Make an RPG with AI: systems are cheap now, scope still is not

They are systems-heavy. Stats, levelling, equipment, inventory, a turn order or a real-time equivalent, dialogue with branches, quests with states, saving that has to capture all of it. Each is a week of work for a beginner and they all have to agree with each other.

Then they are also content-heavy. Towns, dungeons, NPCs, items, and enough of each that the systems have something to operate on.

The first pile is what stops people who can code a bit. The second is what stops everyone else. AI collapses the first pile substantially and leaves the second exactly as large as it was.

The systems. You describe them in plain language — how damage is calculated, what levelling gives you, how many item slots there are, what a quest looks like when it is half done — and Director 1 writes the GDScript in your project.

Ask for saving early. An RPG save has to capture quest states, inventory, position and stats, and retrofitting that into a game that already exists is genuinely unpleasant. Having it from the start costs nothing.

Then play and tune. RPG balance is not designable on paper. You will find the fight that is impossible and the item that trivialises the whole middle of the game only by playing it.

The world. Terrain, water, weather, towns, interiors, dungeons, and where every NPC stands. This is the content pile and it is manual, because a place is something you make rather than describe.

The good news is that world building is the part where the tool is fastest. Terrain you sculpt, water you outline, weather you set, buildings you place. What would be a modelling career elsewhere is an afternoon here.

Assets come from generation, from the web, from your own files, or get built directly as geometry. You will be told which happened rather than left to guess.

The dialogue system — branching, conditions, remembering what you said — is code and Director 1 writes it.

The actual lines are yours, and it is worth being deliberate about that. Generated filler dialogue is instantly recognisable and it is the fastest way to make a world feel dead. Fewer NPCs with lines you wrote is better than a town of people saying nothing in particular.

The classic first RPG is a sixty hour epic and it is never finished. The reason is not ambition, it is that content does not compress: every town is a town you have to build.

A two hour RPG with one town, one dungeon and a real ending will teach you more and will exist. Build that. If it is good, the second one can be bigger, and you will know what bigger costs.

Practically: Stage Engine is a desktop app for Mac and Windows, running your game locally on bundled open-source Godot. Building needs an internet connection; playing does not. Free download and free plan, Indie $10 a month, Pro $50 for more usage.

Quests break at the edges. The player kills the character who gives you the quest. They find the item before being told to look for it. They do step three first. They leave halfway through and come back an hour later.

Every one of those is normal player behaviour and every one of them will happen. Describing what should happen in each case when you ask for the quest is far cheaper than discovering it later in a save file that no longer makes sense.

The general rule is that a quest should be a state the world is in, not a sequence the player is trusted to follow. Ask for it that way and most of the breakage never appears.