How EDGE POKER Inspired POKER Q’z’s Vision for the Ideal AI Poker App

Introduction

As an engineer working on the poker training app POKER Q’z, I spend my days building products and designing algorithms at the intersection of poker and AI.

Over the past few years, Japan’s poker app market has grown rapidly. Apps like EDGE POKER, m HOLD'EM, Poker Chase, EARN POKER, and Poker Fate have made it easier than ever to play poker casually on a smartphone. As both a player and a developer, I’m genuinely happy to see this momentum.

One of the most talked-about titles lately is EDGE POKER, which strongly positions itself around poker and AI. Its standout feature is a proprietary AI called EDGE AI. After trying it myself, I found parts that felt impressive, and other parts where I thought, “With a few changes, this could become even more interesting.”

In this article, I’ll use EDGE POKER as a case study to organize, from a developer’s perspective, how poker apps can evolve going forward and what kind of technology we’re building at POKER Q’z.

Note: This article is not written to criticize. I’m writing it because I want the entire industry to become more exciting.
Note: To respect the terms of service, I’m not including EDGE POKER screenshots.

Point of Concern 1: How AI Is Used

GOD's MOVE

First, let’s talk about GOD's MOVE (a feature where AI automatically selects the optimal action). By using a paid item called “Crystals,” players can use this feature even in the rated mode, “Ranked Games.” In other words, it effectively supports or even replaces the player’s decision-making. If a feature like this is allowed in ranked play, results no longer reflect pure skill alone.

In addition, pay-to-win advantages in ranked games aren’t limited to GOD's MOVE. The game also offers effects that directly influence rank movement, such as maintaining consecutive placement bonuses.

When paid features create in-game advantages and disadvantages, rank starts to represent not “player strength,” but “how effectively someone used features and items.” As a result, the meaning of rank and competitive fairness can weaken.

Poker is fundamentally a game where players compete in decision-making and strategy under the same conditions. Of course, using GOD's MOVE in a non-rated mode like “Single Game” can be a great learning and onboarding experience. But at least in ranked games, I believe a non-pay-to-win, skill-based design should be the baseline. If you use AI, it should not play for you during the hand. It should focus on post-game review and learning support, which aligns better with the essence of poker.

One representative approach is an explanation feature, and at POKER Q’z we’ve continuously strengthened this as a core function. We develop systems that automatically generate explanations based on your decisions, and technologies that analyze your tendencies and play style from hand histories to deliver personalized feedback. These are our proprietary core technologies, and we are currently filing patents.

BUDDY AI

Another feature that caught my attention is BUDDY AI (an AI that imitates your play). The concept itself is technically fascinating: it learns your tendencies and plays on your behalf. But from a user experience standpoint, I still have questions.

Poker is, at its core, fun because you struggle, think, and choose for yourself. That’s why, rather than watching an AI play “instead of you,” I believe the experience of
improving while you play has more value as a game.

If you want to use AI on a player’s data, it may fit better as a coach that:

• automatically analyzes your tendencies
• points out your weaknesses
• explains “this is a better action here” in natural language

In terms of learning and growth, AI as a coach and explainer seems like a stronger match.

Point of Concern 2: AI Performance

Because EDGE POKER positions poker and AI as its main strength, expectations naturally rise around AI accuracy and consistency. While the app includes many convenient features, I also noticed several behaviors that felt “potentially risky” when viewed as a learning tool.

Three points stood out in particular.

Mismatch Between In-Game Advice and Post-Game Review

In some cases, the action the AI recommended during play differed from the “best action” shown in the post-game review. Even though it was the same spot, the conclusion changed, which can cause issues such as:

• users not knowing which guidance to trust
• learning becoming inconsistent
• doubt about whether the AI is actually correct

Consistency is critical for any learning-support AI. If real-time advice and review use different evaluation logic, the tool loses credibility as educational content.

Recommendations That Are Clearly Unnatural

In a few spots, the AI recommended actions that were clearly far from sound theory. For example, it suggested folding a completed straight, which is almost never a realistic decision in practice.

This isn’t just a minor optimization gap. It can actively push players toward misunderstandings, which is fatal for a learning tool. For poker AI, being “strong” matters, but being “clearly not wrong” matters even more. Without basic strategic consistency, users can’t safely rely on the AI.

Unclear Logic for Range Analysis and Hand Classification

I also felt some friction with the range analysis feature. It doesn’t provide a clear, top-down view of overall range strategy, and the information it shows feels fragmented. On top of that, I noticed issues such as:

• unclear criteria for strong, medium, and weak
• no clustering based on hand properties

For example, a very strong draw like a gutshot plus a flush draw sometimes gets labeled as “one pair” because the board is paired, which doesn’t match real-world poker intuition.

Range visualization isn’t just about “classifying hands in a plausible way.” It’s an information design problem that directly affects decision-making. If the display doesn’t align with intuition and practical play, users lose trust in the output and stop using the analysis features. Thinking in ranges is at the heart of poker study, so clarity in classification logic and practical consistency are essential.

