Former Tencent Technical Designer Starts AI Game Development "Maggie's Garden": In the AI Era, No Game Creator Can Afford to Stay on the Sidelines | Creative Notes

This article is authorized to be republished from "Game Artist".

Interview/Text: Xia Qingyi, Wang Chenqian

When it comes to AI games, most ordinary players might think of preset "Chatbot" AI NPCs that can hold conversations, "smarter" AI teammates or enemies in competitive games, or AI acting as a kind of GM to guide players through the game.

However, Maggie's Garden, an AI sandbox game planned for release this year, focuses on exploring the systematic application of AI at the game mechanics level.

In the game, players can import images to automatically generate NPCs, which are driven by AI to generate storylines, dialogue, and voice in real-time. This provides players with a more personalized and immersive experience, stimulates their motivation to create and socialize, and establishes a game ecosystem where users and AI coexist.

Zhang Haoyang, the producer of Maggie's Garden, previously founded startups in the VR/AR gaming and AI digital human fields. He later joined Tencent in 2020, where he served as a technical designer responsible for AI-related work in Game for Peace. During his time at Tencent, he also participated as a producer in developing Eden Island, an AI game similar to the "Stanford Town" concept.

In 2023, Zhang left Tencent and founded Shenzhen AutoGame Company, focusing on the integration of AI and games. The company has already secured tens of millions of yuan in funding.

Zhang believes that future AI games will undergo systemic changes. He divides the development of AI games into four stages and judges that the industry is at a critical juncture, moving from "AI NPC" to "AI UGC."

Among these, AI UGC will revolutionize the mode of content production and consumption, delegating the creative rights of the gaming medium to players, evolving into a brand-new ecosystem where everyone participates and creates together.

01

Do Otaku Prefer It?​

Let Players Create Their Own "Lifelike" AI Companions​

In October 2024, Maggie's Garden released a demo on Steam and began its first closed beta test. The game's novel gameplay and solid quality attracted a large number of players. Without participating in the Next Fest, it shot up to 45th place on the Steam New & Trending list within a day and accumulated over 5,000 wishlists in a month, exceeding the development team's expectations.

The game adopts a pixel art style and a Japanese fantasy story setting. Players take on the role of a retired hero, gathering resources, recruiting companions, battling on adventures, and gradually building and developing their own camp.

The main feature of the game is the unique "AI companions." Players can recruit their own AI companions by uploading images or inputting descriptions. Each companion has its own unique personality and story—they plan their own schedules, complain about the details of camp life, proactively build relationships with the player, and observe the situation and follow commands in battle.

Compared to traditional NPCs, with the support of AI, they are more like living individuals truly existing in the game world.

This "lifelike" feeling creates a stronger customized experience, satisfying players' need for emotional companionship. Some players raise their "OCs" in the game, others create their dream partners, and some have even played the small-scale demo for over 100 hours, just to repeatedly generate different characters to accompany them on adventures.

Zhang told Game Artist that the conception of Maggie's Garden was not a flash of inspiration. While still working at Tencent, he was optimistic about the potential of AI games and began conceptualizing the prototype of this game in 2022.

In 2023, Stanford University published the paper that sparked widespread attention across the industry. It built a virtual "Stanford Town," creating 25 AI agents with different personalities who autonomously act and freely live in the town.

The Virtual Stanford Town

This paper also greatly inspired Zhang, not only confirming his direction to some extent but also proving the trend and feasibility of AI games. So, he led an 11-person team to create a "Stanford Town-like" game called Eden Island.

Maggie's Garden is a further iteration based on Eden Island. He hopes to make it the "Avatar" of AI games, exploring more feasible paradigms for AI gaming.

For example, the sandbox genre was chosen because it can accommodate various mini-games, suitable for AI to shine. In the future, new features like AI-generated items, AI-generated complex tasks, and multiplayer can be added to further enhance playability. Based on these characteristics, sandbox games also have the potential to develop into social products.

