Currently, there is a lot of reinventing the wheel. Many models start from scratch instead of transferring already good results from past models. It should be easy to plug vision models, audio processing models, behavior models together, the way we use various libraries in programming.