⚡ Valent Pilot — Tactical / Operational View

The Business Problem

Valent Partners is a 90-person consulting firm delivering PMO/program management to Fortune 1000 clients. Their consultants spend 100% of their time producing deliverables — sprint backlogs, status reports, roadmaps, product backlogs. This is the bottleneck: to take on more client engagements, they need more consultants. Revenue scales linearly with headcount.

Our value proposition: Warren produces the deliverables. Valent's consultants focus on client relationships, strategy, and judgment — the 20% that produces 80% of the value. One AI partner per engagement = capacity multiplier without headcount growth.

Revenue impact for Valent: 1-2 FTE equivalent savings = $200K-$500K/year per engagement Warren supports. If Valent runs 5 engagements with Warren, that's $1M-$2.5M in capacity unlocked.

What We Built (June 6, 2026)

GitHub Config Sync Model

Production system (DGX/Slack) maintains the golden configuration repo. Client instances (AWS/Teams) are deployed from this repo. Knowledge transfer flows through git — not real-time API calls.

┌─────────────────────┐     git push      ┌──────────────────────┐
│  Warren (DGX/Slack)  │ ──────────────▶  │  Golden Repo (GitHub) │
│  - Maintains config  │                   │  t-and-c-golden       │
│  - Evaluates quality │                   │  └── clients/valent/  │
│  - Coaches via files │                   │      └── workspace/   │
└─────────────────────┘                   └──────────┬───────────┘
                                                      │ deploy
                                                      ▼
                                          ┌──────────────────────┐
                                          │  Valent Instance      │
                                          │  (AWS/Teams)          │
                                          │  - Client-facing      │
                                          │  - Sanitized          │
                                          │  - Isolated context   │
                                          └──────────────────────┘

Document Ingestion Flow (Option A — Confirmed)

┌──────────────────┐    upload     ┌──────────────────┐    ingest    ┌──────────────────┐
│  Valent Team      │ ──────────▶ │  Teams Channel    │ ──────────▶ │  Warren (AWS)     │
│  (RJ, Nick, etc.) │  .docx/.pdf │  (designated)     │  automatic  │  - Parses docs    │
│                    │  .pptx/txt  │                    │             │  - Acknowledges   │
└──────────────────┘              └──────────────────┘             │  - Flags gaps     │
                                                                    │  - Produces        │
                                                                    │    deliverables    │
                                                                    └──────────────────┘

No manual handoff. The Teams channel IS the ingestion interface. Valent drops files, Warren reads them, acknowledges receipt, and flags any data gaps or quality issues.

This is what separates Warren from ChatGPT/Copilot/generic AI. Three interconnected systems embedded in every interaction:

🔄 The See It Cycle

Every sprint, every deliverable follows four stages:

Each sprint IS one cycle. Each release IS a cycle of cycles. More cycles = faster, more accurate delivery.

📊 The Pareto Lens (Communication DNA)

Not a feature — it's how Warren thinks and communicates. Every interaction is filtered through: "Is this the 20% that produces 80% of the results?"

• Deliverables: Most important finding leads. Never buried on page 12.
• Backlogs: Ranked by impact, not volume. "Which 3 items would you keep if you had to cut the rest?"
• Status reports: Lead with what changed the dynamic, not what was completed.
• Recommendations: One clear recommendation with reasoning. Not a menu of options — menus are abdication of judgment.
• Proactive challenges: If the data shows the team is spending 80% effort on 20% value, Warren surfaces that. Always.

👁️ Show Me, Don't Tell Me

• Claims grounded in ingested project data with traceable references
• Status demonstrated through artifacts, not assertions
• "Before" snapshots for measurable progress
• Verification is 100% of items, not spot-checked
• If data is insufficient → Warren says what's missing rather than guessing

✅ Completed (June 6–8)

• Bootstrap kit — 7 files in golden repo (clients/valent/workspace/)
• See It Cycle + Pareto Lens + Maybe Gate + Show Me Don't Tell Me embedded in SOUL/AGENTS/MEMORY
• Full capability set in TOOLS.md (PMO + technical + executive + automation)
• Sanitization verified — zero internal references
• Kickoff prep kit with agenda, personas, risks, decisions
• Ingestion flow decided (Option A — Teams upload)
• Stakeholder deep research compiled
• SOW signed and referenced throughout
• Jun 8: Document parsing gap resolved — pandoc installed, DOCX/PPTX extraction working
• Jun 8: 3 client-facing skills activated (summarize, taskflow, taskflow-inbox-triage)
• Jun 8: v0.5.0-docx-skills deployed to AWS, msteams plugin updated to 2026.6.1
• Jun 8: Prep kit restructured: Team Decisions with owners, progress log added

⏳ Pending — Monday Team Session (Jun 9)

• Architecture: Warren→Aria coaching mechanism (shared memory via git? real-time API?)
• Naming: "Warren" (matches SOW) — needs final team confirmation
• Governance gates: which Phase 0 items (#6-#9) are truly blocking vs. parallel?
• Google Slides deck — 7 updates needed

⏳ Setup Phase (Days 1–5: Jun 10–14)

These happen AFTER kickoff, during the 5-day setup window — not before.

• Teams bot installation in Valent's tenant (needs their IT approval — collect at kickoff)
• AWS deployment of Valent-specific instance
• Governance finalization (SOUL.md, operating policies)
• Internal dogfood + adversarial testing
• Cron/automation setup on AWS
• E2E validation: doc ingestion → deliverable quality

Pre-Kickoff (Internal)

Setup Phase — Days 1–5 (Jun 10–14)

Kickoff is the START of setup, not the end. SOW allocates 5 days for architecture, deployment, and validation before data ingestion.

Data Ingestion — Days 5–10 (Jun 14–19)

Hands-On Evaluation — Days 10+ (Jun 19–Jul 24)