SemanticMind
← Case studies overview
Case study 4 · Original output

Business dialogue with an AI agent

The agent interprets the model, checks conditions and behaves carefully in undetermined situations.

Starting point

A user can interact with the MQLS model in natural language. The AI agent does not only answer questions about documented flows; it interprets relationships contained in the model.

Example questions

Using authorise-shipment as an example, questions may include:

  • Under which conditions may a shipment be authorised?
  • Which information and evidence are required?
  • Which role holds responsibility?
  • Which events trigger the transition?
  • Which subsequent situations are possible?
  • Which rules prevent authorisation?
  • What would be the effect of changing approval rules?

Answers can be traced back to underlying semantic elements. The result is not an isolated AI reply, but a business interpretation of the model.

NoteThis projection is a real, unaltered AI output. It was generated by GPT-5.6 Sol based on the semantic model and demonstrates that the method produces working, inspectable results — not only a concept, but a verifiable artefact.

Decision space of the agent

For an MQLS in state Authorized for Shipment, compatible transitions include final acceptance, rejection and publication of updates — each with its own authorisation and guards. The agent must respect that space rather than invent free-form process steps.

Handling undetermined situations

Where the model does not determine a unique outcome, the agent should make the uncertainty explicit, name missing information or open approvals, and avoid silent assumptions. Controlled behaviour in undetermined situations is part of trustworthy use.

Takeaway

The quality of the dialogue depends on the quality of the semantic foundation. The agent becomes useful because meaning, rules and states are explicit — not because it improvises enterprise policy.

← Case studies overview