The State of AI in Game Development

Artificial intelligence represents the most transformative technology to hit game development since the internet. While its adoption has sparked controversy, AI’s promise to help teams do more with less makes its integration inevitable. Beyond simply enhancing current development processes, AI promises to unlock entirely new types of game experiences and player interactions. However, many developers remain cautious about publicly discussing their AI usage, particularly on PC, where Steam maintains strict disclosure policies, while mobile platforms present fewer restrictions but similar hesitancy.

Many studios aren’t public about their exact usage of AI, so we can’t see everything. But let’s take a look at how we do see AI impacting game development across three key phases.

Pre-Production

Pre-production focuses on game design, production planning, budgeting, developing mock-ups/prototypes, and securing investment.

Market Research & Validation

Large language models (LLMs) have the potential to be extremely helpful in providing a brainstorming assistant available 24/7 for design exploration and ideation. AI can help streamline the typically tedious process of market research and validation, rapidly analyzing trends, competition, and genre performance. Of course, AI has limitations. LLMs can only provide surface-level information since they can’t access paywalled/private information and data. They also don’t engage with games. (This is where Naavik still comes in handy.)

• The tools include:
  • Deep research features in ChatGPT, Gemini, Claude, Perplexity, and Grok.
  • Ludo.ai, a specialized gaming platform offering AI assistance for brainstorming, market research, documentation, asset generation, and prototyping. Notable users include Ubisoft, Voodoo, Say Games, HOMA, Garena, and Unity.

Generating and Updating Design Docs

Good game design documents (GDDs) are essential in the early stages, but they are time-consuming. LLMs paired with GDD templates can accelerate initial documentation and highlight important questions early in development. They also simplify document maintenance during production, keeping documentation current and useful for future AI-assisted tasks.

As an example of using AI for game design document generation, Sergey Zaigraev, the lead game designer at Crazy Panda, uses it for stream-of-thought processing of ideas and discussion into an organized document. He also uses AI to gather feedback and suggestions from the document to improve things.

• The tools include:
  • Ludo.ai, which provides quick GDD generation from concepts and integrates with the market research element.
  • Wayline’s Blueprint for GDD generation.
  • Using an LLM with free online GDD templates.

Rapid Prototyping

Early validation of game concepts and mechanics can prevent years of wasted development. While experienced developers can quickly create prototypes, many game designers lack coding skills and don’t do as much prototyping as they may want or need to. The ability for AI to generate code can enable rapid prototype creation and iteration, accelerating the process of “finding the fun” without excess development overhead.

The process of describing mechanics precisely enough for AI to create prototypes also helps designers refine their concepts. Well-written GDDs can help provide additional context for these AI prototyping tools.

There are two approaches to using AI for coding, one for experienced programmers and the other for complete novices. Experienced programmers can use LLMs like Claude or dedicated coding tools like Cursor to help more efficiently write boilerplate code that takes less contextual understanding and project knowledge.

Novice programmers have instead picked up “vibe coding,” where all trust is placed in the AI to completely handle everything, including its own errors. Both approaches can be used to speed up prototyping, but vibe coding will often only work for simpler prototypes with current limitations.

Beyond simple prototypes, AI coding can also help develop vertical slices for demonstration to others working on the game or potential investors. There is a risk here of misrepresenting the ability to execute quickly, but, in general, the vertical slice would be of lower quality than the team could eventually execute on. In theory, AI can also help with coding during production, but given the current capabilities, it’s best to limit the use to prototypes and vertical slices. However, AI’s capabilities are improving rapidly, so far more will be possible in the months and years to come.

While most game developers haven’t openly discussed using AI for internal prototyping on the coding side, Rovio is all in on using it for prototype art, which it also apologized for, but hasn’t stopped using it. Developers in general should be leveraging AI as much as possible in the prototype stages to “find the fun” and validate ideas sooner. Hypercasual mobile game developers like Homa have advocated for the benefits.

• The tools include:
  • AI code prototyping and sharing platforms for web-based games like Websim and Rosebud.
  • Dedicated AI coding tools like Cursor, Windsurf, and Github Copilot.
  • LLMs like Claude and Gemini with MCPs (essentially APIs for AI) for “vibe coding.“
  • Unity-specific tools that allow creating/modifying content using text prompts like  vibe-gamedev and Coplay.

Key Takeaway

AI excels at accelerating traditionally time-consuming early processes by enhancing, not replacing, creativity. With over 63% of surveyed developers in 2024 already using AI for design and storyboarding, it enables seamless workflows from market research through documentation to prototype generation. The technology enables more consistent and thorough validation while reducing risk.

In the long run, emerging innovations like vibe coding may somewhat disrupt classic game development through lower barriers of entry and potentially by unlocking or accelerating new platforms. This comprehensive approach helps build investor and developer confidence through better planning and validation processes, especially as this eventually gets worked into planning and budgeting.

