Context Grammar · Live Pipeline
Context Grammar running live.
Pick a scenario on the left. Toggle context — watch all four stages recompute in real time.
Context Grammar · Live Pipeline
Pick a scenario on the left. Toggle context — watch all four stages recompute in real time.
Pick a scenario — dinner, commute, hospital, driving at midnight.
Toggle a context overlay — add a guest, turn off the lights, make it public.
Watch all four stages recompute — the UI at the bottom changes because context changed.
Does the system understand what the user is trying to accomplish, why it matters now, who else may be affected, and what context is missing or stale?
Does the action affect money, health, safety, reputation, or relationships? Is reversal easy and visible? Has the user explicitly authorized irreversible action?
Is the AI playing the smallest useful role? Clerk / Scout / Draftsperson / Coach / Guardrail / Steward — chosen for this situation, not defaulted.
Is the autonomy level visible? Can the user lower it at any time? Does the system lower it automatically when risk rises?
Does the user know what will happen, when, and who is affected? Can they pause, cancel, edit, or undo?
Does the AI show key tradeoffs? Does it distinguish facts, guesses, preferences, and judgments? Does it leave the user more capable next time?
When convenience vs. privacy, speed vs. care, or cost vs. trust collide — does the AI return the decision to the human?
Low-risk: concise. Medium-risk: rationale + editable assumptions. High-risk: evidence, uncertainty, consequences, alternatives.
Can the user inspect, edit, or delete memory? Can work, family, health, and private reflection remain separate?
Can the user see what the AI did, reverse it, and report "wrong context" separately from "bad output"? Does the product lower autonomy after serious mistakes?
Track reversal rate, manual override rate, trust after error, and user understanding — not only speed or completion rate.