How You Teach the Organization

Operate

How you teach the organization

The goal is not to remember everything. It is to stop being the organization's memory — and let memory become software you curate by exception.

The three learning paths

The organization does not learn from documentation alone. It learns through three paths, and you teach it through each.

1. Decisions become precedent. Every Action Queue approve, edit, or reject is captured as a decision record in organizational memory (GBrain) and surfaced as advisory precedent on future similar work. The next time the system encounters a similar question, it arrives with prior context — not a blank slate. You are not re-asked the same question; the question was already answered and the answer is on record.

This is the most leveraged thing you do: a single decision in the queue teaches the system for every future recurrence of that class of work.

2. Constraints and clarifications become reusable rules. When you correct an agent, clarify a boundary, or reject a proposal with a reason, that correction does not evaporate. It enters organizational memory as a constraint — a reusable rule that applies to the next agent that encounters the same situation. You answer once; the organization remembers.

The alternative — repeating the same correction each session — is burden that compounds. Constraints in memory are burden that shrinks.

3. The manual itself is the constitution agents read before acting. The Operating Manual is not a human reference document. It is the first input every agent reads before proposing, implementing, or reviewing. When the manual says how to classify a decision, agents classify decisions that way. When the manual names a reserved boundary, agents respect it. You teach the organization at scale by keeping the constitution accurate and current — not by supervising each agent individually.

What you do not do

You do not become the organization’s memory yourself. If you are explaining the same architectural decision for the third time, that is a system gap — a knowledge source that should be ingested but has not been. The remediation is not to explain again; it is to surface the gap and close it through a knowledge ingestion or constraint update.

You do not supervise individual agent reasoning. You govern the constitution the agents reason from.

◇ Target State

Agents arrive already knowing the decision history for the domain they are entering. Prior decisions are cited, not requested. Repeated corrections have become system constraints, not recurring conversations. The manual is the agent’s first input, not a document it was told to read.

An operator’s teaching surface is the Action Queue and the constraint log — not a chat session, not a prompt, not a recurring report.

▣ Current Reality

The teaching loop is built, not proven. GBrain stores decision precedent, but its influence grade is advisory (influence: informed, never grounded) — it cannot auto-ground actions, cannot self-modify, and does not currently reach live reasoning loops. Knowledge documents enter zero live capability reasoning sessions today; the wiring from knowledge ingest to agent context is the next build step, not yet operational.

Decision records are written to GBrain on each Action Queue decision. Precedent retrieval exists architecturally. But the loop from “you decided this” to “the next agent cites it” has not closed in production.

For the precise, dated grade of the knowledge and memory system, see Knowledge & Memory and the Honest Current State. The teaching loop is Phase 4 of the transformation roadmap — it depends on the runtime and product surfaces that precede it.

What this means now: the most important teaching you do today is through the Action Queue and by keeping the Operating Manual accurate. Each decision you record is a precedent the system will eventually cite. You are writing the organization’s memory in real time, even before the retrieval layer is operational.