Arguably it really depends on your DSL right? If it has a semantics that already lies close to existing programming languages, then I'd agree that a few examples might be sufficient, but what if your particular domain doesn't match as closely?

Examples of domains that might be more challenging to design DSLs for: languages for knitting, non-deterministic languages to represent streaming etc. (i.e https://pldi25.sigplan.org/details/pldi-2025-papers/50/Funct... )

My main concern is that LLMs might excel at the mundane tasks, but struggle at the more exciting advances, and so now the activation energy for coming up with advances DSLs is going to increase and as a result, the field might stagnate.