Like software, there's going to be a lot of pressure to use well documented tools within the model's training set. Innovation on the outputs may increase, but infrastructure and tooling will slow down.