The Instinct Most Teachers Have
Most of what students have been taught about their own work is shaped by an instinct that is, on its own terms, correct: do your own work, do not copy, do not outsource your thinking. This instinct exists because in most contexts it is right. A student who hands their assignment to someone else has not done the assignment. A student who copies has not learned anything.
The instinct is doing important work. It should not be abandoned. But the instinct is no longer sufficient for the room a design student now sits in.
Claude Code, ChatGPT, and tools like them are in every Capstone studio whether teachers acknowledge them or not. Students will use them. The question for a teacher is not whether to permit them but whether to teach with them. And teachers, defaulting to the instinct, mostly choose not to. They tell students to do the work alone, or to use AI only for "ideation," or to mention it in a footnote. They build classrooms in which the most consequential tool of the student's professional life is treated as contraband.
This position argues for a different default. Encourage students to build with Claude Code, not around it. Then teach the negotiation.
What Changes When the Partner Arrives
A student working alone is bounded by what they already know how to do. Their work tends to inherit the contour of their existing competence; the unknowns at the edge stay unknown because there is no occasion to meet them.
A student working with Claude Code encounters a different shape of room. The partner produces, at speed, work that is sometimes better than the student could have produced, sometimes wrong in ways the student can detect, sometimes wrong in ways the student cannot yet detect, and sometimes shaped by aesthetic or architectural choices that the student would not have made.
The student's task is no longer to generate output and stop. It is to read what arrived, decide what to keep, decide what to refuse, decide what to correct, decide what to redirect — and to defend each of those decisions to themselves and, eventually, to a teacher who will ask why.
That ongoing decision is the negotiation. It cannot be performed by reflex; it requires the student to articulate what they want, what is good, and what is wrong — articulations that, before the partner arrived, did not need to be made.
A sparring partner who hits harder than you forces you to defend more, faster, and to learn the structure of your own defence in the process.
The Negotiation Is the Curriculum
What a teacher should grade is not the speed, the volume, or the polish of the output. It is the texture of the negotiation. Specifically:
What the Teacher Is Reading For
And, in each case, why. A student who can defend each decision is doing the assignment. A student who cannot — whose answer to "why this?" is "Claude Code did it" — is not.
The wager is that this kind of grading produces a steeper, more durable learning curve than any amount of solo-work-within-comfortable-competence. Most students underestimate how hard the negotiation is. That underestimation is part of why the curriculum needs to name itself explicitly.
The SDC Connection
A casual reader will spot what looks like a contradiction. Koher's foundational architecture argument — Split-Domain Cognition, or SDC — says that production AI tools should keep language work and judgement work structurally apart. AI does pattern recognition and narration; deterministic code does the judgement; a human decides. Doesn't putting Claude Code into the centre of a student's design process violate SDC?
The contradiction is only apparent. SDC's caution is about unconscious conflation in shipped tools. The pedagogy makes the conflation conscious. Claude Code, in a building context, does conflate language and judgement — it produces text, code, and architectural choices in a single stream, with embedded judgements that are not labelled or held apart. The pedagogical move is not to ship that conflation. It is to require a learner to enact the separation.
A semester of pulling apart what the partner is saying from what it is judging teaches, from the inside, why the separation matters in tools that ship.
A student who has done this work for a semester has learned, in the strongest possible way, why language and judgement need to be held apart in the production tools they will eventually build. The negotiation is SDC enacted by the learner, on a partner that does not enforce it on its own. It is the right way for someone who will go on to design tools to learn what those tools are made of.
What This Commits a Teacher To
Three commitments follow from holding the position.
The recommendation in the room. In Capstone, in studio mentorship, in any class led by a teacher who holds this position, students are told to build with Claude Code, not around it. The teacher's job is not to gatekeep the partner; it is to teach the negotiation.
The grading reorients. Feedback attends to the texture of the student's decisions: what was kept, what was refused, what was corrected, what was redirected, and why. A student whose work was produced through visible, defensible decisions is doing the assignment. A student whose answer to "why this?" is "the partner did it" is not.
The artefacts include the trail. The negotiation has to leave evidence — saved prompts, dated versions, short notes on moments of refusal, before-and-after pairs that show what was reshaped. Without the trail, the negotiation evaporates, and a piece of work made through real decisions looks identical to one handed to the partner. The trail is the curriculum's audit log.
What This Does Not Mean
It does not mean every student should build with Claude Code. The wager is that the negotiation produces a steeper learning curve on net for the majority of design students. There are students for whom the partner is overwhelming. There are tasks where the speed of a coding partner is the wrong scaffolding. The default recommendation is to build with the partner; exceptions are read case by case.
It does not mean Claude Code does the design work. The student does the design work; Claude Code produces material the student must then judge. A student who has handed their judgement to the partner has stopped designing. The pedagogy depends on the negotiation actually happening, which means the student must stay in the loop as the rules layer.
It does not mean Koher endorses AI tooling generally. The position is specific: for design education, with this kind of partner, in a setting where a teacher can see the negotiation, the negotiation is the curriculum. It does not extend to the claim that AI tools are good in general, that students should rely on language models for judgement in their adult professional lives, or that the partner is a teacher.
It does not change the architecture of Koher's tools. Production tools still hold language and judgement apart. The pedagogy applies to learning to build, not to what is built. A student who has done a semester of Learning by Negotiation will, when they ship a tool, ship a tool that holds language and judgement apart — because they will know, from the inside, why that separation matters.
The Autobiography
This pedagogy is not a new posture invented for the classroom. The same shape — an attentive practitioner working with material that is partly out of their control, learning the practice through the friction of that negotiation — runs through earlier projects and through Koher itself. Prayas negotiates daily with Claude Code as a co-author whose synthesis he must judge, accept, refuse, and redirect. Koher's published declaration of AI co-authorship names this division of labour explicitly.
A teacher who tells students to do what the teacher does not do is teaching something else. Koher tells students to negotiate with a more capable partner because that is what Koher does. The architecture is autobiography, and so is the pedagogy.
Related: The Boundary and the Loop — the framework this pedagogy enacts. What and why are yours; how is where AI lives; the loop between them is the curriculum.