hobbies at the DSL boundary

Adam Chalmers works at zoo.dev — parametric CAD driven by a text-based DSL. The DSL framing is his: any domain with a formal grammar is now an LLM superpower. The question is which domains.

A friend is trying to figure out the coolest ways to prompt LLMs for new ideas around introspection and mental health, or maybe building up new hobbies. The angle that Adam Chalmers pointed out was that any tech with a DSL is now superpowered because of LLMs. Excel for tables, zoo.dev for CAD, etc. Is there something in there that could help someone rediscover new ways to look at old hobbies?

The Chalmers insight is exactly right, and it generalizes further than most people take it. The key move: a DSL isn’t just a syntax — it’s a compressed ontology of a domain. LLMs don’t just translate natural language into the DSL, they make the ontology navigable by non-experts. Excel’s ontology is cells / ranges / functions. KittyCAD ’s is solid geometry operations. The question becomes: what domains have rich formal structure that people already care about but can’t access without years of skill-building?

Music theory is the killer case. ABC notation, MusicXML, LilyPond — these are DSLs for music. Someone who played guitar in college and dropped it can now say “give me a chord progression that sounds like Radiohead meets bossa nova, in tab format” and get something playable. The LLM navigates the theory; the human brings taste and fingers. That’s not replacing the hobby — it’s removing the bottleneck that killed it (theory fluency) and leaving the part that was always fun (playing).

Cooking has this structure too, though people don’t think of it as a DSL. Flavor compound databases, technique taxonomies, ratio frameworks. “I used to love Thai cooking but got bored — what’s a dish that uses the same aromatic profile but a French technique?” That’s a query over a latent DSL.

Nikki Segnit’s The Flavor Thesaurus maps 99 ingredients as nodes in a flavor graph — that’s the DSL made legible without LLMs. Now you can query it in natural language.

The meta-move: the prompt isn’t “give me ideas.” The prompt is “here’s what I used to love about X — what’s the DSL-adjacent structure underneath that, and what other domains share it?” Someone who loved darkroom photography might discover they were actually in love with process control with aesthetic feedback — which maps to sourdough, to generative art, to ceramics glazing.

The question isn’t “what new hobby should I try” — it’s “what was the latent structure I was always attracted to, and where else does it live?”