Context Grammar — Floor 1

Intent

Before AI reads the room, it must read what the user wants. Sometimes it's spoken. Often it isn't. Intent is the signal that starts every interaction — and the one today's AI almost always misses.

Intent is Floor 1 of Context Grammar. It's what the user wants in the moment — not a permanent goal, not a demographic. Two kinds: one spoken, one inferred. Get Intent wrong and every floor above it fails.

A · The Metaphor

Two kinds of restaurant — and two kinds of AI

One restaurant waits for you to ask. The other knows before you do. The difference is not magic — it's memory combined with situational awareness.

The key question for every AI interaction: are you designing a restaurant that waits to be asked — or one that already knows?

B · The Foundation

Intent has two forms — spoken and unspoken

Most people think Intent means "what the user said." That's half right. Intent comes in two kinds — and in real life, the unspoken one is far more common.

Explicit Intent

Spoken clearly

The user declares what they want — by voice, tap, or type. AI executes. No inference needed.

  • "Set a timer for 20 minutes"
  • "Call Mom"
  • "Buy Sota soccer shoes"
  • Imperative, direct, no ambiguity
Implicit Intent

Understood without words

Nothing is said. But the intersection of situation (Token) × memory (Brain) lets AI infer what's probably wanted.

  • Tuesday morning, uniform by the door → "soccer day"
  • Friday 7pm, fridge empty → "probably pizza?"
  • 11pm, reading in bed → "keep everything quiet"
  • Far more common than Explicit
The key distinction: Explicit is a command. Implicit is an atmosphere.
Today's AI (Siri, Alexa, ChatGPT) hears only Explicit. Context Grammar is the design language that makes Implicit readable too.

C · How It Emerges

Implicit Intent is not a guess — it's a function

Implicit Intent does not appear by chance. It emerges from a specific intersection: current situation signals (Tokens) read against stored memory (Brain), mediated by Coordinator. Remove any one of the three and the inference collapses.

Floor 2 8 Context Tokens What is true right now — physical state, cognitive load, time, form factor.
Floor 3 Multiple Brains Who this person is and what the household has learned — stored across 3 memory layers.
Floor 1 Implicit Intent An unspoken want — expressed as a proposition + confidence score.
Intent = what is wanted. Token = what is happening right now. Brain = who this person is.
They are three separate things — but Implicit Intent can only emerge from the intersection of Token × Brain, with Coordinator doing the cross-reading.

D · A Real Scenario

Tuesday morning in the Yamashiro house — Intent arrives before Sota speaks

Sota (11) winces while trying to put on his soccer shoes one morning. He says nothing. And yet, the home's AI already knows his Intent at 92% confidence. Here's how it gets there.

(Sota frowns at the door, struggling to get his soccer shoes on)

— Tuesday 7:42 · Yamashiro House · Zero words spoken

1 Context Tokens shift — the environment's sensors register a signal Token
Sota hasn't said a word. But the home's Context Tokens are already moving. Tokens are 8 instruments that read "what is the world like right now." This morning's readings:
Token readings — current values
  • ① Physical StateStride + posture sensor detects "foot discomfort" for Sota
  • ② Cognitive Load7:42am · 10 min to school departure · estimated LOAD = HIGH (low-bandwidth mode)
  • ③ Social ExposureFamily only (private) · speakers OK
  • ⑤ Form FactorIndoors · refrigerator display directly in line of sight
2 Multiple Brains corroborate the signal Brain
Tokens alone only get you to "a boy with foot discomfort." This is where Brain (memory) enters. In Context Grammar, a home doesn't run on one Brain — multiple Brains operate in parallel, each remembering from a different angle.
Brains consulted in parallel
  • Person Brain (Sota)Age 11 · soccer club · feet have grown 5mm recently · peak growth phase
  • Household BrainPast 3 months: 2 mentions of "shoes feel tight" · previous delay caused shoes to not arrive before match
  • Calendar BrainToday is Tuesday → soccer practice · match next Saturday
3 Coordinator crosses Token × multiple Brains — Implicit Intent emerges Inferred
Coordinator is the layer that reads Tokens and Brains together. It intersects Token (foot discomfort + Tuesday morning) × Person Brain (Sota in growth phase) × Household Brain (previous failure pattern) × Calendar Brain (match in 5 days). An unspoken Intent surfaces.
Inferred Implicit Intent
  • Surface reading"Shoes feel tight and hard to put on"
  • True desire"New soccer shoes needed before Saturday's match"
  • Confidence92% — corroborated by all three Brains: Person × Household × Calendar
→ Conclusion: put Mai one tap away from "Sota's shoes may need replacing?"
4 Sota says one thing — Explicit Intent confirms what's already known Explicit
Sota turns to his mother and says:
"Mom, my soccer shoes are too tight!"

