Why don't beat LLM-as-a-judge or traditional classifiers

This article misses the most crucial point. Developers don't "use the platform" when it results in a strictly worse end result.