> AI will struggle like with Unity/Unreal/Godot.

I am automating Unity with headless method invocation of agent authored editor scripts. I don't think "struggle" is the word I'd use to describe how GPT5.4 is currently performing.

I can tell the agent things like "iterate over all scenes. Wrap lightmap baking in a 5 minute timeout. Identify all scenes that exceed baking time. Inspect the scene objects and identify static geometry with poorly configured light map scale relative to their world space extents."