The current explosion of embedded systems (i.e., IoT, Edge Computing) implies the need for generating tailored and customized software for these systems. Instead of using specific runtimes (e.g., MicroPython, eLua, mRuby), we advocate that bootstrapping specific language kernels is a promising higher-level approach because the process takes advantage of the generated language abstractions, easing the task for a language developer. Nevertheless, bootstrapping language kernels is still challenging because current debugging tools are not suitable for fixing the possible failures that occur during the process. In this paper, we take the Pharo bootstrap process as an example to analyse the different challenges a language developer faces. We propose a taxonomy of failures appearing during bootstrap and their causes. Based on this analysis, we identify future research directions: (1) prevention measures based on the reification of implicit virtual machine contracts, and (2) hybrid debugging tools that unify the debugging of high-level code from the bootstrapped language with low-level code from the virtual machine.