Fujitsu is testing a new multi-AI agent system that lets companies adjust supply chain decisions in real time without trading sensitive data. The effort seeks to speed cross-partner coordination as disruptions become more frequent and harder to predict.
How Real-Time Agent Coordination Works
Fujitsu has developed a collaborative multi-AI agent technology designed to help companies synchronize operational decisions even when they cannot share detailed data with one another. Rather than pooling proprietary information into a central platform, each company deploys its own autonomous agent that interprets local conditions, inventory positions, production schedules, transportation constraints, and transmits only high-level signals to other agents in the network.
Fujitsu’s system evaluates those signals using a global optimal control model that approximates the preferred conditions of partner companies without revealing what sits behind them. According to publicly available technical documentation, the model identifies a network-wide “best feasible response” by running continuous simulations of how each agent’s choices influence the rest of the chain. This setup enables real-time coordination during rapid changes such as sudden demand spikes, supplier delays, or logistics bottlenecks.
The technology is now entering field trials with Rohto Pharmaceutical and the Institute of Science Tokyo. Fujitsu says the tests will measure not only day-to-day planning efficiency but also how quickly multiple agents can converge on recovery actions following a disruption—an area where manual coordination often slows decision cycles across partner networks.
Securing Shared Decisions Without Shared Data
A central feature of the system is Fujitsu’s secure inter-agent gateway, which acts as a firewall between companies while enabling their AI agents to collaborate. The gateway allows each agent to access a distilled representation of partner models through techniques similar to knowledge distillation, a method in which a “student” model learns from several “teacher” models without receiving their underlying proprietary data.
This structure keeps sensitive operational information, such as production yields, demand forecasts, or cost structures, within company boundaries. At the same time, the gateway evaluates the behavior of every participating agent to detect anomalies, flag malicious actions, or prevent unintended data leakage. Publicly available research from Japan’s Cyber-Physical Systems initiatives indicates that similar distributed architectures are becoming increasingly important as companies adopt AI-driven decision tools but remain wary of exposing competitive insight.
Once deployed, the multi-agent system can run simulations at high frequency, updating its shared understanding of network conditions in near-real time. Fujitsu says this allows partners to shift production, redirect logistics flows, or adjust procurement decisions before bottlenecks materialize.
Two Components Driving Coordination
The platform is built on two core components. The first, global optimal control, estimates the operational preferences of partner firms using limited inputs and identifies a coordinated response that minimizes network-wide friction. This helps agents converge on decisions even when supply chain partners operate under different priorities or constraints.
The second component, the secure inter-agent gateway, governs how agents exchange insights. Instead of trading raw data, the agents share model-derived signals that capture essential constraints and changes. These signals are continuously simulated and refined, allowing the system to adapt to new conditions without revealing sensitive details about any one company.
Fujitsu expects the combination to support more resilient planning across multi-vendor networks, especially in sectors with frequent disruptions or long recovery timelines. Following the Rohto pilot, the company plans to expand testing to more complex manufacturing environments.
Where Distributed Intelligence May Quietly Redraw Boundaries
One emerging question is how autonomous agents will reshape the balance between collaboration and control inside global supply networks. As more companies adopt distributed AI systems, the point of advantage may shift from owning the most data to shaping the rules under which shared decisions are made. Public efforts such as Japan’s Cyber-Physical Systems initiatives, and similar frameworks in Europe aimed at secure industrial data spaces, suggest that governance models could soon matter as much as algorithmic capability. For companies testing agent-based coordination, the next step may be understanding not just how these systems perform, but how influence flows when decisions are made collectively rather than centrally.
