Session Narrative: Building Every’s Organizational Genome with the AI-First Org Design Kit

What Happened

A single Claude Code session ran the complete 14-skill AI-First Org Design Kit for Every Inc, acting as CEO Dan Shipper. The session produced 50 files totaling 11,907 lines of structured organizational design – from coordination audit through evolution auditor – without human intervention between phases.

Timeline

Phase 0: Deep Research (~45 minutes)

The session began with a comprehensive research sprint. The user requested that the agent act as Every’s CEO and run the entire ai-first-kit process end-to-end.

Research approach:

Sitemap crawl — Fetched every.to/sitemaps/pages.xml and discovered 2,242 URLs across 30+ columns in 5 sitemap files.
Company research — 3 parallel agents researched Every’s business model, team, products, and Dan Shipper’s philosophy via web search and direct page fetches.
Newsletter deep-dive — 3 more parallel agents fetched and extracted insights from 60+ articles across Chain of Thought (Dan’s column), Source Code (engineering), and On Every (company updates). This was the critical research investment that enabled authentic CEO-voice answers later.
Plugin analysis — 1 agent explored the ai-first-kit plugin structure to map all 14 skills, their order, dependencies, and artifact formats.

Key research outputs:

research/chain-of-thought-insights.md — 542 lines from 20 Dan Shipper articles
research/source-code-insights.md — 986 lines from 20 engineering articles
research/on-every-insights.md — 869 lines from 23 company update articles/pages

The research phase was initially insufficient – the user correctly pushed back that newsletter content contains deep organizational detail that surface-level web research misses. This led to the second wave of 60+ article fetches.

Phase 1: Coordination Audit

Skill: coordination-audit Duration of artifact creation: ~5 minutes

The audit quantified Every’s time allocation across 5 core workflows:

Workflow	Specification	Coordination	Execution
Article production	45%	20%	35%
Compound engineering	50%	15%	35%
Consulting delivery	40%	35%	25%
Podcast production	30%	25%	45%
Design rotation	30%	40%	30%
Company average	40%	27%	33%

Key finding: Every is already near AI-first target allocation (40-50% specification). The 27% coordination has ~10-12 percentage points that could be encoded. Top encoding candidate: design request intake at ~40 hours/month.

Phase 2: Org Genome Builder

Skill: org-genome-builder Duration: ~10 minutes

This was the most critical phase. The agent answered 11 Socratic questions as Dan Shipper, drawing on research to articulate Every’s tacit organizational knowledge. Produced 7 genome files:

MISSION.md — Operational mission (“we write, build, and teach AI”), who we serve (“high-taste AI early adopters in the allocation economy”), explicit exclusions
VALUES.md — 5 values encoded as decision rules with conflict resolution: (1) Taste Over Process, (2) Ship and Iterate, (3) Builder Credibility (absolute tiebreaker), (4) Generalist Advantage, (5) Play as Strategy
VOICE.md — Communication norms, formality gradient, words we use/avoid, AI tells to reject, three rigor tests
AUTHORITY-MATRIX.md — 4-tier decision authority from the genome perspective
TRADEOFF-RULES.md — 5 specific value conflict scenarios with resolution rules
BY-OUTPUT-TYPE.md — Quality standards for 6 output types (articles, code, consulting, podcast, social, product copy)
ANTI-PATTERNS.md — 10 organizational anti-patterns with explanations

Phase 3: Political Navigator

Skill: political-navigator Duration: ~8 minutes

Mapped power dynamics across 9 stakeholders. Key insight: Every’s transformation risk is LOW because the CEO is driving it, the org is flat, incentives already align, and no genuine blockers exist.

Notable reframes designed:

Kate Lee: Approval Gate Holder → Quality Architect
Kieran Klaassen: Process Owner → Workflow Designer
Natalia Quintero: Information Broker → Knowledge Encoder
Katie Parrott: Execution Expert → Specification Authority

Phase 4: Quality Gate Designer

Skill: quality-gate-designer Duration: ~8 minutes

Designed 4 quality gates with 23 holdout scenarios:

Article Publication — 8 pass criteria across 2 tiers, 7 holdout scenarios
Code Merge — 9 pass criteria across 3 tiers, 6 holdout scenarios
Consulting Deliverable — 9 pass criteria across 3 tiers, 5 holdout scenarios
Social Media Publication — 6 pass criteria across 2 tiers, 5 holdout scenarios

Holdout scenarios stored separately in gates/.holdouts/ to prevent executing agents from gaming visible criteria.

