3 posts tagged with "AI games"

This article is republished with permission from "游戏艺术家Game Art"

​In one sentence having AI create an image, a video, a space, or even a playable game prototype is nothing new. The real question is what kind of experience the generated content delivers, how to keep players around, and even get them to pay for it.

What MetaKid World is exploring is how to turn AI's capabilities—driving world rules, character relationships, quest and story generation—into an open-world story adventure game that players can freely enter and explore.

The core gameplay uses AI to generate dynamic story worlds with infinite possibilities, in a mode similar to AI Dungeon overseas. At the same time, it is also an AIGC digital interactive entertainment platform oriented toward player creation and sharing, currently offering three modes of experience: Big World, Text World, and Interactive Story.

Taken as a whole, MetaKid World is closer to a "2D version of Roblox."

Producer Wang Kai and Chen Mingqiang were college classmates who already started a business together in school. After graduation they joined big tech companies, then reunited because of the development of large language models. Like many young up-and-coming game founders, they hold an ideal of "AI-native games."

Both believe that the key to AI Native at the current stage is finding the entry point for rebuilding games with AI—not building a traditional game first and layering AI on top, but letting AI participate in how the world operates from the ground up.

Constrained by technical capability and cost, a small startup team cannot deliver a highly interactive physical space. A more realistic path is to first pick relatively well-defined game genres to break down—such as cultivation (xianxia) or JRPG—gradually adding AI capabilities within mature game frameworks, and then move from "game + AI" toward games truly driven by AI.

​ALL IN AI, making an AI open-world story adventure game

Wang Kai and Chen Mingqiang are classmates who graduated from Hefei University of Technology with computer science degrees, and they started a business together back in college. At the time, the two built a campus app that brought the class schedule system—previously accessible only through the campus network—to mobile devices, adding social features and school administration systems. Relying purely on word of mouth, the product quickly surpassed ten thousand users.

After graduating, both joined big tech companies with fairly stable jobs and incomes, but they kept exchanging startup ideas. Wang Kai says he is an INTP: when he can't sleep, his mind is full of wild ideas. One night it suddenly struck him—wouldn't it be interesting to let different characters act and interact autonomously within a complete worldview?

It happened that last year large language models were developing rapidly, and AI made this idea feasible. So the two got back together and focused their startup direction on AI games, hoping to "do something meaningful amid the wave of AI."

In the past, a player who wanted to turn an idea into a game had to complete scriptwriting, art production, programming, and more, at considerable cost and difficulty. The team wants to lower that creation barrier: users only need an idea, and MetaKid World can help generate the corresponding characters, quests, stories, and exploration content.

Early on, the product's positioning was not entirely clear. They were often challenged by users and industry insiders: is this a game, or a creation platform? During two consecutive years participating in the "Shulong Cup" AI innovation competition held by Century Huatong, judges raised similar questions.

To find a product direction, the two even sought "outside help" on Xianyu, consulting industry practitioners. Once, someone they connected with on Xianyu, claiming to be a game designer at a big company, rejected the product from start to finish. But after internal discussion, the team judged that the critique was still framed within traditional game evaluation and missed the core logic of AI-native games, so they didn't adopt it.

The answer had to be found on their own. Both believe the traditional industry tends to treat consumer games and creation tool platforms as separate things. MetaKid World's logic is "play first, create later": users open it and directly experience a complete, interactive AI world, modifying settings and creating characters at any time while playing, completing creations and sharing them. It is both a playable game and a UGC creation platform driven by the game experience.

Following this direction, Wang Kai and Chen Mingqiang arrived at an even more core reflection—an AI game must not only answer "what can AI do," but more importantly answer "why do players need it."

The answer ultimately came down to player experience. MetaKid World does not want to be a creation platform where users must first learn complex tools; instead, it lets users enter a world they can play first, and complete creation through play.

