Larian and Microsoft at the center of the dispute: how neural networks are changing game development

AI use in game development has become one of the most contentious issues in the industry. Against the backdrop of scandals surrounding The Alters, Larian, and Expedition 33, some players are demanding a complete rejection of neural networks, but the studios themselves consider them a working tool, not a replacement for people.

Where the dispute came from

The new wave of criticism was triggered by quite specific cases. In The Alters, players found traces of unedited AI content: placeholder strings in dialogue and textures that looked like machine-generated drafts. The developers admitted that such elements indeed made it into the release by mistake, and neural networks were used, among other things, for localizing new content at later stages.

Later, Larian came under fire, and employees of Obsidian, a studio within Microsoft's ecosystem, publicly spoke out against AI. Against this backdrop, journalists and part of the gaming community compiled the usual list of grievances. The main fear is that neural networks will transform from an auxiliary tool into a mechanism for cutting costs at any price.

This anxiety is compounded by mass layoffs in the industry: according to the data cited, from 2022 to mid-2025, roughly 45,000 people were cut from game development. This is why any hidden use of AI is now perceived as a sign of deception.

• Risk of new layoffs for artists, translators, and junior developers
• Fear of conveyor-belt content and declining originality
• Scandals around AI voice acting, textures, and draft assets
• Player distrust of studios that don't explain how they use AI

Where AI already helps

Removing emotion from the equation, AI's strongest positions right now are in areas with lots of routine and repetitive operations. Localization comes first. For small teams and especially for Japanese studios, which often release English versions with a delay, such tools significantly reduce the time between releases.

But this isn't about pressing a "translate everything" button: even a good machine-generated draft still requires editing, cultural adaptation, and final human review. The same logic applies to the visual side. Neural networks can indeed handle low-level tasks: quickly sketching out options, preparing draft assets, helping with sound, skeletal animation, or test materials.

This doesn't mean that a living artist is no longer needed. Rather the opposite: the stronger the tool at automating routine work, the more obvious the value becomes of a person who can select, edit, and assemble a cohesive style from templates. In essence, AI here is closer to procedural generation than to an independent creator.

Why studios won't refuse

The main argument for AI supporters in game development hinges not on hype, but on the economics of production. AAA development gets more expensive every year, development cycles grow longer, and player expectations rise higher. If such tools were completely removed from the pipeline, major games would come out even more slowly.

The author directly links this to the risk for the entire industry: the longer and more expensive production becomes, the less willing studios are to experiment and the more carefully they select which projects are even worth launching. There's also a more practical level of utility — prototyping. The article gives an example where the Qwen model assembled a game prototype in an hour, when a human would have spent a week on a comparable amount of work.

Even if such a result falls short of final quality, it dramatically accelerates idea validation. For a team, this means less time on rough assembly and more time on design, balance, and polish. That's why for many studios the question no longer sounds like "do we need AI at all," but rather "where does the line between useful automation and shoddy work lie."

What this means

The dispute around AI in games will likely not disappear, but the technology itself has already taken hold in production as a way to speed up localization, prototypes, and some routine tasks. For the industry, the key question now is not a complete ban, but transparent rules for use: where neural networks help the team, and where they start to substitute quality with cheap surrogates. The winners won't be those who reject AI loudest, but those who learn to use it carefully and honestly.