There is a chief architect at an enterprise software company, twenty-five years in, one of the small group who know everything about the product.1 The joke inside the company was that they should find a way to duplicate his brain. Every architecture question found its way to his calendar.2 The review board he ran met constantly, argued well, decided things, and the decisions evaporated: a summary here, a recording nobody reopens there, while everyone else worked from whatever version of the reasoning they remembered.2
Engineering organizations already govern by conversation.3 The board meets, weighs the options, rejects most of them for reasons, and moves on. Those conversations steer the humans who were in the room. They steer nobody else, and they certainly do not steer the AI tools the team codes with all day, which reason from the middle of their training data while the organization's actual judgment sits in a transcript.3
That gap is what the pilot closed. The company had months of review-board transcripts already recorded, and the transcripts became the bootstrap: decisions, reasoning, the rejected options and why they were worse, extracted into a perspective without asking the senior engineers for a single new hour.4 The conversations they were already having became the source. On the demo call it was the CTO who saw where it pointed and said it: this could become the review board.5
From then on governance ran through the perspective. A teammate asks about an architecture decision in the channel where they work and the answer arrives cited: who made the call, when, in their words, dissent preserved beside it.6 The coding agents draw on the same record, so the AI writing the code carries the reasoning of the board that governs the code.6 When the board changes its mind, nothing is overwritten. Old and new both stay, and the projection reads the shift. (uncommitted)
The first week told the story. The full invited team authenticated, usage split into recognizable patterns within days, and the heaviest user, running nearly half the tool calls himself, was the chief architect.7 The man whose brain the organization wanted to duplicate turned out to be the one doing the duplicating, one remembered decision at a time. Within a month the pilot converted to a paying subscription, and the follow-on request was quality assurance: the same record that steers the AIs, turned around to check the work against what the organization decided.8
The pilot customer's chief architect is the load-bearing technical lead of the engagement, and he sits at the center of a severe knowledge bottleneck. As documented in the May 1, 2026 taxonomy session, the company ships a very large SDK surface where documentation is sparse:
"only ~3-4 people (the ARB members) actually know what the product supports, what it doesn't, and how features compose."
Everyone else relies on direct access to those calendars, which is exactly the constraint the pilot was designed to relieve.
In the April 13, 2026 demo call, the chief architect outlined the daily pressure of running the company's architecture review board:
"Architecture Review Board — daily meeting, 8am, where everyone comes to the chief architect with architecture questions. His explicit goal: 'have an AI replace me at ARB.'"
The decisions made in those meetings were easily lost. A development manager on the team later extended the point: without structured memory the organization struggled to make decisions survive "next Tuesday," repeatedly re-litigating settled questions and conflating rejected options with decisions.
In a May 2026 strategy session, Scarlet framed the core thesis around how human teams naturally operate:
"Organizational governance and process governance — the ARB is the engineering version; the act generalizes to any leadership or governance process."
That conversational governance historically failed to reach the developer's daily environment. Delivered through MCP, the record becomes a productivity multiplier developers pull directly into existing workflows, so the AI tools they code with are steered by the organization's actual judgment.
During the April 13, 2026 demo call, the chief architect identified a massive shortcut for onboarding the account:
"the customer had ~6 months of ARB transcripts that could be ingested to bootstrap the perspective — a much faster path than formal extraction interviews."
The strategy was operationalized in May 2026 through a structured seeding pattern: a topic analysis across the corpus surfaced the domains, and the architect validated and annotated the extracted decisions and rejected options without a single new interview hour.
On the April 13, 2026 demo call, the customer's CTO immediately recognized where the technology pointed:
"his immediate reaction on the demo call: 'This could become ARB' — not just augmenting the daily architecture review board, but replacing the meeting itself as the primary way people get architectural answers."
He validated the board as the right starting point precisely because of its technical, finite audience: the ideal bounded environment to prove the system before expanding it.
In negotiating the statement of work on May 1, 2026, the chief architect defined success by how seamlessly knowledge would circulate:
"the stated criterion in the SOW is 'answers flowing to other team members' — the agent answers ARB-style questions with sufficient fidelity that team members rely on it before pinging the architect directly."
By June 25, 2026 this flow was fully active across an interface split where engineers query the perspective inside GitHub Copilot, so the AI writing the code carries the exact architectural reasoning and constraints established by the board.
The pilot's first-week metrics revealed an extraordinary adoption pattern in which the chief architect himself drove the system:
"the heaviest user in week one: 69 calls, approximately 47% of attributed activity. He runs the full read-write loop himself: 29 remembers, 22 perspective loads, 12 introspects, 4 cites."
The intense individual usage occurred within a highly active team trial: the full invited team authenticated in week one, and usage split into clear working patterns within days.
By late May 2026 the engagement reached its commercial milestone: the pilot customer signed as customer one on an Organization-tier subscription, with deployment options expected to grow the relationship further. The contract superseded the initial April 30, 2026 transaction, the first paid transaction by an external customer in Aswritten's history, when the chief architect entered a corporate card for a Team-tier trial. The relationship immediately extended into a next phase: deploying perspectives behind AI agents for client intake and internal quality assurance.