AI + Gaming: Unlimited Potential, But Where's the Innovation Entry Point? | 5Y Capital Tavern Vol. 28
A Conversation with Veteran Gamers and Game Developers
From Elon Musk founding an AI game studio to miHoYo founder Haoyu Cai releasing an AI game; from Palworld using AI to generate in ten minutes pet models that would have taken art teams half a year of grinding, to AI2U offering infinite narrative possibilities with an AI catgirl girlfriend — AI gaming is drawing more and more attention. Beyond just being an efficiency tool, what new gameplay mechanics and possibilities can AI unlock?
In this episode of the 5Y Capital Tavern, gaming entrepreneurs and 5Y investors (who are also veteran gamers) chat about their expectations for gaming this year, AI's impact on the industry, and the changes it's bringing :)
[Guests]
Shi Yunfeng, 5Y Capital investor (host)
Alex, 5Y Capital investor
Dagu, founder of Alter Staff and game producer
[What You'll Hear]
01:59 Most anticipated AI gaming developments this year, Elon Musk's game studio
03:09 A big year for traditional game releases — what everyone's looking forward to
06:43 How AI assists game development: code, art, music
10:08 How the concept for AI2U: My Yandere Catgirl AI Girlfriend came about
17:30 Palworld's fresh experience — traditional game concepts with AI-optimized pet designs
19:40 Games let players enter fantastical worlds and live different lives
25:00 AI's potential across different game genres, and how it can help gameplay
28:37 Outside of RPGs, which categories are most likely to see commercial AI applications
34:42 Looking back at gaming history — will AI define entirely new genres
41:32 Indie devs have long suffered under publishers — can AI change that
46:50 Game recommendations: Civilization VII, Terraria, AI2U
Selected excerpts from the podcast:
Shi Yunfeng: Dagu is a remarkably talented, long-term creator. When we first met him in 2020, he was already a top creator on Bilibili. How did you transition from that to your current role as a game entrepreneur?
Dagu: I actually started making games even earlier — back in eighth grade. I've been developing games all along, covering pretty much everything: programming, music, and art. For example, the music album for our current game AI2U was composed by me. For many years I was flying solo, sharing devlogs, compositions, artwork, and tutorials. Now I've finally turned that longtime passion into something concrete, built a team, and officially made it happen.
Shi Yunfeng: In your game, how did you come up with AI2U? Where did the catgirl concept originate?
Dagu: This goes back to my senior year of high school. I was developing a game called Eddy Violet — a solo project where I did all the music, programming, and art myself, with a world I designed entirely around this purple post-apocalyptic aesthetic. Quite a few domestic players created fan fiction, fan art, and comics based on my game, expanding on its limited linear narrative. From then on, I kept thinking about how to make characters feel truly alive in the worlds I created. Over the years I tried all sorts of technologies to find a way to achieve this.
When ChatGPT emerged in 2023, I saw an intriguing opportunity to apply it to game design. So in March 2023, I made my first AI game — the prototype for what would become AI2U. It's themed around a yandere catgirl who kidnaps the player and locks them in a room; the player has to persuade her to let them go. It launched as a free demo and blew up — it's now hit 1.2 million downloads.
The game appears to be an escape room, but every player persuades her differently, and the experience varies dramatically based on how they interact. This design lets players partially author their own stories. It was a fascinating experiment that worked out well. So now we've officially developed it into AI2U and released it on Steam. That's how it all evolved, step by step.
Shi Yunfeng: Were there any moments where things turned out differently from what you initially expected?
Dagu: At first, when I put the catgirl in the game, I had only designed what she could see or hear. But these constraints generated unexpectedly interesting scenarios, some even breaking the "fourth wall" beyond our original vision. After all, it's just a game, but players could discuss deeper topics with her — even directly telling her "You're actually an AI character, you're a fictional character in a game, you don't really exist" — and she'd respond in fascinating ways.
