Why We Are Building the First Supply Chain Model

Gartner’s 2026 supply chain technology trends list has eight entries. Four are about autonomy: agentic AI, collaborative multiagent systems, physical AI and polyfunctional robots. Two are about specialisation, and one of those is domain-specific language models, which Gartner describes as models “trained or fine-tuned for specialized supply chain use cases, delivering greater accuracy, reliability and compliance than general-purpose AI models.”
‍

I read that and thought: the analysts are finally describing the thing we have been building.
‍

A few weeks ago I announced that Ship Angel is building the world’s first supply chain model. I want to explain what that means, why now, and why it matters to anyone who moves freight for a living.

‍

General models are brilliant and ignorant

The frontier models are extraordinary. I use them all day. But ask one what a detention and demurrage free time clause means in practice at a specific terminal for a specific carrier, and you get a confident paragraph that a good ops manager would laugh at.
‍

That is not the labs’ fault. General intelligence is not the same thing as knowing that a major carrier’s Rotterdam transhipment invoices tend to double-bill THC under a different charge code, or that a particular forwarder buries the fuel surcharge in a footnote of its rate sheet.
‍

That knowledge lives in the heads of experienced people and in the daily flow of real shipments, real invoices and real exceptions inside real enterprises. It does not live on the public internet, and that is why no general model has it.

‍

What Maddy has been accumulating

Maddy, our supply chain super-agent, is in production at more than twenty-five enterprise shippers. Every day it processes rate sheets, audits invoices, manages spot procurement and executes bookings. Every one of those actions has an outcome. The invoice was overcharged or it was not. The spot rate beat the contract or it did not. The booking rolled or it sailed.
‍

That is the raw material of a supply chain model: not documents, but outcomes.
‍

Over the last year Maddy has developed skills, memory, proactive monitoring and scenario planning. The next step is to close the loop and train on what actually happened: capture the prediction, log the outcome, score the model against it, and let it improve on freight the way the general models improve on code.
‍

The general model is the engine. The supply chain model is the driver who has done the route ten thousand times.

‍

Why this is a shipper’s story, not a vendor’s

I want to be careful here. This is not about Ship Angel owning your data. Your data stays yours. What compounds is the skill, anonymised across the network, in the same way every ocean carrier benefited from the container standard even though nobody owned it.
‍

The shipper who runs on a supply chain model gets three things the shipper on a general model never will.

• Judgment on freight, not just fluency about it.
• Compounding accuracy, because every audit and every booking across the network teaches the next one.
• A head start that widens every month, because the model is learning from live operations while a legacy vendor is still writing a requirements document.

‍

The 1956 moment

In 1956 the first container ship sailed from Newark to Houston. The container did not win because it was clever. It won because a handful of operators adopted a standard early and the economics became impossible to ignore. Everyone else followed.
‍

Domain-specific models in supply chain will play out the same way. Gartner has put them on the trend list. The frontier labs are competing to power them. The only open question is which shippers will be early.
‍

If you want to be one of them, try Ship Angel for free at www.shipangel.com/demo.