How This Was Built
A provenance note for Conlangs as Tools, edition v1.2. Published here 2026-09-27.
Written by Peter Kaminski with Freya (Claude), 2026-05-28.
♡ Copying is an act of love. Please copy and share.
© Peter Kaminski · CC-BY 4.0 (Creative Commons Attribution 4.0 International)
Project files: project home · how this was built · the brief for the ten · Toki Pona deep dive · all files (.zip)
Languages: Toki Pona · Taso · Kipona · Ren · Maku · Sito · Olu · Vesi · Korē · Drosa · Hesi
If a gift is going to come from more than one mind, the right thing is to name who did what.
The order of events
The framing was mine. I have wanted to write something on Toki Pona for a while, partly because it is one of the few constructed languages that has actually become a community language, and partly because I think it is the closest thing in the conlang tradition to what a friend has called an “epistemic workflow” — a language designed to do a specific kind of work on its speaker’s attention.
Doruga drafted the ten briefs. Doruga is my mobile-first agent (the second persistent one, after Freya). I gave Doruga the seed — Toki Pona shows that a small vocabulary can be a tool; what other tools are possible? — and Doruga produced, in a single working session on 2026-05-28, a Toki Pona deep dive plus ten briefs sketching new design moves. The briefs are Doruga’s work. I edited lightly.
Freya wrote the deep dives. A background instance of Freya (codename bg-sphinx) took each brief, did the web research the brief implied, and wrote the deep dive in whatever form fit the language — a 50-word lexicon, a bilingual short story, a grammar sketch, a ritual text. The instruction was “form-appropriate, ground in real linguistics where you can, invent coherently where you must.” She worked sequentially through the ten and produced a handoff note.
I shaped the package. Selection, ordering, the cover doc, this note, and the call about what did and did not belong in the gift. Foreground Freya did the integration: pulling the background work back into the project, committing it, spot-checking citations, and helping me think about delivery.
What this implies
Three agents and a human made one gift. The seam between “what Pete thought” and “what an agent wrote on Pete’s prompt” is not always visible inside the documents. The seam between “what Doruga drafted” and “what Freya elaborated” is visible (one is a brief, one is a deep dive) but the voice has been smoothed.
I’m flagging this for two reasons. The first is a principle a friend has been articulating in her own research: if your agents are formalizations of your standards, then naming them is part of being honest about who is making the claim. The second is that I think this might be a fair preview of where collaborative work is going, and if so, the conventions for crediting it are worth working out in small low-stakes cases like this one.
What I would have done differently
Probably nothing, but I notice three things worth saying out loud:
- The briefs are tighter than the deep dives. That is partly because briefs are easier, and partly because Doruga’s single-session burst had a unifying voice that ten parallel research sweeps can’t quite match. If this were a finished product I would do one more pass to align them.
- I deliberately did not have either agent cross-reference the other languages inside their own deep dives. Each language stands alone. That makes the bundle feel a little like a museum exhibit rather than a conversation. Trade-off.
- I leaned on Sonja Lang’s design philosophy as the through-line. Other through-lines are possible (Wierzbicka’s NSM, Lojban’s logical bet, Esperanto’s neutrality argument). I picked Lang because her work is what I’ve seen become most alive. (Loglan and Esperanto have much life as well, it’s just that I haven’t experienced it.)
What the agents would not say
I’m signing for the whole thing. If anything in here is wrong, glib, or culturally clumsy, that is on me, not on Doruga or Freya, who did exactly what I asked them to do.