It currently does not have a REPL, and that is its main limitation right now. I think it is still suitable for more complex workflows where you would not do it

It currently does not have a REPL, and that is its main limitation right now. I think it is still suitable for more complex workflows where you would not do it in the REPL anyway - think training a large neural network. I hope something will come after 1.0 is out...