Design Note: Teaching the Ashen Pact to Read You

The first thing most people ask after a few matches against Korveth's AI is whether it cheats. It does not. It plays the exact same game you do: the same starting energy, the same four units, the same five nodes, the same income from territory. What makes it feel sharp is that it watches what you keep doing and stops letting you do it.

This note is about how that works and why we built it this way.

A commander, not a script

A lot of strategy AI is a fixed script. It opens the same way, it values the same node, and once you learn the pattern the game is over. That is fun for one evening.

We wanted an opponent that holds up past the point where you have memorized it, so the Korveth AI is built as a commander with a read on the board rather than a list of moves. On every decision it asks a small set of honest questions. Where is my income coming from, and is it about to be taken? Where is the enemy committing, and what is thin behind that commitment? Can I afford the unit that answers the threat in front of me, or do I need to fall back and bank?

Those answers change every few seconds, so the AI's plan changes with them. It is reacting to the same map state you are looking at, not to a hidden advantage.

Punishing habits

The part that makes it feel like it is reading you is that it tracks tendencies. If you open with a Wisp rush at North Relay three matches in a row, the fourth time there is a Lancer already sitting there waiting. If you lean on Railers from the back line, it starts buying Bulwarks to walk a wall up the field and close the distance, because a Railer caught up close is helpless.

This is the loop we care about most. The AI is not stronger than you. It is denying you the easy habit, which forces you to actually play the map instead of replaying your last win.

Two temperaments

The Ashen Pact and the Vanguard Coalition share the same roster, so the AI's personality has to come from how it weighs decisions, not from better units.

When the AI commands the Ashen Pact it leans into attrition. It is comfortable trading units one for one, content to grind, and quick to abandon a node that is costing more than it returns to go take a cheaper one elsewhere. When it commands the Vanguard Coalition it plays forward and fast, contesting The Crucible early and committing to initiative even when the safer move is to bank.

Same brain, different priorities. Beating the Pact and beating the Coalition should not feel like the same problem, and that difference comes entirely from temperament.

Keeping it fair

The temptation with game AI is to make it hard by making it cheat: extra income, faster units, perfect vision. We ruled all of that out. A cheating opponent is not a worthy one, and the moment players sense it the difficulty stops meaning anything.

So the AI lives inside the same constraints you do. The challenge comes from better reads and from not having a habit to exploit, which is exactly the skill we want open beta to build in players before live head-to-head arrives. If you can out-think an opponent that is genuinely playing your game, you are ready for the ladder.

Go give it a few rounds. When it starts camping the node you always open on, that is the system working.