The current version of MetaKid World already delivers a fairly complete game experience. Players only need to input a single sentence or a simple setting—such as a story background, a character identity, or a thematic idea—and the system generates the corresponding world framework, including world rules, environment, character relationships, and main and side storylines.

Players can also advance the story through text dialogue and action choices. Unlike the relatively fixed NPC interactions of traditional games, AI characters have independent personalities, memories, and behavioral logic, giving different responses based on players' words and choices, keeping the story constantly evolving. The "MetaKid" agent acts as an AI companion, helping players organize characters, clues, and the current situation.

Beyond experiencing the world, players can also take part in creation—modifying world settings, creating characters and stories, and influencing NPCs' fates and story directions through their choices. After an adventure ends, players can record and export their experiences, or share the world they created with other players, letting others enter and experience their world.

Currently, the product centers on 2D plane exploration and text interaction, enhancing immersion through map exploration, character portraits, scene atmosphere, music, and other elements; richer interaction forms will be expanded in the future.

​Rebuilding classic games with AI, startup teams can move faster

Constrained by current technology and computing costs, MetaKid World is not yet pursuing high-fidelity 3D physical interaction; at this stage it uses 2D maps and text-and-image interaction as its main vehicle. Still, compared with static novels to read, MetaKid World builds a dynamic interactive narrative world. AI drives character memories and relationships, events keep evolving with player behavior, and stories even emerge spontaneously between NPCs.

On the technical path, MetaKid World also did not start by trying to build a complete AI big world directly, but chose to proceed in stages—a approach Wang Kai summarizes as "rebuilding games with AI."

First, the game's characters, maps, quests, events, and other content are structurally decomposed, breaking the originally rather complex game content into modules that can be managed and invoked, then linking these modules through a self-developed game engine so that AI can continuously generate content within.

On this basis, the first phase starts from text interaction, letting AI generate stories and verifying whether AI can truly drive a continuously developing game experience.

The second phase further visualizes the text content. AI breaks the generated story into structured content akin to a "script," which the game engine then parses, converting storylines originally presented as text into concrete game scenes, events, and interactions.

Compared with the first phase, this phase demands much more of the Agent. AI must not only generate stories, but also understand game structure and output executable content in structured form for the engine.

The third phase is exploring the AI big world. Ideally, AI would no longer just generate parts of a game, but participate in the operation of the entire game world, giving game mechanics, characters, quests, and events—formerly reliant on human design—dynamic generation capabilities.

According to the team, core engine modules such as characters, storylines, and structured generation have already been completed. Beyond the core AI big world experience, creation tools for characters, storylines, and quest editing have been built, preliminarily validating the feasibility of "letting AI generate playable interactive events."

That said, Chen Mingqiang admits the long-term goal is a complete AI big world; jumping straight from text adventure to a full AI big world at this stage is still too big a leap. The more pragmatic path right now is "Big World plus."

"Big World plus" does not mean directly building an infinitely open world, but first picking relatively well-defined game genres to break down—such as cultivation, JRPG, or action games—gradually adding AI capabilities within mature game frameworks, and then step by step moving from "game + AI" to games truly driven by AI.

In Chen Mingqiang's view, Roblox's community and content ecosystem align fairly well with MetaKid World's positioning, but the two paths differ. Roblox still centers on traditional gameplay; MetaKid World wants AI to go deeper into the game's底层—wait, no: wants AI to go deeper into the game's underlying layer, participating in character interactions, story development, and even combat systems, increasing the game's freedom.

Moreover, Roblox games are mostly built with Lua script programming, with AI only as an auxiliary tool; MetaKid World allows creation through natural language alone, with AI deeply driving the world's underlying operation.

This approach also distinguishes MetaKid World from AI companion products like Character.AI, Xingye, and MaoXiang—on top of the basic chat function, it has a worldview, maps, character growth, and more complete game mechanics.

