By Olga Taranova, online tech lead at Thrylox

On August 18, 2026, the small Korean studio ALIVE released Vapor World: Over the Mind in Early Access, a 2D Soulslike platformer that had accumulated more than 100,000 wishlists before launch.

After release, the discussion quickly shifted from the game itself to its use of generative AI. The developers disclosed that they had used AI for voice acting and cutscenes, while the share of positive reviews shortly after launch stood at around 25%. Many negative reviews explicitly mentioned the AI-generated content.

Director Young Kim explained the decision by pointing to the studio’s limited resources: the team had hoped that AI would make the cutscenes more expressive while allowing them to devote more resources to the game itself. After the negative reaction, ALIVE abandoned the generated cutscenes and replaced them with versions the team created.

This raises a broader question: which parts of game development are appropriate for AI, from both developers’ and players’ perspectives?

AI in Small Studios

Small studios operate with limited numbers of programmers, artists, QA specialists, and game designers. The appeal of AI is therefore understandable: while it does not necessarily replace specialists, it can reduce the technical and routine work they have to do.

According to the GDC 2026 State of the Game Industry report, 36% of surveyed game industry professionals use generative AI. They most commonly use it for research and brainstorming (81%), everyday work tasks and coding assistance (47% each), and prototyping (35%).

Player-Facing AI Content Carries Greater Risks

When AI is used for dialogue, voice acting, art, or cutscenes, players evaluate not only the quality of the result but also the very fact that AI was used. As a result, even technically acceptable content can provoke a negative reaction.

The example of Vapor World shows that even having a team review AI-generated content does not necessarily protect a studio from player backlash. The studio said that all AI-generated content had been reviewed by the team and was included in the game only in pre-rendered form. Nevertheless, this content became a major source of criticism.

Skepticism is also widespread within the industry itself. According to GDC 2026, 52% of surveyed professionals viewed AI’s impact on the game industry negatively.

AI in Internal Tools

Another approach is to use AI not to create content for players, but to build tools that speed up development. I faced this problem at Thrylox, where part of our workflow involved transferring data from Excel into game assets inside Unreal Engine. Doing this manually took a lot of time, so we needed a dedicated internal tool. We built one with Claude’s help, using AI to support implementation while keeping the workflow and final checks within the development team.

For me, this is where AI is much easier to justify. Developers use the output, so they can inspect it and fix problems before anything reaches the player. AI still saves development time, but mistakes remain inside the production process instead of becoming part of the game itself.

What Matters Most Is Who the End User of the AI Output Is

Based on this experience, I think one of the most useful criteria is who the end user of the AI output is. With internal tools, the team itself works with the AI output, so errors or poor-quality results remain a problem for the developers rather than the players. Suitable tasks include working with internal tools and data, automating repetitive operations, assisting with coding and testing, finding errors, and prototyping.

It is therefore more useful to distinguish AI use cases not simply as creative or technical, but by whether the result stays within the team’s workflow or becomes part of the user experience.

Conclusion

The story of Vapor World shows that the important question is not only whether to use AI, but also where it should fit into the development process. Small studios genuinely need tools that can save time and resources. But AI is best suited for internal tasks, where it can speed up the team’s work without becoming a source of final, player-facing content. The more directly players encounter AI output, the more their attitude toward the technology itself matters.