Skip to main content

How does value flow from PRD priority to shipped outcome — and where does time die?

Trigger

From Outcome Map: pipeline has prioritised work, an engineering agent needs to start a session and ship. Currently agents spend 5-10 minutes re-exploring what happened before doing any work.

VALUE STREAM MAP: AGENT SESSION LIFECYCLE
════════════════════════════════════════════════════════════

[PRIORITY TABLE UPDATED]


┌─────────────────────┐
│ 1. DISPATCH │ C/T: 2min Wait: 0-∞ %C&A: 30%
│ Human reads table, │ (Manual: read table, open terminal,
│ opens terminal, │ decide which team, which agent)
│ starts session │
└──────────┬──────────┘
│ ░░░░░░░░░░░░░░░░░░░░░░░░░ (wait: hours-days, human availability)

┌─────────────────────┐
│ 2. CONTEXT RECOVERY │ C/T: 5-10min Wait: 0 %C&A: 40%
│ Agent reads codebase,│ (Re-explore: what was built, what's left,
│ past sessions, PRD │ where did we leave off, what failed)
└──────────┬──────────┘
│ ░░░ (wait: seconds, loading)

┌─────────────────────┐
│ 3. PLAN CREATION │ C/T: 10min Wait: 5min %C&A: 60%
│ Compose plan from │ (Template selection, task decomposition,
│ template + PRD │ estimation — often wrong template chosen)
└──────────┬──────────┘
│ ░░░░░ (wait: human review of plan)

┌─────────────────────┐
│ 4. EXECUTION │ C/T: 60-120min Wait: 15min %C&A: 70%
│ Agent builds, │ (Code, test, iterate. Blocked by missing
│ tests, iterates │ context from other teams — archaeology)
└──────────┬──────────┘
│ ░░░░░░░░░░░░░░░ (wait: blocked on other teams, no signal)

┌─────────────────────┐
│ 5. STATUS REPORT │ C/T: 5min Wait: 0 %C&A: 50%
│ Agent reports what │ (Often incomplete: "done" without evidence,
│ was built │ no commissioning link, no measurement)
└──────────┬──────────┘
│ ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ (wait: human reads, verifies)

┌─────────────────────┐
│ 6. COMMISSIONING │ C/T: 15min Wait: 1-3 days %C&A: 30%
│ Dream team verifies │ (Manual: navigate deployed URL, walk PRD
│ against PRD │ table, compare expectation vs reality)
└──────────┬──────────┘


[OUTCOME: Feature shipped and verified]


TIMELINE SUMMARY
─────────────────────────────────────────────────────────
Step C/T Wait %C&A (est.)
─────────────────────────────────────────────────────────
1. Dispatch 2min 4-48hrs 30%
2. Recovery 7.5min 0 40%
3. Plan 10min 5min 60%
4. Execution 90min 15min 70%
5. Status 5min 0 50%
6. Commission 15min 1-3 days 30%
─────────────────────────────────────────────────────────
TOTAL C/T: ~130min (2.2 hours of actual work)
TOTAL WAIT: ~1-4 days
LEAD TIME: 1-4 days per feature cycle
FLOW EFFICIENCY: 130min / (130min + ~2880min avg wait) = ~4.3%

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

Note: %C&A values are estimates based on observed sessions. Phase 1 (drmg vvfl audit --dry-run) will establish baselines.

Flow Efficiency

Current: ~4.3% (130 minutes of work in ~2880 minutes of lead time)

Target: above 15% (outcome map O3: 2x cycle time improvement requires eliminating wait between steps)

World-class: above 40%. We are at "worse than typical" (typical = 5-15%).

Bottlenecks (Ranked)

RankBottleneckWait TimeRoot CauseOutcome Map Link
#1Dispatch → Recovery4-48 hoursHuman availability gates every session startBlocks O1, O3
#2Context recovery5-10 min per sessionNo memory persistence — agent re-explores from zeroBlocks O1, F1
#3Status → Commissioning1-3 daysBuilder validates own work, dream team manualBlocks O3, O4
#4Execution blocks15 min avgNo block signalling — team waits silentlyBlocks O2, F2
#5Plan quality5 min reworkWrong template, no PRD-to-template mappingBlocks O3

Waste (Toyota Categories)

WasteWhereSeverity
WaitingDispatch queue, commissioning queueCritical — 95%+ of lead time
ReworkContext recovery (redo what was known last session)High — 5-10 min per session
Motion3 separate CLI entry points, context switchingMedium
TransportCode-merge communication between teamsMedium — archaeology, not messages
Over-processingManual commissioning (navigate, screenshot, compare)Medium
InventoryUnverified features pile up between commissioningLow-Medium

5 of 7 wastes present. Waiting dominates.

Improvement Targets

BottleneckPlatform FeatureTarget Improvement
#1 DispatchPriority dispatch (Concern 3, Feature 14)Automated: table update → message in under 2min
#2 RecoverySession bootstrap (Concern 1, Feature 3)Under 30s (from 5-10min) — loads memory + messages
#3 CommissioningCommissioning dispatch (Concern 3, Feature 19)Agent walks deployed URL, captures evidence
#4 Block signalBlock signalling (Concern 2, Feature 11)Under 2min broadcast, help arrives faster
#5 Plan qualityPRD→template mapping (Concern 3, Feature 15)Correct template auto-selected from PRD metadata

Gate

Before moving to Dependency Map:

  • All 6 steps documented with cycle time — YES (2, 7.5, 10, 90, 5, 15 minutes)
  • Wait time measured between steps — YES (0-48hrs, 0, 5min, 15min, 0, 1-3 days)
  • Flow efficiency calculated — YES (4.3% current, above 15% target)
  • Bottlenecks identified and ranked — YES (5 bottlenecks by wait impact)
  • Waste types categorized — YES (5 of 7 Toyota wastes mapped)
  • Improvement opportunities prioritized — YES (5 improvements linked to platform features)
  • %C&A flagged as estimates — YES (note present, baseline measurement planned)

Context