Reliable enterprise agents need transactions, meaning, memory and action — not another isolated chatbot.
Four layers with different responsibilities
The ERP remains the system of record for transactions. The ontology describes business meaning and relationships. Enterprise memory keeps decisions, evidence and learning. Agents use these layers to prepare or execute actions.
Confusing these responsibilities leads either to duplicated data or to agents that act without enough context.
The ERP is necessary but not sufficient
ERP data shows what was ordered, produced, delivered or invoiced. It rarely captures the complete reasoning, market signal or exception that shaped a decision.
Contextual AI should reference ERP truth while combining it with governed documents, relationships and decision history.
Agents need controlled action paths
An answer only becomes operational value when it supports a decision or action.
- read through the user’s identity and rights
- cite authoritative sources
- apply business rules and approval thresholds
- write back through governed tools
- record the action and its outcome for evaluation
Design for change
Models and frameworks will change. The organisation should keep ownership of business semantics, evaluations, critical connectors and operating evidence.
This makes the architecture more reversible and allows new use cases to reuse the same governed foundation.
CONNECT THE DECISIONS
Place this analysis inside the wider AI system
This architecture becomes operational when it combines enterprise memory, a clear business ontology, clear governance of the harness and a real path to production.
FREQUENTLY ASKED QUESTIONS
FAQ
Should the ontology replace the ERP data model?
No. It provides a business semantic layer across ERP and other sources while the ERP remains authoritative for its transactions.
Where should agent actions be recorded?
In governed operational logs linked to the business object, user identity, evidence, approval and result.
