None of these are Priya's failure — they're the shape of knowledge inside any company old enough to have veterans, a research archive, and onboarding. The same three patterns recur in every org.
PriyaB2C Product Owner · 31
MayaDesign Lead · 29
DevEngineering Lead · 34
LeenaBrand Director · 41
KenjiB2B Product Owner · 35
ArmanIncoming PO · 28 (arrives 2029)Four designers are waiting on a color-rationale answer. Leena (Brand Owner, 41) last read Slack at 2:47 PM. Non-urgent questions roll to tomorrow; urgent ones interrupt a meeting.
Consumer_Insights_Q3_2025_FINAL_v3.pdf — last access Oct 2025. A Slack thread just spent two hours arguing a question whose answer lives on page 14. Research wasn't missing. A delivery language was.
Priya at the whiteboard: "There was a rebrand two years ago — not the one in the guidelines, the real one…" Each retelling drifts. This cost never shows up on a timesheet.
The same Context Grammar that quietly runs a household in Project 01 scales to a company. A person has one brain with three layers. An organization has many brains, because thinking is distributed across people, brands, and projects. Five brains, each with the same three-layer anatomy.
Org Brain holds what's true about the company itself — org chart, policies, glossary, history. Brand Brain holds judgment DNA per brand (B2C and B2B run two of these in parallel). Research Brain holds user understanding, with live data flowing in continuously. Project Master Brain is the strategist — it reads the other three and the past project history, and decides what to start, hold, or kill. Each Project Brain is time-bound — born at kickoff, dies at close, leaves its learnings behind.
Inside every brain, the same three-layer anatomy: Identity · Learning · Now. Identity is what doesn't change. Learning accumulates. Now is in flux. Layer is the universal unit of memory; brain is the unit of cognition.
Concept · Brain architecture Org Brain at the foundation. Brand Brains (one per brand) and Research Brain on top. Project Master Brain reads them all and spawns time-bound Project Brains — each with the same three-layer anatomy (Identity · Learning · Now).
A Context Token isn't a description. It's a contract every surface and every agent can subscribe to. Same value, two readers.
{
"id": "atlas.intent_fidelity",
"value": 0.62, // 0.0–1.0
"trend_7d": -0.20, // dropped from 0.82
"confidence": 0.81,
"scope": "project:apollo/sprint:3",
"sources": [
"identity_layer.intent",
"sprint.artifacts.figma",
"sprint.artifacts.copy",
"sprint.artifacts.commits"
],
"trigger_event": {
"kind": "standup_decision",
"ref": "meeting:2026-04-17/dec:#214",
"actor": "dev"
},
"projection": {
"high": { "autonomy_default": "auto" },
"medium": { "autonomy_default": "notify" },
"low": { "autonomy_default": "confirm",
"surface_question": true }
}
}
Backstage · Token contract The schema is the spec. Every surface that consumes Intent Match reads the same JSON; the projection block tells each surface how to render the value at low/medium/high. Add a new surface tomorrow — it doesn't need a meeting, it needs to subscribe.
The foundation under everything else. Org Brain holds the structural facts of the company — who reports to whom, what the policies are, the glossary of internal terms, the history of how the company got here. Slow-changing, all-employee, permanent.
Other brains read from it. Brand Brain references the company's voice principles. Project Master Brain reads team capacity from the org chart. Research Brain references customer-segment definitions from the glossary. Org Brain isn't where decisions happen — it's the ground decisions stand on.
Founding story, mission, legal entity, leadership structure, brand portfolio (B2C + B2B), office locations, time zones.
Org chart history, role evolution, glossary of internal terms, policies that have been refined, headcount and seniority patterns, how reorgs have gone before.
Who's currently on which team, this week's capacity, who's on PTO, today's all-hands message, current org-wide priorities.
Slack, Jira, Figma, Drive, the Project Master Brain app, the laptop reader — none of them store the Brain. They subscribe to it. Permissions are enforced at the API layer, once, not in every surface.
GET /brain/project/apollo
/identity_layer
?role=design_lead
&scope=current_sprint
→ 200
{ "intent": "...",
"decisions": [ ... ],
"open_questions": [ ... ] }
POST /brain/project/apollo
/learning_layer/decision
{
"statement": "Nav: tabs over rail",
"constraint": "discoverability",
"alt_considered": "info density",
"actor": "dev",
"meeting_ref": "2026-04-17:dec:214"
}
SUB /brain/project/apollo
/intent_fidelity
on_change:
threshold: 0.70
notify:
- role: product_owner
surface: iphone
Backstage · Brain API Three verbs, one contract. Surfaces don't argue about who has the truth — they all read the same Brain. Permissions live in the API, not in 11 surfaces with 11 implementations.
