AI and the Future of Poker

Over the last decade, the world of poker has witnessed a systemic change, not just in the way the game is played but in how it’s studied. The shift from intuition-based strategy to data-backed strategy has redefined how poker is approached at every level. But the most exciting changes lie ahead, driven by advancements in artificial intelligence (AI) game theory.

At the core of this revolution is GTO Wizard’s AI engine team, responsible for the research and development of GTO Wizard AI and driven by a bold vision: to solve any poker variant in a few seconds. In this article, we’ll explore the evolution of poker study, where it stands today, and the future thats on the horizon as envisioned by Philippe Beardsell, the lead of GTO Wizard’s AI engine team.

Philippe Beardsell giving a lecture on poker AI at MIT during the 2025 MIT Pokerbots competition

The Traditional Poker Study Era: From Books to Solvers

Before the advent of solvers, players relied heavily on books, strategy guides, and forum discussions to improve their skills. It was an era where intuition and experience were the primary teachers.

With solvers came the ability to explore game theory optimal (GTO) strategies. For the first time, poker could be broken down into mathematically precise strategies and concepts. While solvers unlocked unprecedented insights, they had a steep learning curve. Players needed to invest a considerable amount of time in understanding how to set up simulations, let alone interpret the results. Moreover, the computations of these simulations could take up to hours to converge for complex spots, limiting their effectiveness.

As solvers began to gain widespread adoption, platforms like GTO Wizard emerged, democratizing solvers into user-friendly interfaces and significantly reducing the amount of manual work and costs through pre-solved solutions.

The Future: AI-Powered Study Tools

With the release of GTO Wizard AI, we removed the limitations of pre-solved solutions that constrained players to fixed parameters, allowing you to edit the solving parameters and solve poker spots in mere seconds. The later breakthroughs of custom rake and ICM+bounty solving brought a new and more effective way for players to study cash games and tournaments by reducing the reliance on multiple pieces of software, extensive computing power, and time to solve spots.

We have always believed that speed is as important as accuracy for effective learning. Indeed, for effective learning to happen, players need fast feedback loops to allow them to constantly test and refine their hypotheses. However, AI inherently involves a trade-off between speed and accuracy. Much like virtual assistants such as Siri, which prioritize real-time interaction over perfect accuracy, we see GTO Wizard AI playing a similar role for poker players, offering flexibility in how they approach the game.

We understand that learning styles vary and evolve. What works for one person might not suit another, and preferences shift as players progress. One of our primary objectives is to build flexibility into GTO Wizard AI, accommodating different types of learners. For example, in chess engines like Stockfish, users can adjust the algorithm’s depth to balance the time spent calculating the best move with the quality of the solution. The deeper the search, the better the result. In AI poker solvers, the combination of Counterfactual Regret Minimization (CFR) and Neural Networks has drastically accelerated the solving process. However, current models solve up to a fixed depth-limit—one street at a time—preventing the natural trade-off between speed and accuracy. While increasing the number of CFR iterations might seem like a solution, this approach isn’t viable since too few iterations produce poor strategies, and after a certain point, the solution stops improving. This contrasts with chess algorithms, where longer solving time leads to continuously improving solutions. Looking forward, we envision a future where GTO Wizard AI integrates a growing depth-limit that will allow players to specify the speed-accuracy trade-off that they’re comfortable with. This would allow for instant, near-optimal solutions, while also giving players the option to dive deeper for more precise answers when time allows.

One of the most ambitious goals of GTO Wizard is to solve any poker variant in just a few seconds. Whether it’s 2-player No-Limit Hold’em (HUNL) or more complex multiplayer variants, we are building scalable methods that can handle any game. These methods would allow us to train an agent from scratch to superhuman performance in a few days, for any poker variant that currently exists or is yet to be invented. To achieve this, we are working on algorithms that can scale arbitrarily with compute and data, ensuring that they can handle larger and more complex games without compromising on speed.

Beyond Solving

While our primary goal is to solve any poker game in a matter of seconds, solving alone isn’t enough—players need to understand why the solutions are optimal. Simplifying the solutions to make them more intuitive and easier to study is key. There are many exciting avenues of research for GTO Wizard AI in this direction, such as providing so-called optimal pure solutionsSolutions that simplify optimal strategies that involve mixing different actions for a certain hand to strategies that feature purely a single action for a certain hand. Higher simplicity comes at the cost of performance (EV). These solutions would be optimal in the sense that they trade off the least amount of EV as possible. or forcing the solver to generate simpler and more human-implementable strategies.

Another promising path for simplification and explainability lies in the integration of Large Language Models (LLMs). Engaging with solvers through natural language and receiving explanations in layman’s terms would greatly increase the accessibility of solvers. Combining solvers with LLMs is an exciting field of research for GTO Wizard, offering poker players all the tools they need to understand the complex game of poker.

The future of poker study is incredibly bright, and the advances being made by GTO Wizard will redefine how players approach the game. With a focus on speed, flexibility, and personalized learning, the next generation of poker tools will be more powerful and accessible than ever before.

If you’re passionate about pushing the boundaries of technology, working on cutting-edge advancements, and unleashing the power of state-of-the-art machine learning algorithms, we want to hear from you. Join us in redefining how poker should be studied.

Check out the available positions here.