Notes From Building NextOath

It's been a little over a year since I started building NextOath.

For a long stretch I worked on a much more systems-driven strategy game — a set of systems meant to generate emergent play. The idea was that if I got enough of them interacting — procedural maps, procedural resources, faction overlays, all feeding into each other — interesting behavior would surface on its own.

The systems worked. Interesting things happened. But the UI became a major impediment — it was hard for players to act on the systems and see their impact relative to everyone else's. I spent a lot of time here, trying to make the difficult visible. After many iterations, I found solutions to the UI problems. What I didn't find was the fun.

That's a hard place to be. The systems were getting better, I was learning a lot, and some of what I built there turned out to be useful elsewhere later on. So the question was never whether I could keep improving it. It was whether I'd iterated enough to know it wasn't going to get there. I've been doing this a long time, and I've watched people quit too early and I've watched people refuse to quit when they should have. Telling which one you are, while you're in the middle of it, is the hard part.

Eventually I decided I wanted something more visceral. The map was producing outcomes, but I couldn't connect a player's decision to those outcomes in a way that felt satisfying. I wanted players to see what they did and feel it (Reversi like gameplay). I'm not sure I've nailed that yet, but it feels closer than it did.

Something else that keeps surprising me is how often players get stuck on things that feel obvious to me. Genre is the clearest example. After living inside this game for a year, picking a genre feels like the most natural thing in the world. For a lot of players, it isn't. So "Create Genre" is now "Create World," and I've reworked some of the language around it. Small change, but a good reminder that everyone else is seeing this for the first time while I've been staring at it for months.

I've also spent more time than I expected fighting with LLM model quirks and availability. Early on I assumed structured output from an LLM model was a solved problem. It is not. Different models behave very differently, smaller ones especially, and a prompt that works cleanly on one can fall apart on another. It's all solvable — I've just spent a lot more time adapting than I planned to.

The thing I'm still chewing on is where players stop, and why. Some have gotten several rounds in and then just vanished, and most of them haven't told me why. So I'm left reading the game data, picking up feedback where I can, and trying to fix things before I fully understand them. Still figuring that part out.

Some of the game is far better than it was twelve months ago. Some of it I tore out completely. Some of it I'm still not sure about. For now I'm just going to keep building it, keep listening, and figure out the rest as I go.

Thanks to everyone who's played and given feedback.

p.s. Writing blogs is challenging — I applaud those who make this look easy.

Follow our changelog for updates, and don't hesitate to reach out via our support page with ideas, feedback or questions.