Blue Yonder used its ICON 2026 user conference in San Diego this week to confirm an aggressive pivot toward agentic AI. CEO Duncan Angove told attendees the publisher, which now positions itself as the “AI company for supply chain”, is concentrating its roadmap entirely on a Cognitive Solutions portfolio. A wave of new specialized agents was unveiled, alongside a Nvidia-backed model factory designed to retrain Blue Yonder’s own supply chain models.
A wider net of specialized agents
The new agents span planograms, category management, production planning, scheduling and integrated business planning (IBP). Blue Yonder also introduced custom agents aimed at lifting warehouse team productivity. Angove said the technology compresses implementation cycles for client deployments and, longer term, opens the door to running large portions of supply chain processes autonomously.
Beyond the headline announcements, the strategic message was about scope: Blue Yonder no longer talks about embedding AI here and there. The whole portfolio is being repackaged around agentic capabilities. Customers were told the cognitive layer will sit across planning, execution and warehouse operations, with hand-off points wherever a human still needs to stay in the loop.
Hybrid model approach, Nvidia under the hood
The publisher unveiled its Model Training Factory, built on Nvidia’s Nemotron foundation model. Working alongside human operators, the factory is intended to tune and test highly specialized solutions for supply chain workflows that span multiple steps. “Supply chain has always been a relevant application area for AI. Our research on the performance of agentic models in concrete warehouse and planning decisions has shown us exactly where advanced models hit their limits,” Angove said.
Blue Yonder’s resulting approach is hybrid. Frontier models are called when they are needed. Smaller, custom supply chain models, trained on synthetic data rather than client data, operate alongside them. That synthetic-data stance is also a sales talking point in regulated industries where customer data leakage into model training remains a sticking point.
Pressure on the agentic stack
The ICON announcements raise the competitive heat on Manhattan Associates, Kinaxis, o9 Solutions and SAP, all of which have positioned their own agentic AI strategies over the past year. Blue Yonder’s bet is that being purely focused on cognitive agents, rather than retrofitting them onto legacy planning suites, will translate into faster time-to-value and stickier deployments. The proof will come from customer reference numbers in the next two quarters.
