Model composability is one of Julia's and MLJ's highlights. Library designers tend to take this into account throughout the ecosystem. It helps that Julia does not need to dropdown into a C or FORTRAN wrapper like Numpy.
That being said, MLJ tends to wrap a lot of models from scikit-learn with a composable wrapper, meaning you end up having to manage Python dependencies and you lose Julia features such as multi-threading (Edit: when using Python models, multi-threading is supported when using Julia models).
Long term I am sure MLJ will slowly replace scikit models but right now it can be little painful to use if the model you want is not implemented in Julia.
That being said, MLJ tends to wrap a lot of models from scikit-learn with a composable wrapper, meaning you end up having to manage Python dependencies and you lose Julia features such as multi-threading (Edit: when using Python models, multi-threading is supported when using Julia models).
Long term I am sure MLJ will slowly replace scikit models but right now it can be little painful to use if the model you want is not implemented in Julia.