An AI agent reasoning inside the business space
The agent does not merely answer questions about a documented process. It interprets the relationships in the semantic model and reasons about what is currently possible.
Current situation: Authorized for Shipment
For an MQLS in this state, three state-compatible actions can exist:
- Accept Finally — positive final completion,
- Reject — negative final completion,
- Publish — publish the current state without finalizing the MQLS.
The agent evaluates before acting
It checks the current SituationBook, the authority invariant, user authorization, approval status and application consistency. If data or bindings are incomplete, the result is UNDETERMINED — not true and not false.
In that situation the agent must stop, resolve the missing context and rebuild the situation. The model therefore supports fail-closed behavior instead of letting uncertainty silently become a business decision.
Agent and runtime stay separate
The SemanticMind Core does not execute domain commands itself. The agent selects a valid business action on the basis of the model and passes the command to the responsible domain runtime. The runtime validates and executes the transition and publishes the resulting event.
The loop continues
After the runtime event, the agent rebuilds the situation and reasons again from the new business reality.