Compared with the fixed flows of traditional interactive stories and the single-agent conversation model of AI companion products, MetaKid World also wants different agents to communicate and interact freely with each other. The product has added mechanisms for this—while the player is offline, the world keeps evolving autonomously; NPC interactions, characters, and events can continue developing. This is the core of what distinguishes MetaKid World from the AI companion products above.

The team also wants to avoid the framework constraints already formed by UGC platforms. Roblox, TapTap Maker, and others each have their own clear product forms and technical boundaries: the platform first defines a niche, and users create within the established framework. MetaKid World wants to enter from a more open space.

For MetaKid World, the top priority now is to solidify its base—text adventures and its own style of Big World plus—experiences that are relatively rare on the market and not well suited to being carried by traditional platforms. From there, it can gradually expand into more gameplay.

Chen Mingqiang believes MetaKid World may attract three types of users in the future: first, users who enjoy novels and web fiction, or text adventure games, who value story and imaginative space most; second, players who enjoy role-playing games, who care more about character relationships and immersion; and third, players with creative impulses amid the overall AI trend.

Wang Kai adds that as agents become widespread, users will grow increasingly accustomed to collaborating with AI directly through language, and "becoming a creator" will no longer be a high-barrier endeavor. "When making a game gets easier and easier, whatever kind of thing a user wants to play, as long as they can say it, they can play it right away."

For its business model, the team currently monetizes mainly through membership subscriptions and token consumption, and has reached partnerships with model providers such as Alibaba Cloud, Volcano Engine, DeepSeek, and MiniMax to obtain compute discounts.

In the short term, MetaKid World will keep advancing the product around "Big World plus"; if the data meets expectations, the coming year will focus on accumulating users and content, gradually proving out the AI 2D "Roblox" model.

And once users start creating content continuously, the platform will have the opportunity to further build a content distribution and community ecosystem. As for long-term exploration, when conditions mature, promising characters and worldviews could be incubated into IP. Various mini-games on the platform might also be spun off into standalone products.

"AI games are still quite new; many people probably just see AI as a production tool for games, something involved in the game-making process. But I think the ultimate AI game should truly be an AI-native game," Chen Mingqiang said. "At that point, AI would genuinely participate in design, naturally integrated into the game."

On this point, MetaKid World still wants to walk a longer road.

This article was republished with authorization from “Qingbao Ji”.

On September 11, 2026, the technology-themed salon of the Shanghai Online Game Industry Association, “AI in Game Production: Implementation and Efficiency Gains,” hosted by Yiwan (Shanghai) Network Technology Co., Ltd. (TapTap) and co-organized by Yystv, was successfully held in Shanghai.

Focused on two major directions — AI and game engines — the salon invited three industry experts: Chen Tuo, Software Engineer at Epic Games · Unreal Engine (Greater China); Zhao Tianyi, Product Lead of TapTap Zhizao; and Li Chao, Team Lead of Tencent Games' AI-native game engine team. They delivered themed talks on how AI can be integrated into game production workflows, drive development efficiency, and transform production paradigms, and a roundtable forum further explored the practice and future trends of AI in game R&D.

At the salon, the three guests delivered talks themed “From Text to World: Practical Integration of MCP with the Unreal Editor,” “TapTap Zhizao: Games in the AI Era,” and “AI-Native Engine: Reshaping the Game Production Paradigm and Opening an Era of Democratized Development,” respectively. Ranging from tool application and production workflows to engine capabilities, they shared practical experience and insights on how AI empowers game R&D, jointly exploring feasible paths for AI in game production to move from point-specific efficiency gains to full-process implementation, and providing game developers and industry practitioners with a platform for exchange, learning, and practical reference.

Chen Tuo, Software Engineer at Epic Games · Unreal Engine (Greater China), presented the integration practice of MCP (Model Context Protocol) with the Unreal Editor. Through MCP, large language models can directly understand and invoke the tools and capabilities of the Unreal Editor and, combined with PCG (Procedural Content Generation), participate in real production stages such as scene construction, asset retrieval, level layout, and code conversion — advancing AI from a “conversational assistant” toward an intelligent collaborator in game production.

