Case Study: aswritten.ai builds aswritten.ai
How a solo founder uses organizational knowledge to run a multi-agent product company.
The challenge
aswritten.ai is built by one person — Scarlet Dame, a solo founder with 15 years of experience across narrative architecture, consulting, and AI tooling. The product is an MCP server, an extraction pipeline, a knowledge graph, and a growing ecosystem of workflows. The codebase spans Clojure, n8n, GitHub Actions, and RDF/SPARQL.
The problem: one person can’t hold all of that context in every session. Strategy decisions made in a fundraising conversation need to inform the next product session. Pricing validated during a sales call needs to show up when generating the business plan. Architecture decisions need to persist across branches and tools.
Without organizational knowledge, every AI session started from scratch. With it, every session starts from the organization’s perspective.
The solution
aswritten.ai uses its own product. Every strategy call, product decision, sales conversation, and architecture choice is saved as a memory. The extraction pipeline produces structured knowledge. The perspective loads automatically.
What this looks like in practice:
- A call with a prospect validates the $400/month price point. That memory is saved, extracted, and enters the perspective. The next session — generating a pricing page — cites the validation with full provenance.
- A weekly call with an advisor identifies a new market segment. That insight enters the graph. The sales playbook, business plan, and one-sheet all detect the drift and flag sections that need updating.
- A product architecture decision is made during a dev session. The next content session — writing docs — references that decision without the founder needing to re-explain it.
The results
1,477 nodes in the knowledge graph spanning Opportunity, Strategy, Product, Architecture, Organization, Proof, Templates, and Calibration domains.
Every AI session is grounded. The perspective loads at session start. Recommendations cite specific decisions from specific people. Gaps are flagged, not papered over.
Content generates from the work. The changelog, perspective summary, and graph health pages on this site are auto-generated from organizational knowledge. The one-sheet was generated and cited — 91% of claims traced to specific people and conversations.
Docs stay current. Registered documents are tracked against the knowledge graph. When the perspective shifts — a pricing model changes, a market thesis evolves — the system identifies which documents need updating and at what severity.
Multi-agent coordination. Multiple AI agents (Claude Code, Claude Desktop, n8n workflows) all load the same perspective. No context fragmentation. A decision made in one tool is available to every other tool.
The meta
The product produces its own marketing. The perspective steers the product that builds the perspective. The case study you’re reading was informed by the same organizational knowledge it describes.
This is what “think like your org” means at scale: not one person holding everything in their head, but a structured, version-controlled knowledge graph that makes every AI session as informed as the founder’s best day.