Implementing language runtimes is simple in the ideal world: language specifications are combined with well-abstracted APIs and various approaches to code generation and optimization to form functional and efficient VMs, and multiple such VMs are in healthy competition to perform best in various well-defined scenarios. In the real world, things tend to be more interesting: languages lack specification (or don’t adhere to it), APIs expose implementation details in obvious and non-obvious ways, vast ecosystems need to be supported, and users have unexpected expectations on overall system behavior. This talk will show the experience gained from the language implementations that are part of the GraalVM multilingual runtime, in terms of how the understanding of the languages evolved over time (and what to look out for) and how some of the resulting technical challenges were solved by applying generic and GraalVM-specific solutions.