AI in games has to earn the player's suspension of disbelief

The games industry's conversation about AI currently sounds like two panels happening in adjacent rooms.

In one room, platform vendors describe "living games": agents running characters, economies, and stories that react to each player. Google Cloud's GDC pitch moves from automated playtesting and code generation all the way to worlds that continuously adapt. One featured game lets players describe a hero and generates a personalized version inside the studio's authored style.

In the other room are the people making the games. The 2026 GDC industry survey found 52% of respondents believe generative AI is having a negative impact on the industry, up from 30% last year and 18% the year before. Only 7% called the impact positive. Artists, narrative designers, game designers, and programmers were among the most negative groups.

These are not contradictory results. They are reactions to two very different uses of AI that keep being sold as one revolution.

AI behind the curtain

Games are giant state machines held together by art, scripts, probability, and an heroic amount of QA. There are obvious places for AI to help without becoming the creative center of the product:

Playtesting. Agents can run routes for thousands of hours, probe economies, reproduce edge cases, and identify where players get stuck. This is not a substitute for human taste; it gives human testers a better map of where taste and judgment are needed.

Production search. Finding the animation, dialogue condition, shader, or configuration that controls a bug is exactly the kind of bounded code-and-tool problem agents are getting good at.

Prototyping. Temporary dialogue, greybox textures, and disposable quest variations can make an idea testable before a team commits expensive production work.

Accessibility and localization QA. Models can flag overflowing text, missing descriptions, inconsistent terminology, or control prompts that do not match the active device.

None of these belong on the back of the box. That is a feature.

AI on the stage

Player-facing generation has a harder burden. Games already know how to produce more content than anyone can consume. Procedural generation is decades old. The scarce resource is not variation; it is meaning.

A handcrafted NPC line works because a writer knows what it foreshadows, what it reveals, how long the player has been away, and what emotional beat comes next. An infinite conversation can be locally convincing while making the whole world less coherent. The character says more and means less.

The same is true of generated quests and art. Novelty feels magical for an hour. Then players discover the shape of the generator, just as they learn the seams of any procedural system. If the output does not interact with authored mechanics and consequences, "anything can happen" becomes "nothing matters."

The promising versions put generation inside strong rules: a defined visual language, a bounded narrative role, an economy that cannot be talked into collapse, and authored consequences that survive after the model call ends. The model improvises; the game still directs.

Why the hostility is rational

Developers are being asked to evaluate this technology during layoffs, studio closures, and relentless pressure to make larger games with smaller teams. When an executive says AI will "unlock creativity," workers reasonably hear a second sentence about headcount that was left off the slide.

Trust will not come from another demo. It comes from decisions:

• Train on material the studio has the right to use.
• Credit and compensate the people whose work shapes the system.
• Tell players where generated content appears.
• Give creative teams veto power over uses that damage the work.
• Measure whether AI removes drudgery or merely increases the amount of content everyone must review.

My rule is simple: AI should let a team attempt a more interesting game, not just ship the same game with fewer people.

Games are machines for sustaining belief. Any technology that makes the world cheaper but less intentional is a bad trade, no matter how impressive the generation demo looked at GDC.