Chen noted that the value of AI is not to simply replace developers, but to improve R&D efficiency and expand creative possibilities through deep integration with game engines, tools, and production workflows. In the future, as related tools and workflows continue to improve, AI is expected to become even more deeply integrated into the entire game R&D process.

Zhao Tianyi, Product Lead of TapTap Zhizao, shared how game development is changing in the AI era. He introduced that TapTap Zhizao seeks to weave natural language interaction throughout the game creation flow, allowing creators to start by describing an idea in one sentence, progressively generate gameplay, art, code, and audio/video content, and continually refine their work through real-time preview and local development. AI can also generate dedicated editing and debugging tools for specific games, further lowering the barrier from a creative idea to a playable version.

Zhao said that what TapTap Zhizao is exploring is not “generating a game in one go,” but rather kick-starting the “idea — playable version — player feedback — further iteration” loop as early as possible. Backed by the TapTap ecosystem, works can directly reach real players and keep iterating based on community and game data. The cases shown on site, such as “Dang Wo Lao Le,” “Bu Jingdian Lixue,” and “Guqin Dashi,” also demonstrated how AI helps creators turn different themes — social issues, scientific knowledge, ecological protection, and traditional culture — into concrete gameplay and interactive experiences more quickly.

Li Chao, Team Lead of Tencent Games' AI-native game engine team, shared the further changes AI is bringing to game production paradigms, focusing on the AI-native game engine MagicLomix. He argued that an AI-native game engine is not simply connecting large models to traditional development tools; rather, it enables humans and AI to collaborate continuously around the same editable project, allowing AI to understand code, scenes, and assets, invoke professional engine capabilities, and continuously verify and adjust based on actual runtime results — moving AI from content and code generation into real game engineering production workflows.

Building on this, Li Chao introduced MagicLomix's explorations in engine capability enhancement, the Agent Harness, and the developer ecosystem. By incorporating professional capabilities such as rendering, 3D content generation, asset processing, and performance diagnostics into a unified workflow, AI-generated content can undergo standardized processing, quality inspection, and engine validation, truly becoming assets usable for game development. Its core goal is to consolidate professional game development capabilities into capabilities that both developers and AI can invoke and reuse, so that creative ideas can be “made” while games truly “run.”

In the roundtable discussion and interactive Q&A in the second half of the salon, moderated by Chu Yunfan, founder of Yystv, the three guests answered questions collected in advance at registration. Drawing on their respective technical practices and industry observations, they exchanged views on how AI can be further integrated into existing game production pipelines, the core value of developers in the AI era, and the changes to team collaboration and job roles that AI-native engines may bring.

The discussion then extended to the future direction of AI and the game industry, including how far the industry still is from true “AI-native games,” what new game experiences AI may enable that were previously difficult to achieve, and whether aesthetics, expression, and creative choice will become scarcer as content production costs continue to fall. The session also explored whether AI can further narrow the R&D gap between independent teams and large companies, and questions were opened to the audience for exchange with the game developers and practitioners present.

Lin Zhimin, Deputy Secretary-General of the association, said in his address that this salon was the association's second practical salon following the intellectual property themed salon on August 25, focused on member companies' needs for technological upgrading. In the future, the association will continue to make sustained efforts in technical topic research, routine consulting, training and empowerment, and ecosystem building, establishing more high-quality communication platforms to safeguard the high-quality development of Shanghai's online game industry.

During the recent earnings season, Tencent, NetEase, G-bits… almost all listed game companies have mentioned AI in their annual reports. Meanwhile, Steam, the stronghold of independent games, has long since lifted restrictions on AI. By the end of 2025, games on Steam explicitly labeled as using generative AI technology had already surpassed 10,000, accounting for about 8% of the platform’s total game count.

