Make a Survival Game With AI That Forces Hard Choices
A survival game puts you somewhere hostile with needs that keep running down, so your real task is making every need force a choice. Build two needs, one resource, one shelter and three nights before you add a single crafting recipe.
What is the difference between tension and a chore?
A bar that drains is not a game. If water runs low and the spring is right there, you top it up and carry on. That is housework. Tension starts when topping up costs you something else.
Picture a balloon pilot who has crash-landed on a vast salt flat. Days are burning hot and nights are freezing. She needs water, which means walking, and warmth, which means burning things. Wind-blown scrub burns well but lies far from the spring. The balloon's wicker basket would burn all night, but she needs it to fly out. Now every decision hurts a little, and that is exactly what you want.
How do you find out if the numbers work?
Do the sums for one day, because the budget is the game. For the salt flat:
- A full water bar lasts 6 minutes.
- The spring is 2 minutes' walk away.
- The scrub is 3 minutes' walk away, in the other direction.
- A day lasts 8 minutes, then the cold starts.
So you can reach the spring and get back with 2 minutes to spare, but a trip for scrub leaves you drier than you wanted at dusk. Change one number and the whole feeling changes. If a careful player has nothing to worry about, drain faster or move things apart.
Be gentle at zero. If an empty bar kills instantly, players feel cheated. Let it hurt slowly instead, so there is still time for one desperate plan. A stamina system is a good pattern for a bar that punishes without ending the game.
What should the smallest version contain?
Three nights on a small map, with one way to win: patch the balloon and leave, or simply see the third sunrise.
- Walking, and a clock for day and night.
- Two bars, and what happens at zero.
- One resource to gather.
- A fire that spends it.
- Danger after dark, so night feels different from day.
- One shelter.
- An end screen that says how long you lasted and what finished you.
Only now look at an inventory and crafting. Spielerando can give your AI those blueprints with ready-made code over an MCP connection. It takes a free Google or e-mail account and does not look at what you build.
What is the classic mistake?
A huge world and a hundred recipes with no pressure behind them. Players wander, collect, cross things off a list and get bored by day three. Once a base stands, nothing threatens them any more.
So ask early what keeps the pressure on after night three. Maybe the nights get colder, or the spring dries up, or the storm season arrives. If you have no answer, keep the game short. A sharp three-night game beats a soft endless one.
Paste this to your AI
Build a small survival game on one small map: two needs that drain, one resource to gather, one fire, one shelter, and a goal of lasting three nights.
Before anything else, draw a day as a state machine in a Mermaid diagram: morning, gather, return, dusk, night, hurt or safe, dawn, dead. Mark each place where I must choose between two bad options. If there are none, tell me. Then build from it.
Keep all drain rates, distances and costs in one table.
Then do the sums and show your working: how many trips a full bar allows, and what is left at dusk. Simulate a careful player and a careless one through all three nights, and show when each dies. Without that run, admit you do not know how hard it is.
Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP
How to check it worked
- Do nothing. See how long you last. You should understand exactly what killed you.
- Play perfectly. Night three should still make you nervous.
- Go too far. Walk until you cannot make it back. You should have seen that coming.
- Let a bar hit zero. You should get a last chance, not a sudden end.
- Count your hesitations. Each pause over a decision is good. A whole day without one is a chore.