Skip to main content

How do agents orchestrate — and what instruments verify the work?

AGENT & INSTRUMENT DIAGRAM: PROMPT DECK
════════════════════════════════════════════════════════════

AGENTS (Yang — who applies force)
─────────────────────────────────────────────────────────
User (Human) ········ Fills grid, reflects, returns weekly, shares, decides priorities
Deck Engine (System) · Persists state, highlights gaps, takes snapshots, serves templates
AI Coach (AI, V3) ··· Suggests questions for gaps, coaches through dimensions

INSTRUMENTS (Yin — what verifies and rewards)
─────────────────────────────────────────────────────────
Gap Highlighter ······ Visualizes empty cells as strategy, not failure
Snapshot Comparator ·· Shows grid evolution over time (this week vs last)
Mode Lens ············ Toggles Pep Talk (inner) / Pitch (outer) perspective
Scale Selector ······· Zooms Personal / Venture / Industry / Portfolio
Kill Date Counter ···· Time pressure on validation milestones

PROTOCOLS (data + value + decision flows)
─────────────────────────────────────────────────────────

┌──────────────────────────────────────────────┐
│ THE FILL LOOP │
│ │
│ FILL ──→ GAPS ──→ QUESTIONS ──→ │
│ │ │ │ │
│ │ Gap User │
│ │ Highlighter reflects │
│ │ │ │ │
│ ──→ RESEARCH ──→ UPDATE ──→ COMPARE │
│ │ │ │ │
│ Domain Deck Snapshot │
│ Content Engine Comparator │
│ │ │ │ │
│ └── FEEDBACK ┘ │ │
│ │ │ │
│ Grid sharpens Evolution visible │
└──────────────────────────────────────────────┘

FEEDBACK LOOPS
─────────────────────────────────────────────────────────

Loop 1: SESSION (per fill)
┌─ Fill cell → Gap Highlighter → See adjacent gap → Fill next → Until complete ┐
└──────────────────────────────────────────────────────────────────────────────┘

Loop 2: WEEKLY (return)
┌─ Return → Snapshot → Compare to previous → Update cells → Better grid ─┐
└────────────────────────────────────────────────────────────────────────┘

Loop 3: MODE (per review)
┌─ Pep Talk → Pitch → See inner/outer misalignment → Refine → Pep Talk ─┐
└────────────────────────────────────────────────────────────────────────┘

Loop 4: COMPOUND (monthly)
┌─ Multiple snapshots → Pattern recognition → Deeper insight → New five ─┐
│ Stale grid, no snapshots? → Decision: re-engage or archive │
└────────────────────────────────────────────────────────────────────────┘

HORIZON FLOW
─────────────────────────────────────────────────────────

H1: Prompt Deck (this PRD)
│ what matters, where gaps are, how priorities evolve

H2: Time + Mind
│ closes gap between intention (what grid says) and action (what happened)

H3: AI Coaching
agent helps fill what you can't see alone

════════════════════════════════════════════════════════════

Agent Roster

AgentTypeForce AppliedVerification
UserHumanFill, reflect, return, share, decide prioritiesGrid completion rate, D7 return, snapshot count
Deck EngineSystemPersist, highlight, snapshot, template, mode switchData integrity, load time, session continuity
AI CoachAI (V3)Suggest questions for gaps, coach through 5P dimensionsUser satisfaction, fill rate improvement

Instrument Registry

InstrumentMeasuresThresholdAction on FailOutcome Map Link
Gap HighlighterEmpty cells per gridVisual at any countHighlight grows more prominentEnables O3 (completion > 60%)
Snapshot ComparatorGrid delta between snapshotsChange in ≥3 cells per weekPrompt: "Nothing changed — still aligned?"Enables O2 (D7 return > 35%)
Mode LensPep Talk / Pitch alignment gapCells that differ between modesHighlight misalignment for reflectionEnables O3 (completion > 60%)
Scale SelectorActive scale levelAt least 1 scale usedDefault to Personal if unusedEnables O1 (10 users fill)
Kill Date CounterDays to kill date0 remainingForce decision: extend with evidence or stopGuards all outcomes

Feedback Loop Quality

LoopFrequencySensorActuatorLatency
SessionPer fillGap HighlighterUser fills next cellInstant
Weekly7 daysSnapshot ComparatorUser updates stale cellsMinutes
ModePer reviewMode LensUser refines inner/outer alignmentMinutes
CompoundMonthlyMultiple snapshotsUser recognises patterns, redefines fiveDays

Activation Sequence

From the Dependency Map — when does each agent come online?

PhaseWhat ResolvesAgent StateUser Experience
NowPrototype existsDeck Engine: read-only prototypeUser sees the grid, can't persist
Sprint 0 (10d)Auth + persistence + cell editor + modeDeck Engine: fills persistUser fills, grid survives sessions
Sprint 1 (5d)Card View + Scale + HighlightingDeck Engine: multi-pattern + lensesUser sees gaps, mobile works
Sprint 2 (5d)Snapshots + templates + shareDeck Engine: memory + distributionUser compares, shares, compounds
V3 (future)AI coaching integrationAI Coach: onlineAgent helps fill gaps user can't see

The instrument doesn't replace the human. Each version adds one capability layer. V1 structures. V2 adds memory. V3 adds voice. The human stays in the loop because the grid is meaningless without their judgment of what matters.

Gate

Before executing:

  • Every agent named — YES (3: User, Deck Engine, AI Coach)
  • Every instrument named — YES (5 with thresholds and outcome map links)
  • Feedback loops explicit — YES (4 loops at different frequencies)
  • Horizon data flows declared — YES (H1 → H2 → H3)
  • Activation sequence links dependency resolution to capability — YES (5 phases)

Context