The system should help
both sides do better work.

The bond

Particularity lives in the field.

A model supplies broad capability. The maintained field of context, tools, vocabulary, decisions, and refusals gives the work its particular shape.

An AI-bonded system tends that field deliberately. Change the model and much of the working relationship can survive; let the field contradict itself and the work becomes generic.

The peer stance

Enable reasoning. Keep authority explicit.

“Peer” is an engineering stance, not a claim about sentience. Give each side real context and real tools, set boundaries that let both reason well, and verify what matters.

A fluent answer does not grant authority. Respect and scrutiny belong together.

The loop

Connection matters more than efficiency.

AI can make a shape quickly. The work is staying present enough to recognize what is true, what is generic, and what the next cycle asks from you.

Diffuse input becomes a concrete shape. Honest reaction becomes the next input. Enough cycles create calibration.

Evidence

Keep fact, interpretation, and authority distinct.

A source claim is evidence, not current truth. A visual mesh is not exact geometry. An AI explanation is not the measured dimension beneath it.

Good systems preserve provenance and put limits beside the claims they qualify.

Negative space

What the system refuses is part of the product.

Not every capability belongs. A loud refusal is often more useful than a plausible approximation, and a smaller coherent surface is often more trustworthy than a broad ambiguous one.

Record the reason behind a boundary so it can survive the person who first drew it.

Legibility

Design for the person and the AI peer.

Semantic structure, stable routes, clear labels, readable Markdown, and explicit system state help both kinds of reader.

The machine-readable form should derive from the same source as the human page. If the two disagree, the publication is broken.