Architecture has axioms. Just as TOGAF has the Architecture Development Method and DDD has bounded contexts, A² has the Seven Separations™ — seven architectural constraints that define what it means to build agent-native enterprise systems.
Why Separations, Not Guidelines
Guidelines can be ignored when they're inconvenient. Axioms cannot. The Seven Separations™ are architectural constraints — if you violate them, you've made a structural decision that will compound into technical debt.
The Seven Separations™
Separation 1: Cognitive Concerns
Different reasoning types require different frameworks. Mixing them is the most common architectural mistake in AI systems.
There are three distinct cognitive modes in enterprise automation:
- Retrieval — fetching structured facts from a knowledge graph
- Inference — applying deterministic rules via DMN
- Judgment — synthesizing ambiguous inputs into decisions via LLM
These three modes have different latency profiles, different cost structures, different auditability requirements, and different failure modes. Architecting them in a single monolithic agent makes all three worse.
Separation 2: Domain Boundaries
Domain-Driven Design gave us bounded contexts. The agent-native version is stricter: each bounded context must have its own agent namespace, its own knowledge substrate partition, its own governance scope, and its own evolution rate.
The 7 Bounded Contexts of the A² framework (Strategy & Portfolio, Customer/Stakeholder, Knowledge & Content, Workflow & Decisions, Agent Operations, Risk & Compliance, Platform & Observability) are not organizational convenience — they are architectural constraints with anti-corruption layers enforcing their boundaries.
Separation 3: Process & Decision Logic
BPMN, DMN, and Agent Cognition are three distinct layers. This is the single most differentiating concept in agentic architecture, and the most frequently violated.
When an LLM agent is asked to "decide whether this invoice should be approved," it's being asked to perform a DMN function with a probabilistic tool. The result is approval decisions that can't be audited, can't be updated without redeploying the agent, and fail in unpredictable ways at the tails of the input distribution.
The fix: put the approval logic in a DMN table. The agent's job is to prepare the inputs for the table and handle the exceptions the table can't classify.
Separations 4–7
The remaining four separations cover Architectural Zones (20 zones, strict downward dependencies), Knowledge Types (Graph vs. Vector vs. Skill Library), Control & Execution (the Control Gate™ mandate), and Evolution Rates (Wardley-driven architectural decoupling).
Each is covered in depth on the Platform page. The full specification is available to x2machines clients.
The Test
For any architectural decision in your agentic system, ask: "Which separation does this violate?" If you can't answer that question, your architecture review is missing a framework.
Ready to apply this in your organization?
Book a free discovery session. We'll show you exactly how A² Agentic Architecture applies to your stack, your domain, and your automation targets.
Book a Discovery Session →