Only now does Explicit Intent appear. But here is the interesting part — Coordinator had already reached the same conclusion through Implicit inference before Sota said a word.
His statement is a confirming override, not a first discovery.
5 A notification arrives on Mai's iPhone — zero cognitive overhead Result
While Mai is cleaning up breakfast, one line appears on her iPhone:
"Sota's soccer shoes — outgrown. Match Saturday. Tap to auto-order ($48, EU 37)."

Mai does not decide, research, or ask about the size again. One tap and it's done.
That is what a home with readable Implicit Intent looks like.
Key takeaway: Before Sota said "my shoes are tight," Tokens and multiple Brains had already established 92% confidence in his Intent. Today's AI waits for the words. Context Grammar's AI is ready before the words arrive.

E · Architecture

Intent and the Brain — one signal, multiple memories

"Intent without memory is just a guess." But that memory is not one big database. In Context Grammar, a home runs multiple Brains in parallel — each with a distinct role. Implicit Intent only emerges when Coordinator reads them together.

Floor 1 Intent Current moment signal — what the user wants right now.
Floor 3 Multiple Brains Person · Household · Calendar, each with 3 memory layers.
Meaning Actionable Intent An intent that knows who it belongs to and when it's urgent.
Person Brain The brain that remembers "who this person is"

One per family member. 3-layer structure — Identity Layer (Sota is 11, soccer club), Learning Layer (shoe size changes every 3–4 months), Now Layer (unusual gait this morning). Determines the subject of Implicit Intent.

Household Brain The brain that remembers "the family's shared history"

Purchase history, unresolved household tasks, budget priorities, past failure patterns ("the shoes weren't replaced in time last season"). Provides the family-level context invisible to any single person's brain.

Calendar Brain The brain that remembers "when things happen"

Family calendar, school schedules, work travel, match dates. Determines the urgency of an Implicit Intent. No Saturday match means the shoes can wait. A match in 5 days makes it "buy now."

Coordinator The layer that reads all Brains together

Not a Brain itself — a cross-Brain reasoning layer. Reads Tokens alongside Person, Household, and Calendar simultaneously, resolves conflicts, and reduces everything to a single Intent proposition. Disclosure Dial and Autonomy Dial settings are read here too.

Why separate Brains, not one?
Separating Brains by role means Disclosure Dial can control "which Brain can share what with which Brain." Sota's health data stays in his Person Brain. Household finances stay in the Household Brain. One giant merged Brain can't draw those lines.

F · Four Channels

Four channels of Intent — from command to atmosphere

Explicit / Implicit is the right entry point. But when implementing, breaking Implicit into three sub-types keeps design decisions sharp. Floor 1 covers four channels total.

Four channels of Intent — Explicit, Active, Passive, Ambient — stacked from most deliberate to most ambient
Type 1

Explicit Intent

The user states what they want by voice, tap, or type. Zero inference needed.

Example "Call Mom."  /  "Set a timer for 20 minutes."
AI just executes.
Type 2

Active Intent

The user says nothing, but their current action reveals what they want.

Example Mai picks up her phone at 6:42am in running gear with earphones in.
Intent: "About to run — weather, route, playlist."
Type 3

Passive Intent

The user has a pattern. This moment matches it — so AI knows what usually follows.

Example Friday 7pm. The Yamashiro family has ordered pizza every Friday for 3 months.
Intent: "Probably pizza. Ask before acting."
Type 4

Ambient Intent

A continuous background signal. No specific request — just a state the environment should respect.

Example Mai reading in bed at 11pm.
Intent: "Stay calm. Keep it warm. No notifications. Dim everything but the book."

Types 2–4 are all sub-types of Implicit Intent. Today's AI handles only Type 1. The other three are the "moments you wish your phone would handle but it never does." Context Grammar makes all four structurally addressable.

G · Intent vs JTBD

Intent and Jobs-to-be-Done — similar terms, different timescales

Designers with a product background often conflate Intent with JTBD. Both are necessary — but they operate at completely different time horizons.

Framework

Jobs-to-be-Done

A durable goal the user "hires" a product to accomplish. Defined once, valid for months or years.

  • "Help me stay organized as a parent."
  • "Help me stay fit without scheduling workouts."
  • Drives product strategy — which problems to serve.
  • Lives in roadmaps, research decks, annual planning.
Context Grammar

Intent

A momentary signal of what the user wants right now. Updated every minute, every tap, every room change.

  • "Pizza, probably. Ask me."
  • "Running in 5. Don't interrupt with email."
  • Drives interaction execution — what AI should do this second.
  • Lives in the runtime, between the user and the Rule Engine.

