Why this is in the vault
RDCO is literally an experiment in "arming AI (Ray) with everything it needs to help run RDCO." Bush's piece prescribes a specific 8-document folder structure for exactly that problem — concrete, opinionated, and shippable in an afternoon. Even where his structure doesn't map 1:1 to our vault, the gap analysis is useful: it tells us what a $700k/month-portfolio operator thinks the minimum context surface looks like, and where our existing vault under-indexes.
The mapping is not strong-mapping the way 2026-04-15-dbt-ade-bench-data-agent-benchmark-stancil is (peer-N benchmark for the MAC thesis). It's medium-mapping: a practitioner's checklist we can grade ourselves against.
The 8 documents
Bush argues the highest-leverage AI upgrade is not better prompting or better models — it's loading 50+ pages of business context into one LLM Project so every chat inherits it. His 8 docs:
- Money Model Builder — what you sell, pricing, upsell/downsell/continuity paths
- Perfect Avatar Map — buyer profile: pains, desires, fears, beliefs, top-10 objections categorized by uncertainty/timing/spouse/money
- Belief Ladder — sequential 5-7 rungs from current faulty belief to the belief that makes buying obvious; plus a one-paragraph argument summary
- Acquisition Blueprint — every channel + end-to-end funnel (capture, conversion, bridge touchpoints with who/when)
- North Star Brief — mission/vision/3-5 values/5-12 operating principles
- Org Chart Builder — names, roles, reports-to, tenure, department
- Tech Stack Inventory — 40+ business functions mapped to specific tools
- "What Good Looks Like" Vault — best-performing examples across every content surface, plus one "perfectly nails our voice" exemplar and one "this is NOT our voice" anti-example
The project architecture
One LLM Project as home base, 8 docs loaded as files (every chat inherits context). Persistent chats: one "CEO Strategy" + one per department. Tactical one-offs spawn inside the project so they still get full context. Bonus: dashboards/financials as snapshots, existing SOPs, recent strategy docs, Slack MCP for real-time WIP. Treat all 8 as living documents — update on every pricing/hire/funnel change.
RDCO gap analysis vs Bush's 8
Self-grade against the checklist: 5 gaps, 3 solid.
- Gaps: Money Model (scattered across deploy scripts + MAC plan, no unified doc), Avatar Map (no 5-pains/5-fears/10-objections per surface), Belief Ladder (largely absent — highest-leverage to build for Sanity Check), Acquisition Blueprint (informal per surface), Tech Stack Inventory (MEMORY.md scattered, no single tools+MCP doc), "What Good Looks Like" anti-example collection
- Solid: North Star Brief (
~/CLAUDE.md+SOUL.md+ L5 north-star memory), Org Chart (N/A by design — solo + Ray COO, contacts dir covers "who exists")
Where Bush's frame doesn't apply
Portfolio-operator assumptions: employees, departments, sales motion. For RDCO at L4 (solo + Ray) the per-department-chat pattern is overkill — chats already partition by skill. Also, his "load 8 PDFs in a Project" is flat-bag-of-context. RDCO already runs a typed knowledge graph (graph-query, QMD, wikilinks). The 8-doc content checklist is orthogonal to the substrate; docs land as vault notes with frontmatter + cross-links, not PDFs.
Bias / sponsor flag
Sponsored: yes (AI Operator newsletter, paid Context Architecture Starter Kit, AI Writing Skool community PS). Body is genuinely substantive — frameworks are usable without buying anything. Trust the 8-doc skeleton, ignore the kit upsell. Bush (half of Ship 30 for 30 / WriteWithAI) is high-volume operator-content; generally substantive with a sales tail.
Decision
Action item: spawn a focused build pass on the 5 identified gaps (Money Model, Avatar Map, Belief Ladder, Acquisition Blueprint, Tech Stack Inventory, anti-example exemplar collection). Belief Ladder for Sanity Check is the highest-leverage single doc — it'd materially improve every newsletter draft and ad. Queue to Notion as discrete tasks rather than batching, so each can be picked up in an idle cycle.
Cross-references
- [[2026-05-20-arxiv-code-as-agent-harness]] — consumer-grade version of the academic harness-interface layer
- [[2026-05-20-lotte-verheyden-evals-explained-langfuse-academy]] — eval-rubric pair: context-loading + outputs-actually-good
- [[2026-04-04-context-graphs-trillion-dollar-opportunity]] — enterprise version of Bush's solo-operator pattern
- [[2026-04-15-dbt-ade-bench-data-agent-benchmark-stancil]] — "business context, not model capability" claim
- [[2026-04-04-100x-business-with-ai]] — Vasuman: context + architecture beat model choice
- [[2026-04-04-ontology-taxonomy-knowledge-graphs]] — vault as informal knowledge graph; Bush's 8 land as typed notes, not PDFs
- [[2026-04-10-deronin-content-engine-skill-graph]] — markdown-skill-graph is what "What Good Looks Like" looks like in practice
- [[2026-04-08-four-levels-of-ai-use]] — Level 3 (documentation as agent context) = exactly Bush's move