Concept explanations
Explore at your own pace Select a concept to discover its explanation.
Each card explains a concept, why it matters and the distinctions to preserve. You can explore the two perspectives in whichever order suits you.
Governance constraints The requirements that frame how the system operates, including security, confidentiality, compliance, responsibilities and limits of authority. They specify what every decision and action must respect.
Why does it matter? These constraints make the boundaries of authorized behavior explicit and prevent even a powerful technology from bypassing the obligations of the framework.
Keep the distinction A governance constraint does not, on its own, constitute permission to act.
SMBC-GAF© principles The ten guiding principles establish the foundations of the framework: the central role of the business domain, declared business logic, governance by design, separation of responsibilities, protection of sensitive data, explainability and gradual adoption.
Why does it matter? They preserve architectural consistency as technologies and business domains evolve. Their four reading groups create no normative hierarchy.
Keep the distinction The principles are more than optional recommendations; all retain their normative scope.
Framework objectives They express the strategic aims of SMBC-GAF©: reduce structural complexity, place the business domain at the center, govern interpretation and decisions, make results explainable and enable gradual adoption.
Why does it matter? They guide architectural choices toward lasting value, beyond the adoption of AI technologies alone.
Keep the distinction An objective describes what one seeks to achieve; it is neither a measured result nor permission to execute.
Expected outcomes These are the observable and measurable effects sought through application of the framework, such as operational explainability, verifiable traceability and controlled business actions.
Why does it matter? They make it possible to assess whether objectives have actually been met and whether the architecture delivers the intended value.
Keep the distinction An expected outcome is not evidence of an achieved result. Its achievement must be observed and verified.
Business ontology The ontology defines the concepts, relationships and vocabulary of the business domain. It provides the governed foundation of meaning on which interpretation and decision depend. Business rules and policies are distinct governed artifacts linked to the ontology.
Why does it matter? The system can reason from consistent business representations, rather than from implicit rules scattered across code, screens or instructions to AI.
Keep the distinction The ontology and rules are the responsible sources of business meaning and logic; transactional persistence retains the truth of the actual business state. Sensitive data are never passed directly to cognitive reasoning.
Interpret the intention Understand and qualify a request or situation using governed business meaning. AI may help interpret natural language, identify ambiguity and prepare a structured representation of the intention.
Why does it matter? For example, this allows a naturally worded appointment request to be understood without treating it as an immediate transactional command. In SMBC-GAF©, this responsibility belongs to the Governed Decision Engine, even though it is not drawn here.
Keep the distinction An interaction agent can collect or rephrase a request without becoming the governed decision maker. Interpreting is neither deciding nor executing. This example illustrates a possibility, rather than proving a specific implementation in a vertical solution.
Produce a justified decision Evaluate options using the interpreted intention, ontology, rules and applicable constraints. Cognitive resources, including AI when relevant, may contribute to a candidate decision and its justification.
Why does it matter? Mandatory deterministic validations and applicable qualifications distinguish an admissible decision, a rejected output or a situation requiring clarification. The Governed Decision Engine carries this responsibility, remains explainable and traceable, and never performs the transaction directly.
Keep the distinction A plausible AI output is not automatically an admissible governed decision. An admissible decision does not, on its own, create authority to act.
Business agent The business agent is an orchestration component that coordinates application capabilities and activities based on governed decisions. Depending on context, it may use human, deterministic, AI-based or hybrid resources.
Why does it matter? It brings flexibility to coordination while respecting the applicable controls, confirmations and authorizations.
Keep the distinction The business agent is neither the conversational interaction agent nor a sovereign decision-making authority; it does not decide new rules or directly modify business state outside the intended mechanisms.
Orchestrate a governed action Coordinate and prepare the operations needed for an action, respecting a governed decision, dependencies, rules and the authority to act that is currently exercisable. Cognitive resources, including AI, may contribute to this coordination.
Why does it matter? Orchestration provides a controlled transition from decision to execution: preparing actions, checking conditions, obtaining necessary confirmations and allowing rejection or no action.
Keep the distinction Using AI in orchestration creates no new authority. The two stars in this zone indicate two perspectives, component and responsibility, rather than two guaranteed separate invocations.
Execute a governed transaction Actually perform a business operation through authorized transactional mechanisms, applying required integrity checks and validations and producing a traceable execution result.
Why does it matter? This is where real effects on business state are produced in a controlled, observable way. Rejection or no action can also be a compliant result, depending on the case.
Keep the distinction Cognitive reasoning never directly performs a transactional mutation; execution does not become an autonomous decision maker.
APIs and application services APIs and access contracts define available operations, their inputs and outputs, and the rights and conditions for use. Application services expose authorized business capabilities, including governed queries and commands.
Why does it matter? This separation allows real deterministic and transactional functions to be used without directly exposing the decision core or sensitive state to consumers or cognitive models.
Keep the distinction API ≠ transactional service: the API governs access and the service performs the exposed capability. The decision engine cannot directly invoke a transactional service or access sensitive data.
End-to-end governance Governance, responsibilities, confidentiality, authorizations, explainability and traceability apply across all zones of the architecture.
Why does it matter? Every decision, action or non-action must be traceable to responsible sources, applicable conditions of authority and required records; autonomy to act never transfers the authority to decide to AI.
Keep the distinction Governance is neither a fifth phase nor a final technical box. A PASS check, evidence or an admissible decision does not, on its own, grant permission to act.
Close card · Escape