The company runs a B2C product and a B2B platform side by side. They share an Org Brain — but their voice, their visual judgment, their ethical stance on tradeoffs are different. So Brand Brain runs as two parallel instances: Brand Brain · B2C (Priya's domain) and Brand Brain · B2B (the B2B Product Owner's domain). The Brand Director holds judgment authority over both.
Tuesday afternoon. The Design Lead (Maya) pauses on a tweet draft for the B2C side — secondary color catches her eye. Why this color? The judgment from a rebrand three years ago lives in Brand Brain · B2C's Learning Layer. When she queries, the original reasoning replies — not a style-guide excerpt. The Design Lead has her answer without interrupting the Brand Director.
User UI · Laptop The Brand Brain reply landing in the Design Lead's design tool — source, verification, and original decision context travel with the answer.
Priya is on the 9:00 to London Bridge. She drafts a tagline candidate — "Ship your best work, together." — and asks Brand Brain · B2C to check it. The reply lands before her stop.
Watch what arrives in the screen on the right. Brand Brain doesn't approve or reject — it returns three pieces of context:
User UI · iPhone Tap Replay reveal to re-trigger the animation.
The Research team has been running studies for years. 10–15 a year: ethnographies, surveys, interview rounds, A/B tests. Each one ships as FINAL_v3.pdf, lands in a Drive folder, and is opened by maybe four people. When a project kicks off, someone skims the most recent one. Three weeks in, the kickoff brief overrides what the research said. The work was done; nobody could use it.
Research Brain follows the same three-layer anatomy as every Brain in the system. Old PDFs and live data aren't piled into one bucket — they sit in different layers, and that distinction matters at query time.
Here's how Priya uses it. When a new project is about to start, she briefs Research Brain on the project context — goal, segment, open questions. Research Brain reads across all three layers — methodology from Identity, past studies from Learning, today's data from Now — picks what's relevant to this project, and writes those fragments into the new Project Brain's Learning Layer. For the next six weeks the team isn't searching Drive — they're querying their own Project Brain, which already knows what 2024's segmentation study found and what NPS dropped this morning.
User UI · Laptop Priya briefing Research Brain on her new project. Relevant fragments — 2024 segmentation study, 2025 interview behaviors, today's NPS drop — staged for the new Project Brain.
Sprint 4, a new engineer asks: why did we change the nav in Sprint 2? The Meeting Brain surfaces the 90-second clip where Dev argued for discoverability over density.
Automatic transcript, extracted decision log, the why attached to each decision — not just what was decided, but which constraint it was weighted against.
User UI · Laptop Dev queries the decision log. Each decision carries its why — the constraint it was weighted against — not just an outcome line.
Three brains accumulate knowledge. Project Master Brain (PMB) is different — it thinks with that knowledge. PMB reads Research signals, Brand judgment constraints, and Org capacity. It remembers every project the company has ever run — succeeded, failed, killed, held, never started. When a Research signal arrives, PMB is where deliberation happens: does this become a project? which one? at what cost?
PMB is also where Priya talks to her own portfolio. Monday morning, before the standup, she opens the PMB dashboard. Four projects running. Two held. One signal arrived Friday. PMB has already drafted what to do about it. Priya decides.
"≤ 4 active B2C projects." "Kill criteria: ROI < 2× at midpoint." "B2C : B2B = 60 : 40 of design capacity." The decision rules that don't change month to month.
"Pricing experiments fail 70% of the time." "Rebrands take 9 months, not 6." "Held projects revive 15% of the time." Lessons survived the projects that produced them.
Pointers to all live Project Brains. State per project. Signals just arrived from Research. Proposals waiting for Priya. Dead projects don't live here — they're lessons in Learning, files in Archive.
Coming next Chapter 03 — "A Project Is Born" — shows the PMB dashboard and Priya's Monday-morning conversation with it.
Each Project Brain carries the same three-layer anatomy as every other Brain in the system. Identity holds the project's intent, success criteria, and open questions. Learning accumulates what the team has tried, decided, and discarded. Now holds this sprint's live state. At project close, each layer has a different fate — Identity archives, Learning graduates upward to Org Brain, Now evaporates.
What this project is for, who it's for, what "done" means. Written at kickoff. Referenced before every decision. Compresses into the Archive at project close.
