BusinessJan 2026 4 · 4 min read
AI-Assisted Level Design: What Claude and ChatGPT Get Right (and Wrong)
AI is decent at generating level design ideas and terrible at knowing which ones are actually fun without a human playing them.
Why this matters now
We use AI for breadth — twenty variations on an encounter layout in minutes — then apply human judgment for pacing, difficulty curve and the specific “feel” no model can playtest for itself.
What we're actually seeing
Studios that treat this as a one-time fix rather than an ongoing practice tend to see the gains fade within a quarter — the teams getting lasting results build it into their regular process.
Our recommendation
Use AI to widen the option set early, and keep the final pacing call entirely human.
Written by
Sandeep Kumar Tiwari
Founder & CEO, Enixo Studio
Leads studio strategy and full-cycle production at Enixo — from first pitch call to live-ops handover.
Enjoyed this? Get the next one first.
One email a month from the dev floor — zero spam, unsubscribe any time.
2 comments
JinxTheRogue 2 days ago
This is exactly the kind of breakdown I needed — saved for the next sprint planning.
PixelPilot 1 day ago
Curious how this holds up on lower-end mobile hardware?
Have a topic you want us to cover?
Pitch it to us — we might write the next one about it.
Pitch to us →