Production

Production forms the core of game development, often spanning several years as teams create code and assets. Most AI production tools and techniques also apply to post-production live service content creation as well.

Visual Assets(2D and 3D)

Activision infamously began using AI images for Call of Duty: Black Ops 6 as part of its live ops; it was caught with a six-fingered zombie image. Activision eventually admitted to some AI art images, but only after the controversy died down. They continue to use it for live ops cost/time saving, and simply added the required AI disclosure to the Steam page.

High on Life developer Squanch Games fully embraced the weirdness of Midjourney images by using them for poster art. Despite the lack of AI disclosure on the store, there seemed to be no real backlash against the developer as the integration made conceptual sense.

InnoGames (Forge of Empires, Elvenar, and more) integrated Scenario’s AI into their art workflow to produce in-game assets faster. By training custom AI models on its art style, InnoGames auto-generated content like character portraits, item icons, and background art. According to InnoGames, using AI art tools cut the time to create new assets by up to 50%, while keeping artists in control of final edits.

Project Kara from Keywords Studios is an interesting example of using AI in multiple areas, and especially 3D assets, to try and remaster an older game. In this case, Keywords Studios has used AI to help remaster an older 2021 Apple Arcade game, Detonation Racing. The purpose is just as much to test pipeline integration of modern AI as it is an attempt to actually remaster the game.

• The tools include:
  • Midjourney, Stable Diffusion, Gemini, ChatGPT.
  • Specialized: Scenario (2D), Meshy (3D).
  • Tencent’s continuously improving open source 3D asset generation models.

Audio Assets(SFX and Voice)

Voice acting remains a contentious area for AI, as highlighted by recently settled SAG-AFTRA negotiations. Despite the controversy, AI voices have entered public use in both pre-generated and real-time applications. While currently more expensive than text generation, voice AI costs continue to fall.

The Finals used text-to-speech (TTS) voice generation for the majority of its voice acting, especially for the commentators. High on Life, the aforementioned game which prominently features voice actor and creator Justin Roiland, also used AI voice acting primarily for the prototyping.

With games starting to heavily explore the space of LLM-based NPC dialogue, using AI to also generate the dynamic voices has increased in interest. Fortnite recently used an AI voice trained on the real voice of the late actor James Earl Jones, with permission, to voice its Darth Vader NPC integration.

• The tools include:
  • ElevenLabs.
  • Eden AI.
  • Murf AI.
  • OpenAI TTS.

Level/World Design

While AI may not excel at creative design, it can significantly accelerate initial level population and content placement. It particularly shines in procedural generation for puzzle and casual games, handling complex rule sets that traditional algorithms struggle with. King is doing exactly that: A number of staff members helped train its AI level-design tools, who were then fired, and the AI tools then took over. This is likely to be happening across a number of mobile game studios due to the more systemic level design that many casual mobile games have.

On the frontier side of things, AI “World Models”, which are trained to generate virtual worlds, are slowly developing abilities to simulate game experiences without the development of code or assets. They aren’t ready or cost-effective for real usage yet, but they may unlock new forms of play experiences in the coming years.

• The tools include:
  • Coplay, which integrates with Unity for level population and basic code functionality.
  • Promethean AI, which manages asset collections and world population.
  • World models from OpenAI, Microsoft, and Google.

Narrative & Dialogue

LLMs excel at generating extensive dialogue and narrative content, though they haven’t yet matched human writers’ quality. Recent models like Kimi K2 show significant improvements in creative writing capability. AI can help generate ideas, drafts, or help fill large dialogue needs in massive games. And humans can still help finalize the end product.

As an example of it helping with the scale problem, Ubisoft showed off using AI to help with bulk writing efforts with Ghostwriter. This tool helps write the massive variations of “barks,” which are frequent, short NPC reaction lines, with human involvement after generation.

Many games like Millennium Whisper and InZoi are experimenting with using AI to drive real-time NPC dialogue or behavior and allow for open-ended dynamic responses to players. It’s far too expensive to utilize LLM cloud services. Instead, these kinds of games are baking in open source, locally run smaller models that run on a player’s computer. This is a use case we expect to see expand a lot, but the current implementations leave a lot to be desired in terms of response times and system requirements. Bigger picture, dynamic AI NPCs that operate across game worlds can unlock new forms of engagement and fun, emergent properties.

• The tools include:
  • LLMs with good creative writing output.
  • Locally run small LLM models like Gemma.cpp, Llama, Nvidia Nemotron, and Mistral.
  • AI NPC platforms like Inworld AI.

Development Platforms

Beyond specific asset creation, major engines and platforms like Unity, Roblox, Fortnite UEFN, and Meta Horizon are starting to offer more comprehensive AI integration. These engines and UGC-focused platforms combine asset generation with code assistance, using platform-specific training data to supplement developer expertise. While these systems are still emerging, platforms are racing to capture market share, particularly among younger developers.