What the team has tried, what users responded to, which tradeoffs keep recurring. Pre-loaded at kickoff with relevant fragments from Research Brain. Graduates back into Org Brain when the project ends.
What's in flight right now. Open PRs, yesterday's Slack decisions, today's blockers, the current Intent Match score. The team sees this; the Archive never keeps it. Evaporates at project close.
Every team member queries the Brain in a different mode — for daily work, for onboarding, for hypothesis-sparring, for catching up after a week off. Domain experts get their time back; everyone else gets to a useful depth without booking a meeting.
Concept Identity lives longest and travels farthest — that's what makes intent survive people.
Monday, 08:31 AM. Before the standup, before the inbox, Priya opens the Project Master Brain dashboard. Friday, while she was offline, Research Brain detected an 18% checkout-abandonment spike in the millennials segment. Project Master Brain has already done the work: matched the signal against three past projects, checked Brand Brain · B2C constraints, read Org capacity. A draft proposal is waiting.
This is Cross-Brain Reasoning — the strategic AX pattern unique to enterprise scale. PMB joins Research × Brand × Org × past-project history into a single thinking surface. Priya doesn't ask it; she discusses with it. The decision stays with her. The thinking is shared.
User UI · Laptop · Generative Canvas Priya's MacBook · Monday 08:31. The dashboard isn't a fixed report — layout shifts to surface what matters this morning: new signal, 4 running, 2 held, 1 PMB-drafted proposal. The Ambient feed shows what other Brains surfaced overnight, without anyone asking.
Boundaries didn't dissolve. They moved. Priya drafts three UI directions without opening Figma — not as a spec, as a starting conversation. The Design Lead (Maya) picks one and adds human resonance. The Engineering Lead (Dev) owns architecture and tradeoffs. All three read the same Identity Layer.
Write PRD. Hand off to design. Wait two sprints.
Prompts the Brain for three UI directions against the project's Identity Layer. Maya gets a brief, not a backlog ticket.
Translate PRD to visual. Defend every pixel in review.
Picks one of Priya's drafts. Adds taste, emotion, edge-case judgment. Taste, edge-case judgment, and the call-it moments stay with Maya.
Write boilerplate. Fight state management. Ship late.
First-pass code is agent-drafted against the Brain. Dev owns architecture, performance, and tradeoffs.
The Stakeholder Autonomy Matrix isn't one pattern. It's three patterns plugged together — same way Silent Resolution in P1 was a composition of four. Patterns are vocabulary, not endings.
stakeholder_autonomy_matrix = autonomy_dial // 4 stages: Suggest → Confirm → Notify → Auto × role_boundary // who owns this domain × disclosure_resolution // who is allowed to see this output intent_shared_workflow = identity_layer_read // shared intent, one source × knowledge_at_query // past decisions surface in editor × ambient_absorption // today's draft becomes tomorrow's memory archive_replay = schema_versioning // old Brain still readable × queryable_meeting_archive × intent_fidelity_log // scrub the score timeline
Backstage · AX composition Pattern libraries that flatten everything into one list always run out. A composable vocabulary doesn't — when a new workflow appears, you compose the pattern, you don't add a row.
Agents blurred role boundaries. Priya can produce UI mocks; Maya can ship code. So who has final authority in each domain? The Stakeholder Autonomy Matrix extends the Autonomy Dial to team relationships: five domains × five roles.
Maya owns UI-detail Auto. Priya owns business-direction Auto. Dev has Override on feasibility. Leena holds Veto on brand. Taste, political judgment, and technical intuition stay human — agents stop at Suggest until earned otherwise.
Business direction is mine to set. Brand alignment I confirm with Leena before it ships.
UI detail decisions: I own them. Brand alignment I confirm with Leena.
Architecture is mine to call. On feasibility, I can override anyone — including the agent.
Brand is veto-grade. An agent can suggest, never ship — and cross-brand calls are mine to make.
B2B business direction is mine. Cross-brand decisions I confirm with Leena and Priya.
Concept · Stakeholder Autonomy The Autonomy Dial at the home scales cleanly to the team — same stages (Suggest → Confirm → Notify → Auto), plus Override and Veto for senior roles, applied per role-domain pair.
Wednesday afternoon, daily sync. Priya is presenting Mobile Nav Study findings. On the side panel, an Ambient feed updates in the background — not because anyone asked, but because the Brains are listening to each other. The B2B team's PMB just finalized auth-flow consolidation; the pattern matches the bottleneck in front of Priya. Research Brain confirmed a 3rd consecutive weekly NPS drop. Studio Helsinki (vendor) reports Phase 2 is on track. CFO has just been auto-shared the week's Intent Match summary.
