Synedrix OS separates accountability from execution.

A conventional hierarchy is asked to assign authority, coordinate work, approve decisions, and monitor execution all at once. Each of those jobs gets harder to hold together as a company adds software, tools, and automated workflows, because every addition puts more coordination load on a structure built for people working at human speed.

Most of that load is coordination, not judgment, and separating the two lets each structure do what it is good at.

An AI-native company keeps its human accountability layer (Board, CEO, executives, managers, specialists) and adds a second structure underneath it for how work actually moves: the work graph. Objectives enter the graph, get decomposed into tasks, resolve to capabilities, and return as proposed actions or completed outcomes. The company still has people in leadership roles. What changes is that machines take on much of the coordination between them.

Read as an organisation, the platform's 12 agents form four functional planes: Orchestration, Control, Intelligence, and Memory. Orchestration asks what needs to happen. Control asks who is allowed to do it, and when a human must intervene. Intelligence asks what the data shows, now and next. Memory asks what the organisation already knows, and what it should learn.

This grouping is a projection for organisational reading, not a technical classification. Every agent's family and layer stay authoritative: 12 agents, three technical families, 12 layers, none of which moves. Where this view places an agent beside a plane its technical family doesn't match, the page says so.

The platform coordinates execution. It does not decide strategy, and it does not hold accountability.

Humans keep strategy and purpose, capital allocation, business outcomes, policy and risk appetite, and every high-impact legal, financial, or personnel commitment. They keep the judgment calls where a situation is ambiguous or the stakes are high: a sensitive customer escalation, a contested decision, a policy exception. Humans own outcomes and authority; Synedrix OS coordinates execution.

What increasingly runs through the OS instead is the coordination itself: breaking a goal into tasks, routing them to the right capability, enforcing policy automatically, tracking execution, aggregating information for a decision, and keeping an organisational memory of what happened and why.

Synedrix OS does not dictate a company's org chart. It defines a narrower, machine-readable layer underneath that chart: human control roles, the interface the work graph uses at runtime to know which human authority a decision needs.

The platform does not say “your CFO must approve this.” It says an action requires human_risk_owner authority, and the business running Synedrix OS maps that requirement onto whichever real person or title actually holds it. One person often holds several control roles at once, the same way a CFO might carry risk and compliance authority together.

Six OS-level roles make up this runtime accountability interface. They name the kinds of human authority the work graph can route to when judgment is required; they are neither a staffing model nor the company's full org chart.

Business functions and Domain Packs relate many-to-many, not one-to-one. One business function can draw on several Domain Packs and capability modules together, and one Domain Pack can support several functions at once. Order Operations, for example, might combine a Commerce pack, an Ops pack, and a Finance pack, plus a company-specific extension for its own escalation workflow. No single pack owns the function, and none of them belongs exclusively to it.

A Business Instance is where this becomes concrete: the company-specific policies, systems, and thresholds that turn a generic Domain Pack into a working deployment for one organisation.

Two placements need a note, because both put an agent beside a plane its technical classification does not match.

Audit & Lineage sits beside the control agents because it reads the same governance trail they act on, but it is strictly read-only: it is never invoked by another agent, and it never appears in another agent's tool list. It is grouped here for the organisational reading, not because its technical classification changed.

General Chat sits in the Memory Plane because it is the interface employees use to reach organisational knowledge, and every one of its reads passes through the Memory Access Broker. Its actual technical classification is intelligence, at Layer 4. Its broader role as a single conversational entry point for enterprise work is target state, not current behaviour, pending further governance conditions being met.