am very excited to be working on this, and as you imagine I love to talk about it. If you share this enthusiasm, whether you are a gamer, a game dev, or just an enthusiast, please reach out to me!

You can find me on Twitter, join our Discord, or simply shoot me an email at ramon@clementine.games.

Cheers! Ramon