The pace of AI development isn't slowing down — it's compounding.

Every week there's something that would have been a headline six months ago, buried under three bigger things. So here's what actually matters from this week — the shifts that are going to change how you build, how you think, and what's possible.

The models keep getting better at reasoning

The jump from "can answer questions" to "can work through a multi-step problem reliably" is happening faster than most people expected. What used to require careful prompt engineering and chain-of-thought scaffolding is now closer to default behaviour.

For game AI specifically, this matters. We're getting closer to NPCs that can genuinely reason about their situation rather than just pattern-match against training data.

Local models are becoming real options

Running a capable model on consumer hardware was a novelty twelve months ago. It's becoming practical now. Ollama has made the setup almost trivial, and the models available for local inference have caught up significantly with the hosted options for many use cases.

This is the thing I'm exploring in the second workshop — NPC personalities driven by a local LLM with no cloud dependency.

What this means for game developers

The tooling is maturing. The barrier to experimenting with AI-driven behaviour in your game project has dropped substantially. You don't need a PhD or a research budget. You need curiosity and a willingness to iterate.

That's exactly what the workshops are built around.

Mindplay with Aaron — game AI for engineers who build real things.