AI UX Design · Context Grammar
8 context tokens, a rule engine, and 23 adaptive patterns. 21 real scenarios running live in your browser.
A real person in a real moment — with history, context, and a goal.
Explicit or inferred. Not a text string — a structured goal object.
6 context signals + 2 trust dials — where you are and what the system can do.
Three layers: Identity (months), Learning (weeks), Now (seconds).
Tokens × Brain → 23 AX Patterns. Delegate, escalate, or adapt.
Explore each element of the framework — from the first signal of intent to the patterns that shape AI behavior.
The full pipeline from Intent to AX Patterns — all seven entries, one framework. Start here.
Four channels of human signal — explicit, active, passive, ambient. Intent is a structured goal, not a text string.
Six situation signals and two trust dials — the variables that describe where you are and what the system can responsibly do.
Three layers of memory — Identity, Accumulated Learning, Right Now. Every Brain has the same anatomy.
Tokens × Brain → design rules. Three exits: Delegate, Escalate, Adapt.
YAML schema definitions for Context Tokens, Brain layers, and AX Pattern triggers — making the framework portable across teams.
23 named patterns — Delivery, Escalation, Autonomy, Cross-cutting — describing how UI should behave when the Rule Engine fires.
Autonomy Dial × Disclosure Dial — and the temporal arc that connects them. How the relationship grows from Day 1 to Month 6.
See how Context Grammar plays out across real contexts — home, travel, retail, and enterprise.
Intent decomposition × Family orchestration. Day 1 onboarding to Month 6 autonomy.
Parent-child shoe shopping, with the child as decision maker.
Enterprise Context Grammar. Four brains, one grammar.
Essays on agentic UX, Context Grammar, autonomy design, and what changes when organizations actually build with AI. Published on Substack.
Everyone is building AI. Almost no one is asking what it should actually do for people.
An honest look at the shift in design, without overstating what is new.
How far should AI be allowed to act before asking, confirming, notifying, or doing the work itself?
Read it. Build with it. Get in touch.