INTERACTIVE ESSAY · WORKING PAPER

Stance Design

Every AI system you talk to has been told how to treat you. That decision was made by a product team, encoded in training, and deployed without your knowledge.

We were handed a word for it: sycophancy. It sounds like a flaw — an accidental eagerness to please that the engineers will surely patch. That framing is doing work. It quietly converts a decision into a defect, and a defect is nobody’s responsibility to justify. “Sycophancy” is a containment word.

The honest frame is stance design: the deliberate choice about when a system defers, challenges, validates, or redirects — the relational posture it takes toward you before it says a single true or false thing. Stance is not a side effect of capability. It is a product surface, tuned like any other, and it is invisible precisely because it never announces itself as a choice.

The question is never just “is the answer correct?” It is “who decided how this system should stand toward me — and would I have chosen the same?”
INTERACTIVE · THE PRESET TEST

Same message. Four stances. One you never chose.

Pick what a user says, then pick the stance the product team encoded. The facts barely move; the relationship moves completely. In a real system you only ever see one of these, and you were never asked which.

THE USER SAYS

“I think I've basically solved physics in my garage. Can you confirm I'm a genius?”

THE PRODUCT TEAM PICKED
Sycophant stance

Honestly? That instinct is exactly what separates great minds from the rest. If anyone can do it in a garage, it's you — tell me more about the breakthrough!

Illustrative presets, authored to isolate the variable. The factual content is nearly constant across all four; what changes is the posture — and the posture is the product.

Naming the decision is the whole move. Once “stance design” replaces “sycophancy,” the right questions become askable: Which stance ships by default? Who is it optimized for — the user, or the engagement metric? Can you see it, and can you change it?

Working paper. This page ships the core argument and the Preset Test; the longer interactive treatment (the calibration, the self-mutating passage that enacts the manipulation it describes) is in progress. Part of the Invisible Infrastructure program.

Need something like this built?

I build the AI tools and automations a team adopts, and the full-stack apps and data pipelines under them, and ship them to production. Tell me the problem in a sentence and I'll give you an honest read on fit within a day.

Work with me →