The recent advent of deep learning considerably widened the range of possible applications for machine learning. Music is the most exciting for me. In particular, generative models reached a degree of sophistication and realism inconceivable a few years ago. Consequently, new sonic landscapes will emerge from the evolution of audio synthesis techniques and the workflow of composers and music producers could be strongly impacted by the introduction of tools allowing complex high-level symbolic manipulations of the inner structure of their creations. Crafting tools for musicians with these new technologies is a thrilling challenge. It raises many questions about what music creation could be today, which are our starting point in designing new algorithms. My main focus so far has been to work on a system which automatically performs the arrangement of a piano piece, and my main concern is to make it as interactive as possible.