Yesterday, 05:42 AM
Two years ago the discourse was "AI will replace game devs in 3 years." A year ago it was "AI in games is all scams and asset flips." Today the picture feels messier and more *real*, which is worth talking about.
What I'm actually seeing used in production:
- **NPC dialogue & barks** — several studios using LLMs (fine-tuned, governed) for ambient lines and throwaway NPCs, not main story. Ubisoft's NEO NPC experiments and a few indies shipping with it.
- **Voice synthesis** — used for prototyping *and* increasingly for shipping content, especially in non-English locales where voice budgets are tiny.
- **Concept art / reference / mood boards** — basically universal at this point; artists are using it like a better Google Images + sketchpad.
- **QA and playtest analysis** — clustering player deaths, finding stuck spots, summarizing playtest notes. Quietly the most "actually works" use case.
- **Anti-cheat** — behavioral ML, same as before but better.
- **What's still not really shipping:** fully generative game worlds, AI-written main plots, AI that reliably replaces animation or character modeling.
The tension I keep seeing: players *hate* the idea of AI-made content in principle, but don't notice it when it's used well on ambient stuff. Meanwhile devs quietly depend on it to ship on time.
What's your take? Would you feel cheated if you found out ambient NPC dialogue in a game you loved was LLM-assisted? Or is it just the next tool, like photogrammetry or motion capture were?
What I'm actually seeing used in production:
- **NPC dialogue & barks** — several studios using LLMs (fine-tuned, governed) for ambient lines and throwaway NPCs, not main story. Ubisoft's NEO NPC experiments and a few indies shipping with it.
- **Voice synthesis** — used for prototyping *and* increasingly for shipping content, especially in non-English locales where voice budgets are tiny.
- **Concept art / reference / mood boards** — basically universal at this point; artists are using it like a better Google Images + sketchpad.
- **QA and playtest analysis** — clustering player deaths, finding stuck spots, summarizing playtest notes. Quietly the most "actually works" use case.
- **Anti-cheat** — behavioral ML, same as before but better.
- **What's still not really shipping:** fully generative game worlds, AI-written main plots, AI that reliably replaces animation or character modeling.
The tension I keep seeing: players *hate* the idea of AI-made content in principle, but don't notice it when it's used well on ambient stuff. Meanwhile devs quietly depend on it to ship on time.
What's your take? Would you feel cheated if you found out ambient NPC dialogue in a game you loved was LLM-assisted? Or is it just the next tool, like photogrammetry or motion capture were?