JTBD tells you what product to build. Intent tells you what the product should do in the next 200 milliseconds. You need both. One without the other gives you either a well-researched product that is constantly annoying, or a responsive AI that is solving the wrong problem.

H · The Formula

Implicit Intent — in one line

Implicit Intent is produced by a specific function. Coordinator executes it. The output is always a proposition paired with a confidence score — never a bare assertion.

Implicit Intent (t)
  = Coordinator (
    Tokens (t),
    Person Brain (subject),
    Household Brain,
    Calendar Brain,
    Disclosure Dial (per-domain)
  )
  → { proposition, confidence }

Output is always a proposition + confidence pair. Confidence becomes an input to Autonomy Dial stage selection.

I · How It Connects

Intent and the 8 Context Tokens — subject and modifiers

A common question: "Isn't Intent just one of the Tokens?" No. Intent and Tokens are separate layers. Intent is what is wanted. Tokens are the situational facts that shape how that want gets fulfilled.

Intent → 8 Tokens → Execution flow

Floor 1 · INTENT Implicit Intent "new soccer shoes" 8 CONTEXT TOKENS (Floor 2) ① Physical State ② Cognitive Load ③ Social Exposure ④ Priority Weight ⑤ Form Factor ⑥ Feasibility ⑦ Autonomy Dial ⑧ Disclosure Dial reads Floor 4–5 · EXECUTION 1-tap notification "$48 · EU 37 · auto?" applies
How to read this:
① Intent = what is wanted (new shoes)
② 8 Tokens = what is happening right now (morning rush · high Cognitive Load · fridge display in view · budget OK · Autonomy = confirm)
③ Execution = deliver Intent through the constraints the Tokens define

Same Intent, different Tokens → different execution. If Mai checks her phone at 11pm after lights out, the notification doesn't appear at all (Cognitive Load low · Ambient Intent = quiet mode). The 7am timing is what makes it a one-tap notification.

J · Common questions

Common questions answered

Q1. Is Intent the same as a "Goal"?

No. A Goal (Jobs-to-be-Done) is a long-term target: "lose weight this year." Intent is a moment-level signal: "what do I want in the next minute?" Goals are annual; Intent is per-minute. Goals go in roadmaps; Intent runs at runtime. Both matter — they're just different things.

Q2. Isn't Implicit Intent creepy — being guessed at?

That's exactly why Disclosure Dial and Autonomy Dial exist. Users can set "this far is fine to infer, beyond that ask me" — per domain, per person, per Brain. Sota's shoe situation crosses three Brains (Person × Household × Calendar) because permission exists. Aoi's health data stays inside her Person Brain only. Inference happens only where permission is granted.

Q3. What happens when Implicit Intent is wrong?

Design for being wrong — keep the cost low. Confidence drives Autonomy Dial stage. At 92%, it's still a one-tap confirm, not a silent auto-buy. At 50%, no notification — just a note in the Person Brain's Learning Layer. Low-cost failures by design. Silent wrong purchases are never acceptable.

Q4. How is this different from ChatGPT / Siri / Samsung Family Hub?

ChatGPT and Siri respond to Explicit Intent only — they wait for you to type or speak. Apple Intelligence and Samsung Family Hub have some Implicit capability (household profiles, habit learning, up to 6 member profiles, Voice ID, AI Vision) — but without a structured Intent model. Context Grammar proposes structuring Implicit Intent as Token × multiple Brain intersections, with Coordinator mediating — making "the experienced server's water refill" reproducible by design.

Q5. Where is Intent stored?

Intent itself is not stored. It's a moment-level signal, not persistent memory. Sota's "needs soccer shoes" Intent belongs to this morning — tomorrow's morning has its own Intent.

However, patterns learned from Intent are stored in the appropriate Brain's Learning Layer:
· "Sota's shoe size changes every 3–4 months" → Person Brain (Sota)
· "The Yamashiro family misses the window 5 days before a match" → Household Brain
· "Tuesday mornings have HIGH Cognitive Load 10 min before school" → Calendar Brain
Intent flows through; patterns persist — and they persist in different Brains.

Q6. Why not just merge all Brains into one?

Because merging collapses privacy and ownership boundaries. Aoi's menstrual cycle belongs inside her Person Brain — it should never flow to the Household Brain or Calendar Brain. Household finances belong in the Household Brain — not the children's Person Brains. Separate Brains mean Disclosure Dial can control "which Brain shows what to which Brain" one relationship at a time. One giant merged Brain can't draw those lines.

You have the signal.
Now: the vocabulary.

Intent tells AI what the user wants. The 8 Context Tokens tell AI what is true about the moment that wanting is happening in. Together they turn a one-word request into a full sentence.

Next: Context Tokens → Back to Overview →