It's 2026, Where Are the AI NPCs? A Contrarian Read on the Unit Economics

We wrote a piece not long ago about the adoption numbers: AI-powered NPCs went from roughly 8% of new RPGs to well over 60% in the space of two years. That's a real trend, and the number is accurate. But a fair question sits underneath it, one worth taking seriously rather than waving away: if AI NPCs are genuinely everywhere now, name the last one you played a game specifically because of.

Most people can't, not really. And that gap, between adoption numbers climbing and cultural impact staying flat, deserves a proper look rather than a shrug.

The clearest explanation isn't taste, or the tech not being ready, or players being resistant to change. It's cost, and specifically the direction the cost runs in. With a traditional NPC, the expensive part happens once: a writer scripts the dialogue, a voice actor records it, and every subsequent player who talks to that character costs the studio nothing extra. With an AI-driven NPC, the meter starts running the moment a player opens their mouth. Every exchange is an inference call, and inference isn't free. The more a player enjoys talking to your character, the more it costs you to let them keep doing it.

That's a strange incentive to build a creative product around. Traditional game design rewards depth, the more a player wants to explore a system, the better the game is doing its job. AI-driven dialogue at scale quietly punishes exactly that, because depth of engagement here is a cost centre, not just a design win. A studio can end up in the odd position of hoping players don't get too invested in the NPC they spent months building.

None of this means the technology doesn't work, current systems can hold a coherent, responsive conversation better than they could two years ago, and that's genuine progress. But working technically and working as a sustainable design choice are two different bars, and a lot of the coverage this year has quietly conflated them. Plenty of the 60%+ of new RPGs shipping "AI-powered NPCs" are running the technology in narrow, budget-capped ways, a handful of lines, a limited conversation window, rather than the open-ended, go-anywhere dialogue the marketing around this space tends to promise. That's not a failure of ambition. It's a rational response to an economic reality that doesn't get discussed nearly as often as the adoption stat does.

We'd rather sit with that tension than smooth it over. The interesting design work happening in this space right now isn't "how do we make AI NPCs talk more," it's "where does open-ended AI dialogue actually earn its cost, and where is a well-written, fixed system simply the better tool." That's a harder, less flattering question than a growth chart, and it's exactly the kind of thing we want to dig into on stream rather than skip past.