I wouldn't call it bewildering. We're already seeing some capacity of models to take PRDs and turn them into PoCs. We're not far from being able to literally te

I wouldn't call it bewildering. We're already seeing some capacity of models to take PRDs and turn them into PoCs. We're not far from being able to literally test PRDs by how well the models implement them. In other words, how well can a PM write inputs makes a difference. And we're close to having quasi instant "scoring" for that.