Blue Yonder and NVIDIA introduce specialised AI agents to transform supply chain management

Blue Yonder partners with NVIDIA to develop a Model Training Factory that creates domain-specific AI agents aimed at revolutionising warehouse operations and supply chain decision-making.

Blue Yonder has teamed up with NVIDIA to create what it calls a Model Training Factory, a system designed to produce specialised AI agents for supply chain operations rather than relying on broad, general-purpose assistants.

The software is built around NVIDIA’s Nemotron models and Ne...

Mo tools, and is intended to fine-tune, test and repeatedly improve domain-specific models for areas including warehousing, planning, transport, merchandising and network operations. Blue Yonder says the approach is meant to deliver a repeatable pipeline that can generate agents capable of supporting human staff inside tightly defined workflows, with models assessed before deployment and then refined against set standards.

The first use cases are centred on warehouse management, where operators often have to make fast decisions as conditions shift during a shift. Blue Yonder says the initial agents will tackle allocation shortages, inventory exceptions, due-time urgency and stock movements across yards and receiving trailers. In practice, that means dealing with the kind of day-to-day disruptions that can force managers to reshuffle priorities quickly, from late deliveries to equipment problems.

According to Blue Yonder, the training system combines larger models for broader reasoning with smaller, custom-tuned ones for more specific and cost-efficient tasks. The company is using synthetic data rather than customer information, alongside NVIDIA AI Enterprise for infrastructure management and software support. That, it says, should allow the models to be built and tested without exposing client data.

The move reflects a wider shift in enterprise software away from generic AI chat tools and towards narrower agents designed to act within business processes. Supply chain software, in particular, lends itself to that model because decisions often have to be made in real time across warehouses, transport networks and stores. Blue Yonder argues that a specialist agent can weigh more trade-offs, more quickly, than a human operator usually can.

The company says the factory has been built around its own operational data, workflow logic and decades of supply chain expertise, creating what it describes as a reusable method for generating further models across its platform. Later versions are expected to extend beyond warehouses into the wider Blue Yonder product set.

Duncan Angove, Blue Yonder’s chief executive, said the company’s research showed the limits of large general-purpose models when applied to day-to-day supply chain decisions. He said the partnership with NVIDIA was aimed at building “owned intelligence” rather than depending on “rented intelligence”, arguing that the models should be trained on the workflows and decision logic that actually run a warehouse or planning system.

NVIDIA has cast the project as part of the next stage of enterprise AI, where specialised agents are embedded directly into operational software. Azita Martin, vice-president and general manager for retail and consumer packaged goods at NVIDIA, said the next phase of AI in supply chains requires “specialised, affordable and accurate” domain-trained agents that can operate inside business workflows. She said Blue Yonder was using Nemotron, the NeMo Agent Toolkit and NVIDIA AI Enterprise to build models with proprietary supply chain data.

Blue Yonder says the first models should enter customer production later this year through its Cognitive Solutions offering. The company says it serves more than 3,000 retailers, manufacturers and logistics providers.

Source: Noah Wire Services