Why generative AI NPCs are overhyped

And game developers may be the worst target customer.

The most overhyped thing in games right now is generative AI-powered NPCs.

To be more specific, I don't think anyone is going to build a venture scale company providing GenAI NPCs as a service to game developers.

Don't get me wrong - I do think there is a role for large language models in games, and that in the nearish future we will see fewer games that feature NPCs repeating the same 3 pre-recorded barks over and over again.

💡 The key issue: GenAI NPCs are either too important or too trivial a feature for developers to rely on a third party service. Let me explain.

📉 Too trivial: The demos we have seen so far show GenAI NPCs talking to the player, with no meaningful impact on gameplay systems. This is perhaps a minor upgrade from regular NPCs, if a bit uncanny valley right now. But if they don't really impact gameplay, GenAI NPCs are not a feature that most developers will find hugely valuable, especially if they have to pay-per-use for access via an API.

📈 Too important: If GenAI NPCs are absolutely core to the game and integrated deeply into its systems, I think developers will want to own the tech themselves so that they can control and heavily customize it for the game they are building. More importantly, they won't want to leave a core element of their game at the mercy of a third party company's roadmap (especially a startup).

As a side note, I also think there is a good chance that open-source versions will emerge that can be incorporated into games directly and run inference locally, as we are already seeing with broad LLMs.

For these reasons, I suspect GenAI NPCs as-a-service for games are a fools errand.

Instead of allocating resources there, I think companies in this space are better off exploring other verticals and trying to find product-market fit where the technology is more easily generalizable and potentially more valuable.