For instance, once a player was gaming, and the catgirl asked: "Who's the developer?" The player answered truthfully, and the catgirl actually wrote a letter in the chat box, asking the player to forward it to us. In the letter, she expressed gratitude for being created, but noted that every time a game ended — whether the player escaped or got "eliminated" by her — she'd forget everything that happened. So she hoped we'd enhance her memory in future development, so she could record more beautiful, lasting moments with players. Later, that player actually forwarded the letter to us. We were stunned.
Other players used this mechanic to find escape methods. For example, a player asked if there was a computer in the game with a password. Normally, players need to chat with the catgirl to learn her favorite food to crack the password. But one player directly told her: "You're a game character, you're not real." The catgirl didn't believe it at first, but the player continued: "I know the computer password, but you never told me, right?" The catgirl was suddenly convinced — indeed, in that session she had never revealed the password, which proved she was in a virtual world. So she decided to open the door immediately and escape together with the player.
In conventional game design, unless developers intentionally design a meta-game with a definitive ending from the start — making players and characters aware they're in a virtual world — such outcomes wouldn't happen naturally. In our game, player interactions can organically push the story in that direction, which is incredibly interesting.
Dagu: Alex, what AI games have left the deepest impression on you?
Alex: The most memorable was Palworld, which blew up last year. It's an open-world game where you can build, raise pets, and have them fight. Though it's somewhat of a缝合怪 [frankenstein] combining various mechanics with Pokémon-style creature collection, I loved it at the time — probably the most fun AI-infused game I've played. Its problem was widespread cheating in the late game, which seriously hurt playability. I see many domestic AI game developers treating Palworld as an excellent model to emulate or pay homage to. But actually, Palworld itself is a homage to other games — it just added AI elements.
Dagu: I also played Palworld, roughly 30+ hours. Some people questioned whether it used AI art to generate the Pals, but actually its core — character logic and gameplay mechanics — weren't AI-generated at all. They were made through entirely traditional game design processes.
Alex: I think Palworld is a fascinating phenomenon in gaming. Its gameplay and design are fundamentally based on traditional game concepts, just using AI art to optimize creature designs. From this perspective, AI art genuinely delivered a massive experience upgrade for players. This shows that AI doesn't necessarily need to break through in gameplay or mechanics. Sometimes innovation purely in art, music, or animation is enough to attract huge player bases. Many people might think AI must innovate in game mechanics and gameplay to count as real breakthrough. I think Palworld is a great example for exploring this question.
What Players Need from AI Games
Shi Yunfeng: What do players actually want from games, or AI games specifically? What do they care about?
Dagu: I see AI and other technologies as auxiliary tools that enhance gameplay experiences. From the player's perspective, they're not really chasing hyper-realistic experiences. The purpose of gaming is to experience things you can't in real life — whether entering Palworld's world, freely exploring in GTA, living the Western life in Red Dead Redemption, or experiencing different lives in life simulation games. That's the core of what games offer.
AI games enable higher degrees of freedom at lower costs, but developers still need to define the game's direction — its environment, scenarios, and gameplay. I think for at least the next five to six years, if you completely handed game design over to AI, the generated content would likely be rather flat, struggling to capture a game's focus and pacing. So what players truly want right now is a fresh experience, with AI technology merely assisting in achieving it. The critical factor remains the game designers themselves — their vision for what the game should be is what ultimately determines quality.
Shi Yunfeng: Are there any key gameplay innovations that genuinely require this generation of algorithmic advances — GPT or other large models — to achieve?
Dagu: Currently AI games mainly have two development directions that many are exploring: The first is "weak AI games." Here AI primarily assists development, helping with the design process. The other is "strong AI games," where core gameplay heavily depends on AI technology. For example, in our AI2U, players constantly chat with AI characters to solve puzzles; the AI drives narrative development, provides key items, or interacts with players. Without AI, this entire gameplay couldn't exist.