From the user's perspective, players born in the 2000s and 2010s spend a lot of time on sandbox games and have a higher acceptance of AI, making them suitable as "natives of the AI game era" to be nurtured and cultivated.

It is understood that currently, the users of Maggie's Garden are mainly divided into three categories: one is the sandbox game audience, especially fans of games with similar art styles like Stardew Valley; another is users of AI voice chat products, who see the AI NPCs in the game as an upgraded alternative to AI voice chat products; and the third is the otaku audience, especially players who enjoy fan creations or self-insertion.

Zhang himself is also an "old otaku," a fan of Vocaloid (the virtual singer project to which Hatsune Miku belongs), and his work computer desktop features the female protagonist of Code Geass. He said that the game was not initially aimed at otaku players, but this group generally has a high acceptance of new things or enjoys derivative works. Maggie's Garden allows them to easily create their favorite anime and game characters in the game.

​

02

​

A Snowball Revolution for AI UGC

There are already many games on the market related to the AI concept. They range from small independent game teams creating small-scale games with AI as the core selling point, such as AI card games and AI text adventures, to large companies increasing their investment in AI, integrating it into their existing flagship products and projects. For example, Tencent's Game for Peace has deployed AI teammates in the game to improve player retention.

Zhang said that Maggie's Garden is significantly different from many existing games that incorporate AI. Its entire game world is built on top of scripting languages, and all game logic can be generated and updated in real-time by AI.

Whether it's using AI to assist with part of the text creation or deploying AI teammates in competitive games, the mainstream application of AI in existing games mostly involves embedding AI capabilities into traditional game systems to improve a specific game aspect or function.

In Maggie's Garden, the entire game system is combined with AI, leveraging AI capabilities in every aspect of the design.

The characters' actions and dialogues are all generated by AI

For example, to create an item in the game, you need to consider its interaction logic with other objects and characters, balance its numerical values, and create textures, etc. The various elements in the game system are interconnected. The latest demo of Maggie's Garden can already create this series of game logic through natural language input, with AI completing the work at each stage.

Zhang said that designing AI to be deeply embedded in the game system is intended to achieve a more thorough AI UGC.

He believes there is a supply-demand mismatch between the amount of content produced and the demand for content in the gaming industry. The cost of content production is increasing, while on the consumption side, not only are the quality requirements for content higher, but the consumption speed is also faster. Around 2020, a game of the scale of Genshin Impact would take three months to produce a new version that could sustain players for six months. Now, the same amount of content produced in the same time might be consumed in just two weeks.

In fact, the gaming industry explored this issue several years ago, trying to solve it through UGC. However, pure UGC can also become a drain on users. Adding AI significantly lowers the barrier to UGC and improves efficiency.

But AI UGC is not simply about using AI tools for UGC creation.

Zhang said that previously, tools or products that marketed the concept of AI UGC did not fundamentally change the traditional content ecosystem, mainly because the usage threshold was high, often requiring creators to have coding skills, and top-tier content dominated the attention of most consumers. For example, the top twenty games on the well-known overseas sandbox game platform Roblox might account for 80% to 90% of all creator revenue.

 

Roblox Creation Interface

The AI UGC model he more strongly endorses involves using AI to generate game logic, bringing systemic innovation to the game's production and consumption modes. If AI has the ability to create content for events, items, characters, and other aspects through natural language input, it can significantly lower the barrier to creation.

To form such a model, sufficient training data is needed, including player data reflecting play preferences, game code, etc. Rather than manually finding massive amounts of data to feed the AI, a more reasonable approach is to design a mechanism that continuously generates new code, allowing the AI to iterate and grow on its own.

Games are the best carrier for establishing this AI UGC cyclic ecosystem. Players participate in shaping the game experience and generate new data in the process, helping the AI iterate, creating a "snowball effect" that continuously improves the game world and proactively caters to player needs.

"It's like the hypothetical scenario in the anime Sword Art Online, where the activities of all players collectively form the server's history. There's no need to chase top-tier content because every player has created their own story."