Thoughts on EDGE AI

EDGE AI is designed to highlight the experience of “AI solving poker,” by recommending optimal actions during play and even making decisions automatically.

Reference article
https://note.com/edge_poker/n/n97dbb01b7cd8

Comparing the official explanation with the actual behavior, it seems likely that EDGE AI uses something close to an ML-based approximate solver model, such as:

• training a model on GTO strategies computed via CFR and similar algorithms as teacher data
• extracting features from the current situation using rule-based logic
• feeding those features into the model to predict EV for each action and selecting the highest-EV action

This is a very standard approach and technically interesting. At the same time, after playing with it, I also felt the unique challenges that come with poker AI. Below are the key points that stood out to me.

There Aren’t Nearly Enough Training Situations

The first issue is a lack of training coverage. In poker, even small changes can dramatically shift the optimal strategy. Position, stack depth, bet sizing, board texture, ICM pressure, and many other factors matter, and the state space is effectively close to infinite. That makes “training sufficiently on every situation” an extremely difficult problem.

Once the model enters an area not covered well by training, accuracy becomes unstable. This is less about whether the algorithm is good or bad, and more about a structural difficulty: poker is a game where training data coverage is inherently hard to achieve.

Feature Design Doesn’t Match Real-World Poker

Another major point is the design of input features.

Based on the behavior, it appears to use basic information such as table-wide stack conditions, current made hand category (high card, one pair, etc.), and whether draws exist. But in real poker, that’s not enough.

In practice, decision-making depends heavily on context such as:

• hand strength that includes future potential
• range narrowing based on action history
• blocker effects
• proper defense frequencies
• ICM

In poker AI, feature design essentially determines the “resolution” of strategic understanding. If features remain too coarse, the model will struggle to reproduce decisions that feel accurate in real play, no matter how much training you do.

You Can’t Build Poker AI with Engineering Alone

This was the strongest takeaway for me. Building poker AI isn’t just about machine learning or model accuracy. Many critical design decisions come earlier:

• what features to include
• which information truly matters for decisions
• how far you can abstract while still expressing strategy

You can’t answer these well without deeply understanding poker itself. In other words, you need not only the ability to implement algorithms, but also real, practical poker knowledge. Only when both come together can you build a poker AI that’s actually useful.

And at POKER Q’z, many team members bring both engineering and player perspectives. That’s why we can build products that directly improve the learning experience.

The POKER Q’z Philosophy

Let me briefly share the mindset behind how we build POKER Q’z. Our goal is simple: use AI to deliver the best poker learning experience possible.

We don’t just want to create “a poker app you can play.” We want to build a product where you naturally get stronger simply by playing. Poker is a deep, strategic, intellectually rewarding game. At the same time, it’s also true that the learning curve is steep.

• You don’t know what the correct play is
• You don’t know why you lost
• You don’t know what to fix

This vague uncertainty becomes the biggest barrier to improvement.

That’s why we prioritize helping users improve as efficiently as possible. At the same time, we don’t want it to feel like studying. We want a learning experience where you’re having fun playing, and before you know it, you’ve improved.

We want poker to become not “a difficult game,” but “a game where you got stronger without realizing it.” That’s the ideal we aim for at POKER Q’z.

Strategic Accuracy That Supports Learning

The most important foundation is strategic accuracy. While we continue developing AI and algorithms, we’re building the infrastructure to present strategies that are as theoretically correct as possible in any given spot. Even the clearest explanation is meaningless if the underlying strategy is wrong. As a learning product, we believe the minimum requirement is being able to provide answers users can trust.

AI as a Coach, and Natural-Language Explanations

Another key element is how we communicate. Simply showing the correct strategy doesn’t make players truly stronger. Players improve when they understand why an action is good and what thought process supports it.

That’s why we treat AI not as something that plays for you, but as a coach that supports your thinking. We avoid dumping numbers and jargon. We explain in words that even beginners can grasp intuitively. We aim to communicate not only what happened, but also why it happened. By using generative AI and natural-language technology, we work toward an “explanation experience” that helps players internalize the thinking process.

Inside the app, we also include an AI assistant that users can ask freely about strategy and specific plays. Many users have responded positively to it as a learning support feature. We continue improving it to provide more accurate answers and clearer explanations for beginners. We also offer exercise-style quizzes with explanations tied to each lesson theme, so users don’t just consume knowledge but reinforce it through practice.

We don’t want AI to simply “give you the answer.” We aim to deliver a learning experience that helps you build the thinking process that makes you stronger.

Final Thoughts

We want the poker industry in Japan and around the world to keep growing. We still have a lot to learn, but we’re seriously looking for partners who want to help grow the poker world.

• Companies and partners interested in collaboration
• Influencers who want to work with us

And we’re also actively recruiting people to join POKER Q’z, including interns, in areas such as:

• Business development (Japan and global marketing, creative production, and more)
• Engineering
• AI and algorithm research

“Poker can be even more fun.”
“AI can create a better experience.”

If those ideas excite you even a little, feel free to reach out.
Let’s make Japan’s poker scene even more exciting together.