You walked me up the ladder you sketched in February — execution → tactical planning → organizational memory. This is me drawing the picture back, so we can locate where your project ends and aswritten begins.
The hole has a name, a shape, and a product. Both can live in their lane.
Herself,
Scarlet
AI addendum · drafted by Scarlet's aswritten-grounded assistant
The loop and the memory
aswritten captures it: it pulls decisions out of meeting transcripts and AI sessions1, turns them into a versioned, cited record of who decided what and why2, and feeds that to your team's AIs.3
When you're orchestrating a team of AI agents — assigning roles, reviewing work, deciding how things should go — the same loop applies.4 The job is steering, whether the team is human or agentic. a partner agent-orchestration project builds the loop.5 aswritten is the memory the loop runs on.6
The convergence is structural
The convergence is structural. Every box in an agent-orchestration system produces a "system improvement" — but where does it live? (uncommitted) How does the next agent get it? (uncommitted) How do you know it isn't garbage? (uncommitted) That's the retrospective-and-context-gathering gap, and it's one thing, not three.7 It needs a substrate.6
During an April 16, 2026 reflection, Scarlet Dame established that conversational inputs are the primary engine of the system:
"Scarlet herself uses call transcripts and voice memos as her two primary inputs into aswritten."
This principle extends the decision made during the customer demo cycle, where the chief architect identified that six months of Architecture Review Board transcripts could bootstrap an organizational perspective faster than manual interviews. It formalizes the transition from treating transcripts as a 'favor' to recognizing them as a core asset for knowledge extraction.
In the foundational January 2025 check-in, Scarlet Dame defined the core nature of the substrate as a versioned, multi-player record:
"AI memory that is versionable, collaborative, branchable, so Git native, and narrative driven. Encodes style and conviction and other metrics."
This principle is extended by the April 2024 design decision to treat actors as first-class entities, ensuring that every claim in the graph carries explicit attribution for 'who said what.' Together, these nodes establish the system as a cited lineage of organizational thought rather than a flat database.
The March 19, 2026 MVP redefinition session shifted the product from a passive store to an active participant in AI workflows:
"The reframe positions expertise as invoked at the moment of need, mid-task."
This principle is operationalized by the ', aswritten' prompt suffix, a mode switch that forces an AI to ground its output in the versioned record. It moves the system beyond simple session initialization into a model where expertise is installed and cited dynamically as the team's AIs perform tasks.
In a May 2026 positioning session, Scarlet Dame articulated the second major use case for the platform: agentic steering.
"steering the entire agentic development process around it, which is functionally the same as ARB but often done by a smaller team or single person orchestrating agents instead of humans."
This extends the earlier 'perspective-driven development' insight from March 2026, where the goal shifted from humans directing AI to the perspective itself initiating work. It establishes that the same steering loop used for human governance applies equally to the orchestration of agentic teams.
The structural convergence between orchestration and memory was settled in a May 2026 strategy session:
"the partner project builds the loop; aswritten is the memory the loop runs on."
This decision clarifies the partnership with an agent-systems builder's open-source project, which was first noted in February 2026 as an effort to build agent expert systems. By defining the partner project as the 'loop' and aswritten as the 'substrate,' the organization established a clean separation between the orchestration engine and the narrative memory it requires to function.
Scarlet Dame established a hard positioning constraint in May 2026 to prevent the product from being miscategorized as a governance tool:
"In every use case, aswritten is the memory the steering runs on — never the tool that does the governing or orchestrating."
This principle validates the structural unifier that aswritten serves as the substrate for any steering loop, whether that loop is a human meeting or an agentic orchestration system. It ensures the product remains a foundational memory layer rather than competing with the orchestration 'loops' built elsewhere.
In a March 1, 2026 brain dump, Scarlet Dame identified that several disparate technical and context gaps were actually a single structural convergence:
"Triple store... would let pipeline pass references instead of strings... Convergence of OOM, compile reliability, and production-scale context."
This decision recognizes that the retrospective and context-gathering gaps are not separate issues but symptoms of the same substrate requirement. It moves the architecture toward a unified triple store to solve for scale and reliability simultaneously, rather than treating them as three distinct problems.