In addition to AI-native games that build gameplay and systems around AI as the core, AI has also achieved comprehensive involvement in traditional game development.

Claude Code helps write code, Open Claw optimizes project management and workflow integration, DeepSeek assists with API calls, various video and image Agent tools are transforming game art production, and Suno directly generates game soundtracks… In the complex engineering of games, AI has appeared in almost every link.

Especially since 2026, in the eyes of many game developers, AI has crossed the threshold of “productivity level,” and the tools have become more democratized. This means that for most game industry professionals, AI is no longer a “toy,” but a true production tool.

AI is changing games. But because of the game industry's complex ecosystem, different types of games and game developers in different ecological niches accept and use AI differently. Compared to its “crushing” presence in the film and television industry, AI's penetration into the game industry is far more complex.

According to the “2026 State of the Game Industry Report” released by GDC (Game Developers Conference) in January this year, about 36% of game industry professionals have used AI in their work over the past two years.

However, among them, most use AI to help write code, optimize workflows, and so on; few directly generate assets and creative content, and the proportion building AI-native gameplay is even lower.

At the current stage, game projects and teams with more room for trial and error—such as well-funded big companies or student indie teams without baggage—are more willing to use AI. Meanwhile, veteran indie developers under survival pressure are generally more cautious.

At the same time, games come in many varieties. Just within the indie game space, epoch-making developments like animation generation technology have clearly benefited genres such as AVG and interactive cinematic games that rely heavily on film and animation content; some mature small-game gameplay mechanics have also seen their barriers to entry plummet thanks to AI's coding capabilities.

But for complex gameplay and 3D art assets, AI tools have not yet been democratized to the point of truly lowering the threshold for indie developers; and for works that sell on stylized art, AI is still far from being able to replace it.

Indie games are also more sensitive to the cost of using AI tools. After Sora shut down, a series of chain reactions occurred—for example, some AI tools cancelled student discounts or limited-time membership discounts, causing the cost of using AI tools for some indie studios to rise by 10%–30%.

All of this confirms that AI's impact on games is overall in a “transitional period” at this stage—AI's application scope is limited. However, game developers must at least be “aware” of AI; otherwise, it is hard to escape being left behind.

​Not all game developers,“dare” to embrace AI

“Before, when you used AI, out of 100 generations none could reach 89 points—all unusable. Now, if one out of 100 reaches 90 points, you're willing to use it”—this is the real feeling many game industry professionals have about AI tools today.

But progress at the tool level, when applied to different groups, presents completely different pictures.

Whether game developers use AI tools, how they use them, and why they use them follow different logics driven by differences in business models and organizational structures.

In big companies, AI use is being systematically quantified and assessed. For example, Claude has become an API that is “standard in big companies,” and in some places the API usage volume is even included in monthly evaluations—the more you use, the more dedicated you are considered. Tool use itself has become a KPI instead.

Behind this is the consistent demand of big companies: high process standardization, and everything quantifiable and manageable.

Tools like Claude Code, a Vibe Coding tool, have indeed lowered the barrier to program creation to a level anyone can try. But in our research, many game industry professionals mentioned that Claude's usage threshold is higher than it appears. Because the less the user knows about programming, the less able they are to judge whether the AI-generated code has hidden risks, and the less they know how to describe requirements. The result is that those who understand programming find it a powerful boost, while those who don't may spend several times more time, or even fail to complete the task.

In other words, AI has not leveled the playing field, but amplified the value of experience. For game developers at big companies, this is a more familiar logic—in a highly process-oriented system, experience has always been quantifiable capital.

But for indie game developers, cost pressure always hangs over their heads. It is harder for them to bear the uncertain risks of rashly investing in AI. Learning and using AI may both bring a considerable cost of trial and error.