Key Takeaway

The industry’s use of AI as a force multiplier for labor-intensive tasks, while maintaining human creative direction, offers a potential solution to persistent challenges of high costs and extended development cycles. This technological shift may prove particularly advantageous for smaller development teams that can adapt more nimbly to AI integration compared to AAA studios constrained by large workforces and more scrutinizing player bases.

Post-Production

Post-launch support encompasses updates, fixes, balance changes, DLC/IAP, events, and ongoing maintenance.

QA Testing

Given many games still release full of bugs, there’s an opportunity for AI to play an outsized role in quality assurance (QA). There are a variety of approaches for using AI, such as simulating AI players to identify bugs and observing game play to catch bugs. More complex games with lots of moving parts and dynamic situations have the most potential to benefit, as testing all the various combinations and contexts can be near impossible to stay on top of manually. See our podcast episode on the topic from last year as well.

• The tools include:
  • modl.ai.
  • Razor’s AI QA Copilot.
  • Ixie.

Translation

Multilanguage translation for localization is important to maximize any game’s geographic coverage and increase its audience size. Unfortunately, it’s also time consuming and error prone, both when done manually and when using AI. AI-driven translation can be useful for first passes, with human, manual corrections. But improvements over time will reduce manual evaluation needs. It’s important not to use this internally to rush translation just to have more languages available on launch, a mistake made by The Alters that publicly backfired.

• The tools include:
  • LLMs that have the desired language support.
  • Services from AlocAI, Andovar Gaming, and KantanAI.

Player Sentiment

While providing any form of live service, or even just patches, it’s important to understand player feedback and sentiment. Between reviews and social media, there is far too much chatter to humanly consume while having any time to act on it, so AI-powered sentiment analysis has shown up to help. It won’t be perfect when it comes to some elements of satire and context, but it can be a massive time saver when used at a higher level and then diving in when needed for more detail. AI can better translate different kinds of sentiment into metrics that can be far better than basic methods like keyword counting. These concepts also can apply to many forms of in-game sentiment, including in-game chat and behavioral patterns for increased in-game visibility beyond KPIs.

• The tools include:
  • GGWP for sentiment analysis.
  • Levellr for Discord insights.
  • Gaming Charts for Steam review analysis.

User Acquisition

Developing and testing the massive numbers of creatives needed to compete in mobile user acquisition (UA) simply puts additional strain on an already-hyper competitive industry. The constant, rapid fire testing of different trends, memes, and unique ideas is far more manageable with an AI-generated pipeline than humans can even begin to compete with, especially with such a low quality bar and the cost of AI video. Usage in mobile game UA is already exploding, and anyone still making most of their creatives manually will absolutely fall behind.

• The tools include:
  • Creative generators like AdCreative.ai, Reforged Labs, and Gamelight.
  • AI video creation tools like Runway and Veo.

Key Takeaway

The game-as-service model is starting to accelerate AI adoption across the industry. While outsourced work and user acquisition are likely to see the earliest transformation, a significant impact lies in post-production support, where AI tools can help studios optimize operations and scale their responses to player needs while managing costs. As AI tools and services continue to evolve, their integration into game development workflows promises to reshape how studios approach both development and live operations.

Conclusion

AI is rapidly transforming game development, and studios not actively exploring these technologies risk falling behind. While we’re still in the early stages, teams across the industry are already discovering valuable applications that enhance existing workflows.

The adoption landscape varies significantly as expectations shift. Mobile gaming, driven by optimization needs and less AI-sensitive audiences, is seeing aggressive implementation. Meanwhile, PC and console development requires more nuanced approaches due to their hardcore player base’s heightened sensitivity to AI usage.

While some AI hype may be overblown, the technology offers crucial solutions for addressing skyrocketing production costs and development risks. Small, agile teams are particularly well positioned to benefit by leveraging AI to compete more effectively with larger studios. These tools, when used to augment rather than replace creative talent, are already showing promising results as evidenced by over 8,000 games on Steam disclosing AI usage. However, larger publishers often face more organizational friction in adoption, potentially creating a growing capability gap between nimble independents and established studios.

We also expect increasing adoption of the tools as developers gain proficiency, find more consistent use cases, and overall cost vs quality improves. Best of all, improvements in this and related technology, including world models, will help unlock new approaches to game development and distribution alongside unexpected new types of player experiences. Every game studio today should be experimenting with low-risk, high-impact areas for AI use like prototyping, building up expertise through pilot projects, and establishing usage guidelines that align with audience expectations.

Best of all, improvements in this and related technology, including world models, will help unlock new approaches to game development and distribution alongside unexpected new types of player experiences. While there are plenty of ways to view AI as a sustaining innovation that unlocks efficiencies, this shouldn’t blind us from how AI — like all tech waves before it — enabled entirely new forms of gameplay, growing the industry in the process. Ultimately, the largest opportunities are what we can’t exactly see yet.

Source : Devin Becker/Naavik, Original link:  https://naavik.co/digest/the-state-of-genai-in-game-development/