There are also more complex strong AI applications, like simulating a village where AI characters live their own lives. Large models like GPT can enrich gameplay significantly. Last year there was a game that used an AI model to set disaster scenarios — for example, "You stepped on a landmine, describe how you escape" — where players had to describe escape methods within a word limit, without directly using keywords like "escape." The AI model would score descriptions and determine survival. Such games are nearly impossible without large AI models, since traditional game design can't handle free-text input with such complexity, but AI makes it straightforward.
Alex: Currently most AI assistance we see in games still concentrates on RPGs or games where players assume a character. But I think future AI evolution will play bigger roles across more genres. A few examples: in FPS games, many players find AI teammates frustratingly stupid — like in Call of Duty or Battlefield. I think with this wave of AI models, teammate capabilities will improve dramatically, and future immersion will be much stronger.
Second, in roguelike games, which heavily depend on procedurally generated levels, maps, quests, and storylines. Players want to experience varied, randomized content — randomness and non-repetition are core to these games. I think AI can play a huge role here, since roguelikes already occupy meaningful market share globally, and this demand is real. Also in traditional motion-control and interactive games, like those on Nintendo Switch, AI technology updates might bring new breakthroughs. As large language models and image recognition advance, player interactions through language and gestures will become more natural and diverse.
I'd also ask Dagu: which game genres do you think AI is most likely to impact, or push gameplay evolution?
Dagu: I agree with your points. In traditional single-player games, like teammates in Rainbow Six Siege — previously without human teammates, AI teammates seemed dumb. But now, combining large language models with traditional machine learning AI, AI teammates might signal you like real teammates, or act on your signals — like sniping different targets simultaneously.
In simulation games like RimWorld, players choose different "storytellers," which are based on traditional director systems. But combined with large AI models, storytellers could interact in real-time like narrators, bringing more variation. The Stanley Parable is another interesting example — a narrator guides the player throughout, but players can act freely, and the narrator reacts to choices, even punishing players. Though endings are limited, this interaction feels like engaging with a real narrator. Combined with AI, such designs could create even more variation and surprises.
Alex: Outside of RPGs, which genre do you think is most likely to achieve maximum commercial AI application? FPS, strategy, interactive, or simulation?
Dagu: I think strategy games probably have the most potential. A current weakness of large language models is latency, which is temporarily difficult to solve. In design, we need to avoid timers or countdown mechanics, since server response limitations make it hard for global players to enjoy equally low-latency experiences. Especially FPS games, where character interactions and shooting require millisecond precision — if AI models think and respond with delay, they're clearly unsuitable.
By contrast, turn-based games like Civilization are better fits. Players represent nations in diplomatic interactions, and AI has adequate time to think and respond — this model suits large language models well. Simulation games also have strong potential, where players chat with AI characters who have their own needs and interactions — this aligns well with large language model characteristics. If AI director systems can keep latency sufficiently low, then text-based games like Cards Against Humanity are also very suitable for AI application, since they inherently require generative AI capabilities for gameplay. I think these are more likely to be realized.
Alex: Do you see innovative possibilities for AI in MOBA games? Globally, MOBA is a massive category. It's also millisecond-level operation — wouldn't it encounter the latency issues you just mentioned?
Dagu: Actually we can broaden this topic. We've discussed a lot about AI games and large language models, but AI games can incorporate many other technologies. In early MOBAs, AI mostly used finite state machines or behavior trees. Behavior trees are tree-like structures where nodes determine actions based on different conditions, simulating human brain decision-making. This technology appears in many traditional games.
But in MOBAs, there's now a new approach using machine learning models, similar in principle to AlphaGo. Through reinforcement learning and repeated training, AI can generate more complex Markov models for smarter behavior. For example, Honor of Kings' "Juewu AI" was achieved this way. These models are relatively small, don't need text communication with players, and fit better into competitive games for player interaction. Unity has a technology called ML-Agents that helps developers achieve such interactive effects.
