Could an LLM create a full Domain-Specific Language?

This was the question I asked myself to prepare the invited talk: “Who will create the languages of the future?” for OOPLE’25.

So we decided to give it a try and see whether I could use a LLM (to be more precise, Claude 4 with Cursor as “front-end” to interact with Claude) to create a small Domain-Specific Language (DSL).

In particular, I aimed to create a DSL for specifying FUNDING files used to specify how to manage sponsorship money in an open source project. I asked the LLM to create:

• The abstract syntax (i.e. the metamodel) of the language
• An ANTLR-based syntax and a TextX based one (together with the parser and the transformation from the AST to a set of objects instance of the classes of the metamodel)
• A graphical notation and a graphical modeling environment for the language.

The whole process (including the prompt I used, the generated artefacts) , the final results and my thoughts about the implications of the experience for the future of language engineering are all summarized in the presentation below.

The TL;DR version would be “yes, you can vibe your DSL, especially for the textual syntax, not so much for the graphical one and there are quite a few tricks you can use to improve the result”. But please check the slides for the full details.

Happy to keep the discussion going!

Jordi Cabot

FNR Pearl Chair. Head of the Software Engineering RDI Unit at LIST. Affiliate Professor at University of Luxembourg.  More about me.