AI-Native Enterprise Architecture
Adapting SABSA and TOGAF for the age of autonomous agentic systems.
The Challenge
Traditional Enterprise Architecture (EA) frameworks like SABSA (Security) and TOGAF (Structure) assume a world of humans using tools.
But in an Agentic Work System, agents are not tools—they are cognitive participants. They make decisions, execute code, and influence strategy. Legacy frameworks break when “The Builder” is an AI.
The Strategy: Mapping the New Reality
I created a comprehensive mapping of the Agentic Work System to the SABSA 6x6 Matrix.
Key Divergence: The “Who” Column
In traditional SABSA, the “Who” at the Operational Layer is an IT Admin. In my architecture, it’s a swarm of Haiku agents.
- Implication: Trust models must shift from “Identity Verification” (User ID) to “Outcome Verification” (Drift Detection).
The “Trust Gradient”
I introduced the concept of Calibrated Autonomy. You don’t just “trust” an agent. You grant autonomy based on observability:
- Level 1 (Drafting): High Autonomy.
- Level 2 (Committing): Medium Autonomy (requires tests).
- Level 3 (Deploying): Low Autonomy (human gate).
Why This Matters
This isn’t just theory. This mapping allows large enterprises to adopt agentic workflows without throwing out their existing governance/risk/compliance (GRC) playbooks. It bridges the gap between the AI Frontier and the Boardroom.