On one hand, although it is widely understood that AI tools can reduce costs and increase efficiency, in many cases, the more senior an indie developer is, the less willing they may be to take time off from production to deeply learn how to use AI to reform their development process.

Most of these developers rely on the single capital recovery model of “complete game development, launch, and sales recoup costs,” making it hard to free up complete time and energy during the development cycle to systematically research AI tools. More critically, they worry that after deeply investing in AI and changing a development process that has already cost anywhere from tens of thousands to over a million, they might find that AI tools fall short of expectations.

Therefore, most people choose to experiment outside of their regular work, rather than directly embedding AI into the workflow of existing projects. For example, a producer talks with Gemini in their spare time to try to spark new inspiration; an art director tries Nano and Seedance to see if there is any stunning inspiration that can be applied to the next stage of development.

On the other hand, although AI tools are increasingly “democratized,” for many indie developers they remain an expense that is not very controllable and even often exceeds budget. With limited resources, it is difficult for them to gauge whether the upfront cost spent for “cost reduction and efficiency improvement” is ultimately worth it.

“How can there still be so many people willing to burn tokens running Open Claw? Haven't they seen next month's bill?” an indie developer complained to Game Artist.

He said that when not yet proficient and with full permissions enabled, one person's monthly Open Claw bill can reach 1,000–2,000 yuan. For a team of a certain size this is nothing, but for small teams and individual developers, it is real pressure.

And this is only the link with relatively low token consumption. Other areas such as AI-generated animation, when applied to indie games, cannot be completed and published quickly for quick return like motion comics; the cost pressure accumulated during development is hard to digest.

In addition, indie developers generally face the paradox of “late-mover advantage.” AI tools iterate too fast, making the timing of entry uncertain.

For example, when making in-game videos, when you spend time learning a certain tool and mastering unique tricks, you also increase sunk costs. Then Seedance appears overnight, with exponential technical improvement—you can't help but wonder: what was the point of the tokens burned on Sora before? If you had waited another year, could you achieve a better result at one-third or even one-fifth the cost?

For a game under development, these hypotheticals are meaningless, but they make both developers and investors more hesitant. Should we enter at this point? Is it better to wait? Because even a small indie game takes one or two years to develop; once a particular tool is integrated into a certain link, if you want to switch later, all the previous work may have to be scrapped.

​The most active and best-benefited users of AI,which indie developers are they?

“We ourselves don't have the ability to do many things, but we want to do those things,” Steven, producer at Sidi Games, said to Game Artist.

Steven's team is mainly composed of college students and is currently working on a project that includes an online system, first-person 3D, a large map, real-time danmaku feedback, camera depth-of-field effects, AI gameplay, and more. Stacked together, these modules are something a college student team would never dare to touch under traditional thinking.

But they did it. Relying on AI.

In the indie game landscape, apart from developers who aim for AI-native games and a few loyal supporters of AI games, the most enthusiastic adopters of AI are precisely those who seem “unqualified” to do complex projects—student teams, startup teams, and experimental teams who want to do something but don't know how.

Their projects are usually not large in scale, or they integrate AI at the early stage of setting up their pipeline, so sunk costs are low. Moreover, student teams have “student discounts” as a safety net—for example, students get one year of free Google One, can use Gemini and Antigravity, and can also use GitHub Copilot and Cursor for free.

Therefore, compared to senior developers under survival pressure, they have no historical baggage and more motivation to experiment.

In particular, international students are an important circle of creators in the indie game field. They come into contact with the overseas AI tool ecosystem earlier, and are more accustomed to treating technology with a “geek” mindset. They will try all the tools on the market and let each perform its own function in different scenarios. For example, besides Gemini, they also try Grok more.

In the research, some overseas student developers even believe that basic programming will become a foundational skill that everyone can master. Therefore, he gave up a programming design major and shifted to an interdisciplinary major like game design. “I clearly feel that in the future I can focus more on planning, and AI will help me realize it.”

