I have played games where the system was immaculate and the story was nothing. I have played games where the rules barely held together and I still think about the campaign years later. The difference was never the system.

This might sound obvious. People who play tabletop RPGs will usually tell you the story is the point. Then a lot of them will go build an AI RPG tool that treats story as the output of a random number generator, or hands narrative control to the model, or optimizes for mechanical completeness over what the session actually felt like.

I spent enough time building the wrong version of Wanderhold to have a clear opinion about what the right version looks like. The short version is this: narrative is the point, the player has to own it, and AI is only useful here if it is doing the one thing it is actually good at.

Story is why people play

Even in crunchy systems, the roll mattered because of what it meant. The critical hit was memorable because of who was in the room when it happened, what was at stake, and whether the character survived to carry the consequence forward.

The system created the conditions. The story is what you remember.

This is not an argument against mechanics. It is an argument for knowing what mechanics are for. They are the infrastructure that makes decisions feel meaningful. Without them, the story has no stakes. But the mechanics themselves are not the thing people are showing up for.

When I ask someone what their best campaign was, they describe a moment. A choice they made, a character they lost, a scene that went wrong in the right way. No one describes their best campaign as the one where the dice math was cleanest.

The player has to be in control of where it goes

Here is the instinct that produces bad AI RPG tools: let the model narrate freely. Give it a premise and let it drive. The AI is generative and fast, so more AI involvement looks like more product.

It is backwards. If the model decides where the story goes, the player is reading a story, not playing a game. The distinction sounds pedantic until you are in the middle of a session and realize you have not made a decision in twenty minutes.

The job of the AI in this context is not to author the story. It is to write the consequences of your decisions in prose that makes those consequences land. Those are different jobs. The first job puts the model in the player's seat. The second job leaves the player in control and asks the model to serve that control.

Wanderhold is built around the second model. You choose. The mechanics resolve. The AI writes what that resolution looks and feels like in the world. The prose is the AI's contribution. The direction is yours.

This sounds like a constraint on what AI can do. It is. The constraint is what makes it a game rather than a demo.

AI is genuinely good at prose

Here is the part that took me a while to say out loud, because it felt like a concession: AI is not good at improvised rules. Ask a model to enforce dice mechanics without structured scaffolding and it will accommodate you, which is the opposite of enforcement. Ask it to maintain consistency across twenty sessions without state management and it will invent things that contradict your campaign's history and dress the contradictions up in confident prose.

These are real limitations and they matter.

What AI is genuinely good at is prose. Texture. Atmosphere. The specific way a character speaks when they are cornered, the way a location feels when you have been there before and something has changed. Given something real to work with (a resolved mechanic, a structured world, a campaign history it can actually access), the prose output is remarkable.

Most AI RPG tools are using AI for the thing it is bad at and leaving the thing it is good at on the table. They hand the model control of the game rules and then wonder why the mechanics feel soft. The model is not a rules engine. It is a prose engine, and a good one.

The right architecture is to give the model a real input (a specific situation, a resolved mechanic, a character with history) and ask it to write that situation in vivid prose. That is the job it can do well. Everything before that moment in the pipeline is structure that the model operates inside, not improvisation it is asked to generate.

A path back to reading

This one is a side argument rather than a central claim, but I think it is real.

Narrative-forward games ask you to read. Not as an incidental outcome, but by design. The session produces prose. You follow it.

A lot of people who have drifted away from long-form fiction (because of time, because of the attention the format demands, because life has moved in other directions) find that they are still reading when the prose is delivering a story they are inside of. The stakes are different when it is your character and your decision that produced the scene.

I do not think this is the main reason to care about narrative in solo RPGs. But I think it is a real thing that happens, and I think it matters that it happens.

---

The case for narrative-forward design is not that mechanics do not matter. It is that mechanics serve the story, and if the design gets that backwards, the players feel it immediately even if they cannot name it.

Build the system so the mechanics work. Then make sure they are working in service of something worth remembering.

That is what Wanderhold is trying to do. Start a campaign at playwanderhold.com/create and see whether it lands.