AXIS

AXIS is a lightweight symbolic grammar for communicating with AI systems. It introduces a small set of operators — glyphs — that mark intent, scope, and structure within a message. The result is not a different way of writing; it's a structured layer on top of natural language that makes what you mean unambiguous.

The problem it addresses is simple: unstructured language creates drift. When you write to an AI the way you'd text a friend, the model is guessing at intent, scope, and weight. Sometimes it guesses right. Often it doesn't — and the correction loops, retries, and misread outputs that follow are the cost of that ambiguity. AXIS removes it at the source.

The protocol is language-agnostic, model-agnostic, and built to work anywhere you can type. No install. No API. Plain text.

AXIS came out of a practical problem: the standard way of communicating with AI systems is informal, and that informality creates cost. Ambiguous language generates guessing. Guessing generates drift. Drift generates correction loops, retries, and wasted time — for both sides of the exchange.

The goal was a more efficient shared language. Not a new interface, not a wrapper, not a plugin — just a small, agreed-upon grammar that both human and AI could use to structure what was being said. Less friction. Less confusion. A more equal, less coercive dynamic: both parties working from the same map, rather than one constantly trying to interpret the other.

The operators that make up AXIS were arrived at gradually, through sustained practical use — not from theory. They represent the structural moves that most often determine whether an AI exchange succeeds or breaks down: opening the field, transferring material, marking genuine questions, signalling a single next action.

AXIS is developed under ensō, a research project exploring what it takes to build coherent, trustworthy working relationships with AI systems over time.

AXIS makes a measurable claim: structured prompts produce fewer correction loops, shorter completions, and less compute waste than unstructured equivalents. The PRAXIS measurement protocol defines how to test this — baseline agent versus AXIS-governed agent, identical tasks, logged token counts.

The claim is testable. We intend to test it publicly. Results will be published when available.