Human-Governed Multi-Agent Orchestration for On-Demand Supply Chains
On-demand supply chains must turn incomplete, sometimes ambiguous customer requests into commitments over money, production capacity and delivery. This paper develops a human-governed multi-agent reference architecture for that problem. It assigns interpretation and task delegation to specialised AI agents, binds execution to server-enforced authority and authoritative business state, and assigns policy ownership and exception adjudication to humans. The design links each delegation to an executable permission, a transaction outcome and a recoverable evidence trail. The industrial case is Nima 3D Print, a live Berlin-based platform for custom manufacturing and white-label fulfilment. Its internal technical documentation describes multi-channel order handling, shared capacity, payment and carrier integration, and several bounded AI services. The operator separately reports live inter-agent delegation, customer self-service including refunds, risk-based escalation and an internal ledger. A worked white-label order and a worked refund show how to apply the design, including under partial failure. The paper derives testable safety properties and gives a comparative evaluation protocol. Evidence of comparative operational performance remains unavailable.
Authors
- Aydin Habibi Javanbakht (ORCID: https://orcid.org/0009-0005-1371-8120)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24404489
- Primary Topic
- Multi-Agent Systems and Negotiation
- Type
- preprint