The model picker is a dead end

Lovable argues that letting users choose AI models is a dead end. Instead, the company builds a 'control plane' that dynamically assigns different models to different parts of a build, based on real-time progress and context. This model-independent approach involves tailoring instructions and tools per model, and even training custom models for specific tasks. The goal is to optimize for the finished application, not benchmark scores, and to adapt as the frontier evolves.