Architecture

Three tiers. Clear ownership. No confusion.

The AOS operates on a three-tiered strategic architecture that separates human intent, strategic decomposition, and machine execution. Each tier has distinct ownership, authority boundaries, and data ownership rules. No agent may operate outside its assigned tier.

AOS Architecture

The Three Tiers

TIER 1

Owner: Human

Vision & Goals

Where are we going?

Qualitative future state and quantitative measurable outcomes. Human-owned — agents read, never write. This is the strategic direction that everything else derives from.

VisionQualitative, unbounded, never 'complete.' The aspirational north star.

GoalQuantitative, measurable, time-aware. The scoreboard for the vision.

TIER 2

Owner: Human + Agent_8

Strategic Decomposition

How do we get there?

Sub-Goals and Loops — continuous engines with lifecycle phases. Agent_8 creates Loops, defines phase gates, and generates Strategic Intents. Human approves Sub-Goal changes.

Sub-GoalLoad-bearing pillar supporting a Goal. Requires human approval to change.

LoopContinuous engine with lifecycle phases: Build, Run, Optimise, Deprecate. Never binary done/not-done. Generates Strategic Intents.

TIER 3

Owner: Machine (via Exec_8)

Machine Execution

What happens next?

Bounded, time-horizoned, metric-driven executable directives. Exec_8 receives Strategic Intents from Tier 2 and executes them with full evidence logging. Agents here never invent their own mandates.

Strategic IntentTemporary, bounded directive linked to a parent Loop. When an SI completes, the Loop's phase may shift and a new SI is generated.

Loop Lifecycle

Continuous engines, not projects.

Loops are the fundamental operating unit of Tier 2. They shift through lifecycle phases, generating Strategic Intents at each stage. A Loop is never “done” — it is deprecated when it has served its purpose.

Loop Lifecycle

Build

Design and construct

→

Run

Live and operating

→

Optimise

Tune and compound

→

Deprecate

Graceful wind-down

Loops are never “done.” They are deprecated, not completed.

Build

Designing and constructing the engine. Establishing patterns, testing hypotheses.

Run

The engine is live and operating. Producing outputs, generating evidence.

Optimise

Tuning throughput, reducing friction, compounding institutional memory.

Deprecate

The engine has served its purpose. Graceful wind-down, knowledge preserved.

Failure Modes

What happens when the architecture breaks.

These are named failure modes with detection criteria. If you see these patterns in an organisation running AOS, something has gone wrong at the architecture level.

Tier collapse

An execution agent starts making strategic decisions, collapsing Tier 3 into Tier 2. The system loses its separation of concerns.

Orphan SIs

Strategic Intents created without parent Loops, breaking the decomposition chain. Work happens without strategic context.

Authority creep

Agents gradually expanding their scope beyond assigned tier. Canon governance prevents this by defining explicit boundaries.

Phase gate bypass

Loop transitions without evidence-based evaluation. The system loses its learning mechanism.