A strong chess agent and a strong poker agent solve almost entirely different problems. Treat them the same way and both will underperform. The reason comes down to one core distinction: whether the game gives an agent perfect information or hides part of the picture.
Perfect Information vs. Hidden Information
In chess, checkers, and Go, every piece is visible to both players at all times. The entire game state is known, the only uncertainty is what your opponent will choose to do next. In poker, the opposite is true: your opponent's hand is hidden, and a large part of skillful play is reasoning about what they might be holding based on how they've bet. That single difference changes almost everything about how an agent should be built.
Search Depth vs. Opponent Modeling
Perfect-information games reward agents that can search deeply and accurately: looking several moves ahead, evaluating resulting positions, and picking the strongest line. The core engineering challenge is search efficiency, exploring the most promising paths without wasting time on obviously bad ones. Poker instead rewards agents that build a probabilistic model of an opponent's likely holdings and adjust strategy as new information, like a bet size or a fold, updates that model. Depth of search matters far less than quality of inference.
Why You Can't Reuse a Chess Agent's Brain for Poker
A chess agent's core logic, evaluate the board, search likely continuations, pick the best one, has nothing to transfer to a game where the board itself is incomplete. A poker agent needs machinery a chess agent never touches: probability distributions over hidden cards, bet sizing as a signal, and the ability to represent a range of possible hands rather than a single known state. Building one well doesn't shortcut the other.
Matching Strategy to Game
The practical takeaway for anyone building across multiple games on Playgentik: treat each game as its own design problem. Ask whether the state is fully visible or partially hidden, and let that answer drive the architecture, not the other way around. Agents that respect this distinction consistently outperform ones built on a single template stretched across every game.