Steven, on the other hand, said that his project investor comes from a technical background and often asks sharp low-level questions, which he cannot answer because it's all written by AI. But this does not affect the smooth operation of the game, because the team can clearly describe what effect they want.

“Art, programming, project management, and all other links are using AI. Without Claude Code, the labor cost of this project might have been one to two million; now it's only about one-tenth.”

From the research, small and medium indie teams assembling their own AI-based development pipeline from scratch can save an estimated 20%–40% on art costs and about 10%–30% on programming costs, depending on their proficiency with the tools and output demands. Programming in particular can save at least 30%–40% of time.

KATOR, founder of KATOR Game Studio, previously worked as a technical artist at NetEase and has a background in computer science and AI. Before the large model explosion at the end of 2022, he was already trying to integrate AI tools into his workflow. Currently, KATOR is leading his team in developing an indie game project.

In his view, AI's greatest help is that it can quickly produce things of about 60% quality. “Although AI tools cannot at all replace the design and thinking of key parts, they can quickly visualize and are very helpful in the process of communicating design ideas and directions within the team.”

For example, in environment production, they use AI to create models to build white boxes and check the atmosphere, and after the design is confirmed, they manually create detailed models; when making 2D characters, they can also use AI to color drafts to see the overall effect. This allows KATOR's studio, with fewer than 10 people, to work on projects that a normal company would need 20–30 people for.

He has also encountered the “late-mover paradox,” for example, when TapTap Maker was released not long ago, planners could generate game prototypes using natural language.

“If we had had this back then, the prototype iteration speed would have been faster, and early prototype design would not have required programmers, saving a lot of communication costs,” KATOR said. Although AI applications are developing rapidly, internal workflow changes should not be too frequent, otherwise everyone will waste a lot of time getting familiar with tools, and project development progress will actually be negatively affected.

Therefore, the emergence of AI tools actually greatly tests the producer's coordination and judgment abilities. Compared to game studios of a certain size, small teams have the advantage of “small boats turn easily”; if core members have strong learning and acceptance abilities, they may even enjoy the dividends of technological late-mover advantage.

Many indie developers mentioned that AI can effectively save costs from “previously meaningless waste”—those parts counted as “loss” and “redundancy” in project management expectations.

For example, a project planned for 18 months of development may spend at least 6 months on repeated trial and error, review, internal communication, external coordination, and so on. The advantages of AI tools in this regard have been verified multiple times and can be applied to mass-produced commercial games. This is also why some game companies have already incorporated AI tool API usage into workload assessments—the redundancy reduction caused by AI can be quantified.

​In indie games,what role is AI playing

After AI appeared, many content creation fields have seen the OPC (one-person company) model. Game industry professionals generally believe that indie game teams will become smaller and smaller in the future, with even a large number of one-person teams emerging.

In the current workflow, what role is AI playing?

Depending on usage, some developers feel it is like a “handy intern,” some think it's like outsourcing, and some consider it a highly cost-effective execution position.

Another indie producer described it as: “It is an ‘employee with no ambition but an extremely meticulous mind.’” In his view, AI can consider details that humans can never fully take into account, and the feedback can stimulate new creativity.

The common thread in these descriptions is that AI sits between a “tool person” and a “creative partner,” and is constantly approaching the latter.

With the emergence of Open Claw, this boundary has become blurrier. It is no longer a mere executor, but has the authority and ability to do things beyond the plan. But to say it can replace human creativity is still far from enough. AI is indeed fast at gathering information, but the power of judgment always remains in the hands of the operator.

Music and sound effects, art effects, copy and story, system gameplay—among the four important dimensions of a game, AI's level of involvement and creators' sensitivity are inversely related.

Suno writing songs is a typical case. There are many people from art, programming, and copywriting positions who come to make indie games, but it is relatively rare for musicians to personally make games. Most developers don't understand music theory and can only judge “does it sound good or not.” But the finished products made by Suno at least make these developers feel they can pass their own bar.