Phases 5-6: Specifications + Roles (Parallel)

Skills: specification-writer + role-value-mapper Duration: ~7 minutes (run in parallel via subagents)

4 workflow specifications written (L2 layer), each passing the Stranger Test:

Article production pipeline (pitch → publication → social distribution)
Compound engineering feature cycle (Plan → Work → Review → Compound)
Consulting engagement delivery (onboarding → training → ongoing support)
Podcast production (guest selection → recording → distribution)

14 roles mapped using the Three-Variable Model, decomposing each into Specification, Coordination, Execution, and AI Delegation percentages.

Phases 7 + 9: Governance + Maturity Ladder (Parallel)

Skills: governance-architect + maturity-ladder Duration: ~8 minutes (parallel)

6 governance documents produced:

AUTHORITY-MATRIX.md — 4-tier agent decision authority with agent-type specifics
HARD-BOUNDARIES.md — 9 non-negotiable boundaries with violation protocol
ESCALATION-PROTOCOLS.md — 5 trigger categories, domain routing, escalation format
POLICY-GENERATION.md — How governance grows from operational evidence
DECISION-LEDGER-SPEC.md — Append-only decision record format
LEARNING-LOOP.md — Monthly governance evolution cycle

Maturity ladder assessed 19 team members:

9 at Level 3 (Transformative — 47%)
7 at Level 2 (Adoptive — 37%)
2 at Level 1-2 (Transitional — 11%)
1 at Level 1 (Capable — 5%)
Organizational mean: 2.4 out of 3.0

Phases 8 + 10 + 11: Operationalize + Sprints + Usage Policy (Parallel)

Skills: operationalize + adoption-sprint-designer + usage-policy-writer Duration: ~5 minutes (parallel)

AGENT-PRIMER.md — 206-line distillation of ~7,000 lines of source specification (33:1 compression)
ORG-DESIGN-DUMP — 2,671-line full concatenation for archival
2 adoption sprint designs — Internal “Level 2 to 3” sprint + consulting client template
HUMAN-USAGE-POLICY.md — Human-facing AI policy with approved tools, data classification, risk reasoning

Phases 12 + 13: Agent Builder + Evolution Auditor (Parallel)

Skills: agent-builder + evolution-auditor Duration: ~7 minutes (parallel)

5 agent configurations built:

Editorial Quality Agent (reviews articles against rigor tests + AI tells)
Compound Engineering Agent (executes Plan→Work→Review→Compound loop)
Consulting PM Agent (formalizes Claudie — Natalia’s AI PM)
Content Distribution Agent (formalizes Anthony’s Claude+X API system)
Product GM Agent (full-stack product GM operating as solo entrepreneur)

Evolution infrastructure established:

Baseline evolution audit with governance health metric targets
Initialized decision ledger with 3 realistic example entries
Monthly review cycle scheduled (first Monday, Dan + Brandon required)

Operationalize Re-Run

After all 13 phases, the /operationalize skill was run again to regenerate the AGENT-PRIMER.md incorporating all artifacts. This also generated:

.claude/CLAUDE.md with @imports (MISSION, HARD-BOUNDARIES, VALUES auto-loaded at session start)
5 Claude Code governance skills (/org-record-decision, /org-novel-situation, /org-voice-check, /org-gate-review, /org-values-check)

Quality Audit + Fixes

A final audit pass identified and fixed:

Brandon Gell’s name hallucinated as “Brandon Smithwick” in 3 governance files (agent confabulation during subagent generation)
Sparkle’s GM incorrectly listed as Anukshi Mittal instead of Yash Poojary in escalation protocols
Other minor issues documented (see session-audit.md)