Here's a thing that surprises people: the enemies in Pac-Man each had a distinct "AI" personality. Blinky chased you. Pinky tried to ambush. Inky was moody and unpredictable. Clyde wandered off on his own tangents. That was 1980.
Games have been quietly using AI since before most of us were born. The recent generative-AI wave is the first chapter in that history where the AI also writes the script, designs the levels, and chats with the player — but the craft of making it feel good is continuous with forty years of thinking.
Let me talk about three places where AI in games is getting genuinely exciting, and one place where it's making me nervous.
1. Non-player characters that remember you
The classic RPG companion has always had a problem: the second time you see them, they greet you with the same thirty lines of canned dialogue. That wall is coming down. Modern language models, grounded in a game's own lore, can produce companions that remember what you said last session, react to the choices you made, and develop genuine-feeling personalities over time.
The best implementations I've seen recently use a hybrid: the authored dialogue (written by human writers, voiced by human actors) still does the heavy lifting for the story. The AI fills the gaps — the small talk on the road, the reactions to unexpected player choices, the filler that used to sound canned. It's an unglamorous use of the tech, and it works.
2. Procedural content that's actually good
Procedural generation is old. But most procedurally generated content used to have a telltale sameness — an endless corridor of variations on the same theme, technically infinite but experientially boring.
Modern diffusion models and LLMs are changing this. Terrain that varies genuinely. Quests whose twists feel authored. Dungeons whose layouts have narrative logic, not just spatial logic. The Dwarf Fortress generation of procedural content was brilliant-but-inscrutable; the new generation is brilliant-and-legible.
The key insight here: the AI isn't replacing level designers. It's giving them faster iteration. "Generate me twenty variations of this shrine, following this tone" used to take a week. Now it takes an afternoon, and the designer picks the one that feels right.
3. Accessibility that isn't an afterthought
This one I'm proud of. Real-time voice cloning means players with speech difficulties can participate in voice chat without their difference being the first thing anyone hears. AI-driven live captions make competitive games playable for Deaf players. Generative assistants help players with cognitive disabilities understand complex mechanics at their own pace.
None of this replaces dedicated accessibility design — the excellent work done by studios like Xbox Game Studios or Naughty Dog is still the gold standard — but AI is democratising it. Small studios can now ship accessibility features that used to be available only to AAA budgets.
What makes me nervous
The commodification of creative work. Generative AI is being used, in some studios, to cut the number of writers, concept artists, and voice actors they hire. That's not just an economic concern — it's a craft concern. Games are a genuinely new art form, still only fifty years old, and the humans who've been inventing its language deserve to keep making a living.
I think the good version of AI in games involves more work for human creatives, not less — because AI tools let each creative be more prolific, and the industry grows accordingly. But that's a choice the industry is making right now. It can go the other way.
Player-hostile personalisation. AI can be used to tune difficulty and rewards to individual players. Most of the time this is lovely — it means the game meets you where you are. But AI can also be used to maximise engagement and monetisation in ways that border on predatory. A slot machine tuned to each individual gambler. That's a road nobody wants to be on, and regulators are starting to notice.
What I'd tell a studio considering AI
A few things we've been saying:
- Use AI to amplify your creatives, not replace them. The best games have strong authorial voices. AI is a brush, not a painter.
- Playtest the AI-generated content with real players. It'll fail in interesting ways. Factor that into the schedule.
- Be transparent. If AI is involved in voice, dialogue, or art, say so. Players are increasingly alert to this, and trust matters.
- Keep the humans in the credits. If an artist's work was used to train a model that later generated assets for your game, name them.
Games, more than most media, have always been about play and joy. It'd be a shame to use the most exciting AI wave in history to make games that feel more like spreadsheets. If you're working on something in this space, we'd love to chat.
