Argued
The commitment, argued.
Position statements on AI, judgement, and how the work of a domain should shape the tools built in it. Two of these carry the whole practice; the rest argue it in specific cases.
The commitment itself
Koher's deepest commitment: keep something real outside the model in the room. Two disciplines follow — judgement in code anyone can read, and the voice between people kept human.
When a tool sits between two people, a model must not become the one they both defer to. Each side gets its own private AI; the talking stays human — the witness half of the commitment.
Argued in specific cases
Everything these models speak with was once somebody's. Three debts, three speeds of repayment — and a response to the louder rooms made by demonstration, not enlistment.
Designing as if the user has nothing better to do is a quiet insult. Attention is a sliver of a crowded day — the demand-side twin of the parameterisation gap.
"Knowing the user" is the parameterisation gap lodged in the founding verb of a discipline. When the parameterised thing is a person, declaring the reduction is not enough — refuse to accumulate it.
What and why are yours. How is where AI lives. But the boundary is not a waterfall — the doing changes the knowing, and staying inside that loop is what AI use most often costs a student.
Koher's growth law: density is not mass. The number of nodes — tools, conversations, students, occasions — can grow without translating into more weight on any single relationship. The architecture is the mandala, not the tower.
A pedagogical position for design education. Encourage students to build with Claude Code, not around it. The negotiation between student and a more capable partner — what to keep, refuse, correct, redirect, and why — is the curriculum. SDC enacted by the learner, not contrary to it.
Three Koher tools are under MIT. Every tool released after 5 April 2026 is under AGPL-3.0. The split is a settled position with specific reasons. Why two licences, and what the split means for you.
The industry is racing to make AI output better — more models, better orchestration, higher benchmarks. But better output and visible judgement are different things.