LATEST · EP 153 · No AI Doesn't Mean What You Think It Does
Player Driven
──── PILLAR GUIDE

AI in Games

Generative AI, AI ops, AI policy.

12 topics36 pieces of content
──── WHY THIS PILLAR MATTERS

AI in games has meant different things in different decades. NPC behavior in the 90s. Procedural content in the 2000s. Machine learning for matchmaking in the 2010s. And now — generative AI for assets, code, dialogue, even whole gameplay loops. This last wave is doing in two years what we expected to take twenty.

The pillar covers three live problems at once. (1) Production AI — can you ship faster, cheaper, with smaller teams, using LLMs + diffusion + voice synthesis? (2) Runtime AI — what does it mean when your NPCs can actually talk, react, remember? (3) Policy AI — who owns model outputs, what trains on what, what disclosures do players deserve. None of these are settled.

Every studio is either using AI productively, or losing time arguing about whether to.

Reality of the 2025 game industry

What's clear: the studios that are pragmatic about it (real workflow integrations, real policies, real player communication) are pulling ahead of studios that are either over-reverent (we don't touch it) or over-zealous (we use it for everything). The interesting work is in the middle — understanding which problems AI actually solves and which it merely reframes.

──── THE BREAKDOWN

12 topics in AI in Games

Each bar is a topic in this pillar. Bar length is content volume — how much we've published about it. Tap any topic to drill in.

──── HOW GAMES USED THIS

Three studios. Three lessons.

Inworld + partners · 2022–present

Inworld AI in Nvidia ACE / Magic Leap

LLM-backed NPC characters that hold real conversations, remember player history, and react in tone. Demos blew minds; production deployments are still finding their feet — latency, hallucination, voice acting union concerns are all live issues.

AI NPCs are a fantastic demo and a real engineering challenge. The gap between Twitter clip and shipped feature is bigger than it looks.

Various, including Activision experiments · 2024–present

Promethean AI / asset generation pipelines

Diffusion-based 3D asset generation pipelines that compress weeks of artist work into hours. Real productivity wins on prop generation and environment dressing; less successful on hero assets that need brand consistency.

AI hits the long tail of content production hardest. The hero assets still need your best humans.

Industry-wide · 2024–2025

The SAG-AFTRA video game strike

Voice and motion-capture performers struck primarily over AI consent and compensation. Outcomes shaped how studios approach voice cloning, performance capture rights, and synthetic actor policy.

AI policy at your studio isn't optional. The talent unions are setting the bar; pretending you haven't decided is itself a decision.

──── THE OPERATOR'S CHEAT SHEET
↳ WHAT YOU MEASURE
  • ·% of production pipeline using AI tooling
  • ·Cycle-time reduction on AI-augmented tasks
  • ·Player-facing AI disclosure rate (do players know when content is AI?)
  • ·Legal exposure: trained-on dataset licensing audits
  • ·Quality regression rate: bugs/issues from AI-generated artifacts
↳ WHO OWNS THIS

Currently chaotic — varies wildly by studio. Best practice emerging: a dedicated AI Lead reporting into the CTO, with cross-functional reps from production, legal, and player experience.

↳ SIGNALS YOU NEED TO INVEST
  • ·Your competitors are shipping faster with smaller teams and you can't explain how
  • ·Production timelines are getting compressed by leadership without commensurate budget
  • ·Players are asking which assets in your game are AI-generated and you don't know
  • ·You're being asked to sign vendor contracts that mention 'model training' and your lawyer can't translate them
  • ·Your art / writing team is anxious — that anxiety needs a policy to land in
↳ COMMON MISTAKES
  • ·Adopting AI tools without a policy on training data provenance
  • ·Letting individual contributors decide what's OK case-by-case
  • ·Confusing demo capability with production reliability
  • ·Underestimating the QA burden — AI-generated content fails differently
  • ·Not communicating to players. The trust hit from a 'gotcha' discovery is bigger than the trust cost of upfront disclosure

Climb the AI in Games track.

Every piece of content in this pillar you finish earns credits toward your AI in Games level. See the full system at /level-up.