03

From Big Companies to Small Teams

Everyone May Find Entrepreneurial Opportunities in the AI Game Era

Before founding AutoGame and developing Maggie's Garden, Zhang had two previous entrepreneurial experiences, involving VR/AR games and the AI digital human field. In his second startup in 2017, the focus was on using NLP (Natural Language Processing) technology to create smart speaker products that could interact with 2D virtual idol characters.

His previous entrepreneurial experiences and his time at Tencent made him realize the huge potential of combining AI with games, leading him to start his third venture.

In Zhang's view, the process of AI changing the form of games can be divided into four stages.

Stage 1.0 is called AI Novel. At this stage, AI only generates images and storylines, mainly used in text adventures or tabletop RPG-like games. Participants are primarily traditional algorithm engineers rather than game creators, and game design often has flaws.

Stage 2.0 is called AI NPC. At this stage, AI's function is to intervene in NPC behavior logic, providing players with emotional value and real-time feedback.

Stage 3.0 is AI UGC, where AI learns how the game operates and creates new content autonomously. However, due to insufficient data, human participation is still required.

Stage 4.0 can be called the "God Model," where AI understands both player needs and the rules of the game world. It can generate world events, new game rules, etc., without human intervention, directly changing the game form and creating new game content.

According to this four-stage division, AI games globally have reached stage 3.0, moving from the second half of AI NPC to the early stage of AI UGC. The ecology of future game content creation and consumption will change, and opportunities lie within.

Currently, there is no shortage of well-known game developers entering the AI game field. Former miHoYo CEO Cai Haoyu released a trailer for Whispers from the Star, a multi-modal real-time interactive AI game, on X, sparking huge discussion. Former Giant Network CEO Wu Meng and Zulong Entertainment's star producer Zhao Tongtong also left their jobs to start AI game ventures, both securing funding.

Whispers from the Star

Compared to star producers investing in large-scale projects like AI GTA or AI wuxia worlds, Zhang is currently focused on smaller prototype validation. For small startup teams, many niche AI game designs are already highly feasible from a technical standpoint.

Zhang said that the development of technology actually surpasses ordinary people's perception, but game products haven't yet fully utilized these technologies. AutoGame itself has overcome many technical challenges during application.

For example, to solve the problem of large language models not being able to upload unlimited contextual data, they conducted in-depth research on RAG (Retrieval-Augmented Generation) technology. This gives AI better memory capabilities, allowing it to remember and understand historical events, main storyline progress, settings, and item connections in the game. At the same time, they need to continuously receive player feedback to iterate and optimize prompts. A single prompt engineering file might be thousands or even tens of thousands of words long.

Fortunately, many underlying technical issues of models are naturally resolved as external technologies develop. For example, early AI lacked advanced matting techniques, but in 2024, Layer Diffusion's author Zhang Lvmin solved AI matting, making it directly applicable to the real-time operation of games. This allows startup teams to focus more on solving problems in vertical niche areas.

04

Conclusion: In the AI Era,

No Game Creator Can Stand Aloof

"AI won't eliminate people; it will only eliminate those who haven't used AI," Zhang said. "That's what I thought before, and later Nvidia's Jensen Huang publicly said something similar."

Although there are still many controversies and risks associated with using AI in the gaming field, especially in art, the trend is significant. In Zhang's view, AI will completely reshape the entire gaming industry, and no game creator can remain outside it.

He reflected that in less than a year, AI's coding level has evolved from a junior/mid-level programmer to nearly an architect level. AI is getting smarter. Believing in technology, adapting to it, and leveraging it might be the better choice.

The capital market has recognized his views. AutoGame is advancing financing to support subsequent R&D. As of the first half of this year, multiple rounds have raised tens of millions of yuan.

Next, the team will go all out to develop Maggie's Garden. It is currently scheduled for October 13, 2025, to officially debut at Steam Next Fest, and may be ported to mobile platforms based on market feedback.

The game's release is not the end. Zhang and his team plan to continuously add new features, gradually polishing it into the ideal form they envision.

He hopes to be a pioneer on this path, to be the first to touch the future.