Disclosure Dial in 4 directions: cross-team (B2C ↔ B2B sanitized) · AI-to-AI (vendor's agent, status-only) · internal-to-exec (auto-summary) · human-in-the-loop (anything sensitive stops at Suggest). Visibility without meetings — the org-scale version of P1's Silent Resolution.
Slack, Jira, Linear, Figma, Drive — every team already pays for these. the Brain isn't a tool replacement. It's a memory layer that subscribes to all of them, vendor-neutral. Like Matter for smart-home, OAuth for auth, OpenTelemetry for tracing — federation is the precondition for a layer above tools, not a service from one vendor.
Backstage · Federation Pick no winner among Slack / Jira / Figma / Drive / Calendar. the Brain reads them all, writes the synthesis to one Brain, and lets the team change tools without losing memory. That's the federation argument.
First, what is Intent? On Monday morning, when Priya kicked off this project, she wrote one sentence into Project Brain · Identity Layer: what this project is for. That sentence is the Intent. For example: "An exploratory checkout that welcomes the user pausing to think."
Intent Match (%) is a number from 0 to 100 that tells you how close the things you're building right now (Figma screens, copy, code) are to that one Monday-morning sentence. 100 % means the work matches the Intent. 0 % means you're building something completely different.
Friday afternoon, Sprint 3. The Brain surfaces this week's number: 62 %. Last week it was 82 %. The Brain has already traced the cause: the design has slid from "exploratory UI" toward "efficient shortest path." The trigger was the performance-optimization call Dev made during a Tuesday standup. Nobody did anything wrong. Dev's call was a sound engineering call on its own. The distance from Intent just widened. The drift is visible. The Brain doesn't decide — it hands Priya a single question: "Accept this drift, or rewrite the Intent?"
User UI · Laptop Priya's dashboard, Friday afternoon. Only the team sees this — the Brain monitors its own output's distance from Intent. 62 % this sprint. The drift is visible.
Same Identity Layer intent. Same sprint artifacts. The Rule Engine compares, scores, and traces the drop back to the standup that caused it. the Brain surfaces one question — never makes the call.
# Rule Engine · Intent Match intent_vec = embed(identity_layer.intent_statement) artifact_vecs = [embed(a) for a in sprint.shipping_artifacts] # figma, copy, commits fidelity_now = mean(cosine(intent_vec, a) for a in artifact_vecs) fidelity_prev = brain.read("atlas.intent_fidelity@last_sprint") delta = fidelity_now - fidelity_prev # trace the drop to the meeting that caused it trigger = meeting_brain.find_decision( correlated_with = artifact_diff(prev, now), within = "7 days" ) # surface a question, never a verdict if fidelity_now < 0.70 and delta < -0.10: surface_to(role="product_owner", surface="iphone", body={ "score": fidelity_now, "drift_from": identity_layer.intent_summary, "drift_to": describe_artifacts(now), "trigger": trigger.summary, "question": "Accept the drift, or update the Intent?" }) else: pass # silent — no notification
Backstage · Rule Engine Embedding model is vendor-swappable; the contract is the score and the trace, not the weights. The Rule Engine is small enough to read in one breath — the trust comes from seeing it.
The Autonomy Dial starts at Suggest for everything. Over twelve weeks the team lets the Brain do more — per domain, at different rates. By Month 3, one Brain, four relationships.
Reads the Identity Layer aloud. Suggests the week's sprint goal. Priya accepts, rewrites, or declines — all three are teaching signals.
Priya's UI-direction prompts produce drafts worth reviewing. Maya accepts one in three. The Learning Layer remembers which taste won.
Meeting Brain captures everything. Decision log, not transcript. Dev stops taking notes. Priya stops being the forwarding address.
Research = Auto (omakase). Brand = Confirm (Leena's domain). Code = Notify (Dev's ceiling). Intent = Suggest — that one stays human.
When a project closes, the three Project Brain layers don't share a fate. The Now Layer evaporates — yesterday's blockers, today's draft slack, the in-flight sprint state. The Learning Layer dissolves into the Org Brain — patterns survive, the project doesn't. The Identity Layer compresses into the Archive — queryable years later when someone asks "why did we do that?"
Concept · Animation Auto-advancing loop. Now evaporates · Learning dissolves into the Org Brain · Identity compresses into the Archive. The next project starts on higher ground because the Org Brain is now richer.