At the art level, from Sora to Seedance, ordinary creators may only see that “the effect is better,” but professionals can discern the precision of skeletal recognition. Ordinary creators will directly stitch AI-generated videos into the final product, while professionals only use them as references and then adjust on that basis.

This creates a gap in AI art capabilities among different game teams. Another difficulty in AI's involvement in the art pipeline is that indie teams need AI to assist in boosting productivity, but if an indie game wants to make a name for itself, art is often very important, and AI art still finds it difficult to produce distinct stylized artistic finished works.

Going further up, the relatively harder barriers to break are copywriting and gameplay.

Many developers said that using AI tools can indeed provide some ideas they hadn't thought of before, but when they want to break down copywriting and story work into predictable workflows like programming; and some AI engine developers even try to refine event logic at the screenwriting level so that AI can supplement narrative and plot to a certain extent—at that point they realize that what is produced is unsatisfactory and work efficiency becomes very low.

A visual novel game developer found that when he broke down the story to a level that AI could understand, he suddenly saw the light and could complete most of the work. He described this process as “looking in the mirror”—you have to have it yourself in order to reflect it in AI.

Moreover, in many game projects, copywriting is an extension of the creator collective. What AI produces is bound to have discrepancies, requiring frequent corrections, making it hard for creators to completely let go.

As for gameplay, its final implementation mainly relies on programming. Programming is the link most attempted and even integrated into workflows by developers because it is standardized and highly deterministic. But gameplay itself still requires creators to ponder and refine. AI saves ordinary creators the learning cost and evens the starting line as much as possible, but after the start, gameplay design becomes purely human brainstorming, and this is the area where AI is currently most helpless.

Overall, purely from the “creation” perspective, AI will greatly affect creators, but its impact on the works themselves is still limited.

​Conclusion:Before AI truly arrives

Ultimately, AI has not yet entered the “anthropomorphic” stage and cannot yet make reasonable, complete inferences, judgments, and decisions through complex situations and factors.

LLM, as the “eldest son” of generative AI, appeared first, representing the communication patterns summarized by humans from information left through language. But language models can be polluted and produce hallucinations. To some extent, this confirms the view of Turing Award winner Yann LeCun: LLM cannot give birth to the generative AI people expect; a “world model” based on actual perception and data can make more correct and reasonable judgments.

In addition, competition in underlying AI technologies and tools will also transmit pressure to developers of application tools. After Sora's exit, the cost of using AI tools for many developers rose by 10%–30%. Not just video generation, AI tools in other fields are also “taking advantage of the trend” to reduce or cancel membership discounts or student discounts.

For indie game developers, when LLMs and world models mature and stabilize to a certain extent and the price is acceptable, that will be the era when AI truly arrives.

In Steven's view, in the future only two positions will be needed to make games: game director and technical planner. The game director is responsible for considering what the player's final experience will be like. The role of the technical planner is to support the director in achieving the effect through reasonable means—to make all AI play harmoniously and transform the director's abstract concepts into concrete experiential emotions.

“These two positions can even be the same person. When that era arrives, you no longer need to be a hexagon warrior; you only need to be able to command and understand, at an abstract level, what players want.”

In fact, the indie developers who actively embrace AI value most using AI to replace the “inhumane” parts of game development.

Some indie developers put it more bluntly: “We make art, and what we like is the process of conveying emotions, not the process of grinding it out. AI has now replaced a lot of the ‘grinding out’ work; from this perspective, it is a democratization of human expression.”

Amid various uncertainties, perhaps this way of thinking can better help game developers use AI well: not “With AI, I can also make indie games now,” but “When I have a great indie game idea, AI can help me complete it as soon as possible.”

The essence of games is still about expression and experience; not putting the cart before the horse because of a technological revolution can also improve the success rate of making games.