Disunified agents can create conflicting Supply Chain decisions across ERP, WMS and logistics systems
Independent agents can optimise locally rational objectives while producing an incoherent end-to-end order decision.
An Australian Supply Chain analysis argues that independently deployed AI agents in ERP, warehouse and freight systems can optimise incompatible objectives. An ERP agent may prioritise customer urgency or revenue, a warehouse agent may optimise labour and picking efficiency, while a customs or freight agent may hold the same shipment for compliance.
The proposed response is a control-tower architecture, potentially supported by Model Context Protocol, that gives agents shared context across order management, warehouse execution and logistics workflows.
The architecture problem, however, is not simply interoperability. It is decision arbitration. Value requires an explicit authority model defining which objective prevails when revenue, warehouse efficiency and compliance conflict. The main failure mode is an integration layer that gives every agent more data but still provides no rule for resolving contradictory recommendations.
The article is conceptual rather than empirical. It presents no implemented MCP workflow or measured operating results, so it should be read as an architecture hypothesis and failure-mode analysis rather than proof of deployment.