Supply chains are generating more signals, more complexity, and more interconnected decisions than ever before. Yet many organizations still struggle to bridge the gap between insight and action, especially when decisions must be coordinated across suppliers, production, inventory, logistics, and customer commitments in near real time.
That challenge is driving the next evolution of enterprise AI.
Kinaxis®, a global leader in supply chain orchestration, and NVIDIA are exploring how long-running AI agents could help organizations continuously optimize and adapt supply chain decisions at scale.
This work builds on the integration of NVIDIA cuOpt, an open-source GPU-accelerated optimization engine, into Kinaxis Maestro®, helping organizations reduce optimization runtimes from hours to minutes in complex scenarios. That acceleration opens the door to long-running AI agents capable of evaluating alternatives, running repeated optimization cycles, and continuously refining decisions as conditions change in near real-time. Through ongoing collaboration around optimization and agentic AI workflows, the two companies are exploring how orchestration platforms and GPU-accelerated computing can work together to help organizations solve increasingly complex operational challenges.
From AI assistants to long-running agents
Most enterprise AI systems today operate as assistants—answering questions, generating recommendations, or summarizing information.
But supply chains don’t operate in isolated moments.
Decisions around sourcing, inventory, production, and transportation evolve continuously as conditions change across the network. Solving one issue often creates tradeoffs somewhere else.
That is why Kinaxis and NVIDIA are exploring long-running AI agents that can iteratively plan, reason, optimize, and refine recommendations until user-defined objectives are achieved, rather than stopping after a single recommendation.
These agents could:
• Create and evaluate planning scenarios
• Re-run optimization models as conditions evolve
• Investigate sourcing or inventory tradeoffs
• Explain why decisions were made
• Refine recommendations collaboratively with planners
• Learn from prior scenarios and optimization outcomes to improve future recommendations
Over time, these agents could leverage reusable skills that help them perform specialized supply chain tasks more effectively. Rather than starting from scratch each time, agents could apply proven approaches for activities such as inventory optimization, sourcing analysis, and network tradeoff evaluation, helping accelerate decision-making while maintaining governance and transparency.
These skills act as reusable capabilities that allow agents to perform specific planning and optimization tasks consistently and at scale.
The goal is not autonomous AI operating without oversight. It is governed AI working within the orchestration and approval workflows already embedded within Kinaxis Maestro.
Why supply chain orchestration is well-suited for agentic AI
Many enterprises remain cautious about agentic AI because of governance and operational risk concerns. Supply chain orchestration offers a different model.
Kinaxis Maestro was built around scenario-based workflows, allowing organizations to safely test, compare, and validate changes before operationalizing them. That creates an environment where AI agents can explore decisions within controlled scenarios while still maintaining human oversight.
As supply chains become more dynamic, this combination of orchestration, optimization, and explainability could become increasingly important for managing complexity at scale.
Exploring agentic workflows and skills with NVIDIA cuOpt
As part of this exploration, Kinaxis is prototyping long-running agent capabilities designed to help planners manage increasingly complex optimization challenges using NVIDIA open technologies such as Nemotron open models, NemoClaw, cuOpt library, and cuOpt agent skills running on NVIDIA GPU infrastructure.
What makes this exploration particularly compelling is the convergence of technologies that historically operated separately: GPU acceleration, advanced optimization libraries, AI-driven workflows, and deep supply chain domain expertise. Together, these capabilities could allow organizations to evaluate operational tradeoffs and orchestrate decisions at a speed and scale that were previously impractical in real-world supply chain environments.
In practice, changing business conditions constantly create new planning challenges. A supplier disruption, demand spike, transportation constraint, or sudden inventory imbalance can require organizations to evaluate thousands of possible responses. Long-running agents paired with GPU-accelerated optimization have the potential to automate much of this work, continuously assessing alternatives and recommending actions as conditions evolve.
By leveraging optimization skills built around NVIDIA cuOpt, agents can invoke sophisticated optimization capabilities as part of larger decision-making workflows.
For example, one objective assigned to an agent could be to reduce inventory costs while maintaining current service levels.
The agent could then:
• Create multiple scenarios
• Re-run optimization models
• Test sourcing alternatives
• Evaluate downstream impacts
• Compare tradeoffs across the network
• Present a refined recommendation with supporting rationale
Another potential use case involves inventory balancing and just-in-time strategies, where a long-running agent continuously adjusts plans based on changing operational constraints.
Over time, these workflows could evolve into reusable AI-driven operational skills capable of handling recurring supply chain challenges, from sourcing analysis to optimization tuning and network tradeoff analysis.
Just as importantly, these capabilities have the potential to make sophisticated supply chain decision-making accessible to a much broader set of users. By combining agentic workflows with optimization skills, organizations could empower planners, analysts, and business users to solve complex operational problems without requiring deep expertise in supply chain modeling or optimization techniques.
The result could be faster responses, broader participation in decision-making, and better-informed decisions across the enterprise.
While still early-stage, these explorations point toward a future where supply chain teams can collaborate with AI agents to manage increasingly complex decisions with greater speed and transparency. Much of this work remains exploratory, but it represents an important look at the art of the possible as AI, optimization, and orchestration continue to converge.
Building toward the next era of supply chain AI
As supply chains become more volatile and interconnected, organizations will need more than static workflows or isolated AI assistants. They will need orchestration platforms capable of coordinating decisions, data, systems, and execution across the enterprise.
That is the future Kinaxis and NVIDIA are helping shape together: combining optimization, agentic AI, reusable skills, and operational orchestration to help organizations move from insights and decisions to coordinated action across the supply chain. As these technologies continue to evolve, they have the potential to transform how supply chain decisions are modeled, optimized, orchestrated, and executed across the enterprise.
We’re excited to continue exploring what’s possible with NVIDIA and sharing more as this work evolves.