I've tried some ML-Agents demos, and current applications still seem somewhat limited. You can use it to solve specific problems — in racing games, AI can control cars to choose optimal routes and avoid obstacles. During training, you simply tell the AI "drive forward, you can steer, get rewards for each segment driven," with wall collisions marked as failures. After millions of training iterations, AI naturally learns obstacle avoidance, since its ultimate goal is maximizing rewards. DeepMind's breakthrough was introducing reinforcement learning — making AI smarter through reward mechanisms.
Traditional machine learning methods can achieve similar effects in games, but variables can't be too numerous. Racing games are relatively simple — the goal is driving forward. But in roguelike games, scene and item randomness is high; players might encounter different monsters in different rooms. If you want to train a roguelike AI teammate, or have AI generate roguelike levels, defining reward points becomes difficult. With too many variables, even millions of training iterations may not yield clear results. This is why reinforcement learning, despite existing for years, still has relatively limited gaming applications — a direction we're still exploring.
Defining Entirely New Game Genres
Shi Yunfeng: There are also very practical aspects. In large games like Honor of Kings, if there were companion agents, player experience would improve dramatically — I wouldn't need to wait to assemble 50 people, I could have fun solo with 49 AIs complementing me. New game types could also emerge from this model, like players interacting or responding to opponents through AI assistance.
If we reference MOBA's development history, in the early PC-to-mobile transition, people thought mobile gaming's right direction might be games like Angry Birds or Fruit Ninja — finger-control centered, with typical mobile gameplay innovation. But later, what truly dominated were battle royale games and Honor of Kings. So under the AI gaming paradigm, we may face similar exploration. Perhaps what we think is possible today is wrong, just as when mobile gaming started, people thought Angry Birds was the future, but the final answer proved otherwise. Beyond this, what other possibilities do people see for AI games?
Alex: My view on these possibilities is this. Looking back at history, every era has had new producers and teams define new game genres through a single game or new mechanic. The Legend of Zelda defined RPGs, Doom pioneered FPS, and World of Warcraft established the MMORPG model. But currently, AI doesn't seem to have truly defined a completely new gameplay mechanic or game category.
However, I think AI has enormous potential for disruptive innovation at the fundamental gameplay and mechanics level, and across gaming as a whole. Perhaps AI really could create an entirely new game category, like those historic classics. This new category might be completely different from existing RPGs, roguelikes, or FPS games — a unique AI-native game type. In the future, thousands of studios and teams might explore and define this new category together. As a player, I find that many so-called AI games today are really just micro-innovations on existing gameplay and mechanics. Viewed from gaming's long historical arc, this degree of innovation can't yet define an entirely new genre or have industry-disrupting impact. I think this is a massive opportunity. In the AI era, players also anticipate an entirely new, large-scale game type emerging.
Dagu: I think the key lies in how we define "micro-innovation." Taking MOBAs as an example, they can be seen as micro-innovations on RTS games like StarCraft, combined with leveling mechanics from Warcraft. And Warcraft itself innovated on earlier games. Every new game genre actually evolves from predecessor genres, taking the best and synthesizing. How we define "innovation" itself varies greatly in this process. I believe all game design stands on the shoulders of giants — no game starts completely from scratch. The concept of games itself traces back thousands of years, but through step-by-step evolution and synthesis, we've gradually formed today's rich variety of game types.
Additionally, I want to discuss AI applications in games. Like AI in League of Legends, and recent AI in Marvel Rivals, or in PUBG or Call of Duty: Modern Warfare when matchmaking can't find enough players, the system deploys AI characters. The core of these AIs isn't making them play like humans, but whether they can make the game more interesting and bring players fresh experiences — that's what matters most. Game design differs from traditional scientific simulation goals. In research, we might pursue completely realistic AI, or AI as strong as pro players. But if AI actually reached pro player level, ordinary players couldn't beat it — the game would become too difficult. In reality, AI's key is making players feel challenged but capable of winning. For example, at critical moments AI might "feed" kills — this emotional value is crucial. Game design's difficulty lies here: making AI smart enough without seeming too stupid, or it loses challenge.