When the project closes, the three Project Brain layers don't share a fate. The Now Layer evaporates. The Learning Layer merges into the Org Brain — so the next project starts on higher ground. The Identity Layer compresses into the the Archive, queryable years later.
Priya reviews the archive-readiness cascade on his phone — the two approvals only he can give before the Brain seals the project.
User UI · iPhone Friday 4:48 PM. Two cascaded items wait for Priya's tap; nine others completed automatically.
Learning Returns Home is the default — the Learning Layer quietly dissolves into the Org Brain. Archive Replay is the second door: Identity Layer + decision log, scrubbed on a timeline. Rewind to Sprint 3, and that moment's Figma, Meeting Brain, Intent Match score, and team composition come back.
Audit teams can replay the moment a typography decision was made — meeting transcript, Figma diff, and Intent Match score intact.
User UI · Laptop An auditor's MacBook, three years on. Scrub to any sprint — the archive reconstitutes that moment's Figma, meeting, metrics, and team composition.
Project Brains graduate to the archive. Five years on, an audit team can replay them. That's only possible if the schema is versioned and migrations are deterministic.
{
"schema": "identity_layer/3.2",
"intent": { "statement": "...", "rationale": "..." },
"success_criteria": [ ... ],
"open_questions": [ ... ],
"stakeholders": [ ... ],
"_migrated_from": "identity_layer/2.x"
}
migrate(v2 → v3):
intent.rationale = derive(
v2.intent_statement,
v2.kickoff_meeting_brain
)
success_criteria = v2.success_metrics
open_questions = v2.unknowns ∪ v2.risks
preserve("_migrated_from")
Backstage · Schema versioning Forward compatibility isn't a feature — it's the precondition for an archive that's worth keeping. Every schema bump ships with a deterministic migration; every Brain remembers where it came from.
A 200-lawyer firm. A senior partner who's been at the firm 22 years. An associate three months in. The same problem the Brain was built for: knowledge dies with people, drift compounds across matters, and onboarding is a re-derivation tax. Same four Brains. Same three layers. Different vocabulary.
Generalization · proof, not appendix A senior partner is on 32 hours of meetings a week — exactly like Leena. An associate three months in is doing exactly Priya's "let me check" walk. Same architecture, different vocabulary. The Brain doesn't care which world it ground itself into; it cares that intent survives the people who held it.
"I want to be the steward of the intent — not the answering machine for the team.
"I don't want to defend every pixel in review. I want my taste to be the differentiator, not the bottleneck.
"I'm not the brand's memory. I have a job. I shouldn't be the only person who knows why we picked this color three years ago.
"Day one, I want a real project to start — not a week of catching up on what I missed before I joined.
Methodology · simulated-archetype interviews informed by published HCI research on AI-assisted enterprise authoring (2024–2026), composite voices from 12 conversations with PO / Design Lead / Eng Lead / Brand Owner contacts during portfolio research. Quotes are composite, not direct attributions. The point of this section is the shape of the need, not the specific n.
Methodology · simulated · architectural projection Modeled with one team across a 12-week sprint cadence, against a baseline of tool-isolated workflow (Slack + Jira + Figma + Drive, no shared Brain layer). None of these are deployment numbers. They reflect what the architecture predicts when the four-Brain memory layer is in place. Real-world figures depend on team size, tool stack, and Brain maturity. The point isn't the exact figure — it's the shape: a memory layer compresses coordination cost by an order of magnitude.
Senior portfolios show reasoning, not just output. Six decisions where the call wasn't obvious — three kept, three cut.
Arman (incoming Product Owner, 28) kicks off Project Orion Monday morning. The the Archive surfaces itself: "Similar intent lineage detected — Project Master Brain (2026). Reference?"
Priya is long gone. His intent, Leena's original brand reasoning, the Sprint 3 drift, the two approvals he gave on his way out — all queryable. Arman doesn't start from zero. He starts from Priya's Month 3.
User UI · iPad Arman's iPad, Monday 9:04 AM. The Archive opens beside Project Orion — same intent lineage, three years older, waiting.
Trustworthy designers name their own edges. Four questions the Brain doesn't yet answer.
When intent survives
the people who wrote it,
the project still knows what it's for.
Project 04 · Context Grammar at work
Project 04 is Project 01's Context Grammar scaled to the company. One Brain, three layers, split into four specialists. Disclosure per role. Autonomy per domain. An archive that graduates when the project ends.
At home, the timescale is twenty years — Aoi inherits the family's memory at 35. At work, the timescale is three years — Arman inherits a product's intent on day one. One design language. Home and office. Outliving both.
