Generative AI technologies are becoming mainstream. They summarize reports, draft emails, and support analysts. Useful, but not fundamental.
What is now emerging is more consequential. Agentic AI systems—models that can pursue goals, retain context, and interact in structured ways—are beginning to move toward operational decision-making. Instead of serving as assistants, they can act as representatives: negotiating, reasoning, and adjusting decisions across organizational boundaries. A potential game-changer.
In our recent paper in the International Journal of Production Research, my co-authors Valeria Jannelli, Stefan Schoepf, Alexandra Brintrup, Matthias Bickel, and I explored what happens when AI does not just optimize decisions within firms, but actively coordinates decisions between them.
Supply chains are decentralized systems. Each actor has its own objectives, constraints, and information. Many coordination failures, like the bullwhip effect, are less about computation and more about misaligned human-to-human bias and negotiation.
We asked a simple question: Can AI agents negotiate their way toward better supply chain coordination?
A simulation study of agentic supply chain coordination
We built an experimental multi-tier inventory setting and replaced traditional replenishment rules with LLM-based agents. Each supply chain stage was represented by an agent with defined objectives and constraints. The agents exchanged structured messages, proposed replenishment quantities, justified their reasoning, and revised decisions based on feedback from others.
In essence, we allowed AI agents to conduct structured negotiations over operational decisions. We then compared their performance to conventional policies, focusing on system-level outcomes such as inventory behavior and demand amplification.
The agents in our study were able to converge toward coordinated decisions without a central controller. Through iterative bot-to-bot dialogue, they moderated extreme positions and moved toward solutions that balanced local and global objectives. In doing so, they reduced inefficiencies such as demand amplification compared to baseline rules.
Our study shows that AI agents can potentially automate, or at least augment, supply chain tasks that have traditionally been mostly human work.
Opportunities and risks of supply chain agents
If agentic systems enter real supply chains in the not-too-distant future, coordination could become continuous rather than episodic. Replenishment parameters, capacity allocations, and exception handling could be renegotiated dynamically through digital representatives operating at machine speed.
The opportunities are significant. Coordination costs could fall. Smaller firms could access advanced negotiation capabilities without the need for large, centralized systems. Decentralized supply chains could—at least in a theoretical simulation—coordinate more effectively without giving up autonomy.
But the risks are equally real. Objectives must be encoded carefully. Governance, transparency, and accountability become critical when autonomous agents negotiate operational commitments. Poorly designed incentives could scale misalignment rather than reduce it. Because there are significant financial opportunities, someone will surely start gaming autonomous AI agent systems.
Supply chains are essentially systems of negotiated commitments. If negotiation itself becomes programmable, the structure of coordination may change. We are still early, and substantial work remains. But the trajectory is clear: the AI agents are coming to supply chain management.
Read the full paper: Jannelli, V., Schoepf, S., Brintrup, A., Bickel, M., Netland, T. (2025) Agentic LLMs in the supply chain: Towards autonomous multi-agent consensus-seeking, International Journal of Production Research, https://doi.org/10.1080/00207543.2025.2604311
What's on your mind?