Koher builds free tools, and a practice, that keep a real outside in the room — and refuse to let an AI take the seat where that outside belongs.
For a long time Koher named this commitment by one of its halves: separate language from judgement. That half is real and still holds. But it was always one instance of something larger. This is the larger thing, named so the practice can be read whole.
What a "real outside" is
An outside is anything you cannot quietly adjust until it agrees with you. It comes in two forms. One is a standard you set in advance — a criterion you commit to while clear-headed, that then stands against your later wish to be agreed with. The other is a real person across from you — someone with their own stake who can tell you, in the moment, that you have gone wrong, and answer for saying so. Both are outsides because neither is yours to re-tune. A system built to please you is the opposite of an outside: ask it to judge and it returns your own view at high fidelity; put it between two people and it tells each a version they can accept.
The two disciplines
Keeping a real outside in the room shows up in two places, and Koher holds a discipline in each.
Code holds the judgement. Where a consequential judgement can be written down as explicit rules, Koher keeps it in code anyone can read, and confines the model to bounded language work — so a model can't hand you a verdict you stop checking. This is Split-Domain Cognition: separate language from judgement. It is one half, not the whole.
People hold the conversation. Where a tool sits between two people, Koher keeps the voice between them human, and gives each person their own private AI that speaks only to them — so a model can't become the third party they both defer to. See People Hold the Conversation.
These two are co-equal halves of one commitment — and deliberately different in shape. The code half scales: a criterion is written once and runs over anything. The people half does not, and is not meant to — it holds only at the scale of two.
They also compose. The private tool each person uses in the people half must itself carry the code-half discipline — describe, don't flatter — or the agreeable mirror simply re-forms in private, inside each person's own tool, and the move fails quietly. Keeping a real outside in the room means refusing that mirror everywhere: not in the judge's seat, not in the voice between two people, and not in the private tool either. No agreeable mirror anywhere, shared or private.
This is what Koher builds — not a rule for how anyone must use AI
This is a specific stance about how Koher builds its own tools. It is not an across-the-board recommendation about how anyone should use AI, and not an argument against AI. AI in its lane is kept everywhere here — it reads signals, narrates decisions already made, privately augments one person's thinking. Koher chooses, for the two named places above, to keep judgement in code and the voice between people human. It does not prescribe that choice for all AI use, and does not claim other uses are wrong. It makes the choice, for itself, and explains why.
Not a law of mind
This is a discipline Koher chooses, not a law of mind it discovered. Nothing about cognition forces these splits — Koher holds them because of what they protect. Said as a found law, the claim overreaches. Said as a commitment you can argue with, it is honest. Each half marks its own boundary: the code half declines, by design, the judgements no rule can carry; the people half declines any setting larger than a willing pair.
Legibility for contestation, not measurement for control
Keeping judgement in readable code is not a call for more measurement. It is the opposite move: it forces the value-choice into the open, where the person it binds can see it and contest it, instead of hiding it behind a fluent verdict. The capacity to inspect is meant to be held in common, not owned by whoever decides. Used as a tool of optimisation rather than a discipline of refusal, this same machinery can be pulled into audit culture — Koher names that risk rather than claiming to be immune.