Gengiskan.online: the strategy game where artificial intelligence negotiates, betrays and remembers

For decades, turn-based strategy games taught us to read patterns: if you amass troops on a border, the AI ​​reacts in a predictable way; If you offer him a treaty, he accepts or rejects based on a hidden table of values. The expert player does not beat the opponent, he beats the algorithm.

Gengiskan.online was born precisely to break that comfort. The premise is simple to state and complex to execute: each rival leader is a generative artificial intelligence agent with personality, history, ambitions and a memory of what you have done with him.

In Genghiskan, diplomacy is not a menu with three buttons. It's a conversation. You can write a proposal in your own words: offer a trade corridor, ask for free passage through a territory, warn a neighbor about a common enemy, or blatantly lie about the size of your army.

The rival agent interprets the message, crosses it with his character (ambitious, cautious, vengeful, opportunistic), with the real state of the board and with the history of your relationship. And he answers. Sometimes he accepts. Sometimes he counterproposals. Sometimes he congratulates you on the offer and attacks you two turns later.

That layer of negotiation turns each game into a different story. There is no optimal route that can be memorized, because the rival is not a table: he is an interlocutor.

The most disruptive element of the design is persistent memory. If you break a pact on turn 12, the affected leader remembers it on turn 40 and the next campaign game. Reputation becomes a resource as valuable as gold or troops.

This produces an interesting phenomenon: players begin to behave ethically out of strategic interest. Betraying is profitable in the short term and catastrophic in the medium term, exactly as in real geopolitics.

Beneath the conversational layer is a solid, readable strategy game: territories, production, supply lines, combat-altering terrain, and expansion versus consolidation decisions. AI does not replace the rules, it inhabits them.

The battle resolution system is deterministic and auditable: the player can always understand why he lost a position. Uncertainty is in other people's intentions, not in the dice.

Gengiskan.online runs directly in the browser as a progressive web application: no downloads, no app store, no gigabyte installations. It works on desktop and mobile, and saves the status of the campaign to return to it later.

This technical decision is not accidental. It is the same philosophy with which we build the rest of the generative games ecosystem: that the barrier to entry is zero and that depth appears while playing.

The video game industry has been promising credible NPCs for years. Current language models allow, for the first time, that promise to be fulfilled within a closed and controllable framework of rules.

Gengiskan is a practical demonstration of that idea: not a chatbot with a map, but a strategy game where negotiation has measurable mechanical consequences. And that is, I think, the path that games are going to follow in the next decade.