The Julia programming language aims to offer a new approach to scientific high-performance computing. It attempts to combine the accessibility of Python/MATLAB with the performance of C or Fortran, and puts an emphasis on composability and reproducibility. In this talk, we will try to evaluate how these claims stand up to scientific reality by looking at Trixi.jl, an adaptive high-order numerical simulation package for conservation laws. Trixi.jl was designed as an extensible software library and focuses on usability and performance. We will show how Julia's package management infrastructure facilitates collaborative development workflows and allows one to interface with binary libraries. After an assessment of the serial performance of Julia code, we will review our experiences with using Julia for MPI-based parallel computations. Finally, we will discuss limitations we encountered in practice and share some of our lessons learned.