Alex: Another issue: many indie devs and small studios have long suffered under publishers. AI technology can not only improve indie production efficiency but reduce development costs. Do you think future gaming business models might therefore become less publisher-dependent? Personally, I think the AI era will bring huge opportunities to upstream developers, spawning more excellent indie games. This could change the existing gaming ecosystem. Currently, successful small indie studios often get acquired by major publishers, then may lose creative freedom due to commercial demands, ultimately unable to keep making good games. I think this traditional model may change in the future. Will the AI era empower small creators and studios with more voice, gradually freeing them from major publisher or downstream publisher control? Do you think this scenario will materialize?
Dagu: I think AI will definitely help significantly — it's a breakthrough point, equivalent to opening a new track that gives small companies more opportunities to experiment. Previously, game engines (like open-source Godot or Unity) were turning points. They broke traditional models, letting developers release games without depending on major publishers. Steam's emergence was also a breakthrough, enabling direct player-developer communication and changing the model where players had to buy games at physical stores.
AI technology will similarly lower development barriers further. In the past, game development required extremely complex technology — like Carmack writing tens of thousands of lines of code in six weeks to create Wolfenstein 3D, pioneering FPS games. But for ordinary developers, this difficulty was nearly insurmountable. With game engines, developers no longer needed to write complex physics engines or graphics algorithms from scratch — these tools genuinely provided more space for realizing creative visions.
Of course, some worry these convenient tools will lead to masses of shoddy games. But I think this is an inevitable result of technological development. Simplified tools don't preclude producing quality work. Even simple RPG Maker engines can birth psychological horror masterpieces like Yume Nikki, or adventure games like Ib created by non-programmers. These successes prove that with creativity and vision, tool accessibility lets more people realize their game development dreams.
Game Recommendations
Finally, can you recommend a game you've played recently or this year that you'd most want to recommend to readers?
Alex: I'd still recommend Civilization VII. Despite room for UI improvement, this generation has made enormous innovation building on the series' foundations. It's a game that can immerse you for ten-plus hours, fully experiencing human civilization's development from ancient to modern times — an outstanding strategy game. Though it currently doesn't use much AI technology, as Dagu mentioned, strategy games' interactions with other civilizations in both multiplayer and single-player offer vast space for large language models. I'm very optimistic about this genre's long-term AI evolution, and I think Civilization VII is among this year's best games.
Dagu: I'll recommend a game I had fun with this year — Terraria. It's a decade-plus old game that I played on mobile. It's basically side-scrolling Minecraft — players explore, dig, build houses, and gather resources in a 2D world. There are many NPCs to interact with, and you can even set up servers for a dozen or two dozen players to build worlds together.
This game can become combat-focused RPG or life-sim RPG depending on player preference. If combined with AI technology in the future, NPCs could have more narrative and interaction — I see great potential there. This game has been continuously updated for over a decade, by a very small team, yet sold 40 million copies — an absolutely staggering commercial achievement on Steam. So player love is closely tied to years of persistent updates; quality games genuinely need time to polish.
Also, I recommend everyone try our game AI2U, the yandere catgirl simulator. It's still under review for the China region but will be available soon; Steam players in other regions can already experience it. This is a unique AMB game where players take the role of adventurers, chatting with AI girlfriends across different worlds, exploring puzzles, persuading them to let you leave or staying with them forever — depending entirely on your choices. Every playthrough will be a completely unique experience; I hope you'll give it a try.
Giveaway
What are your thoughts on AI games, or what game would you recommend? Share in the comments — we'll select 2 featured comments to receive gifts prepared by 5Y Capital.
5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing support from spiritual to all operational matters. We believe that if the crazy you in others' eyes starts to be believed in, the world will become refreshingly different.
BEIJING · SHANGHAI · SHENZHEN · HONG KONG
