The conversation around agentic AI in supply chain has mostly focused on what agents can do. The Supreme Court’s unanimous ruling in Montgomery v. Caribe Transport II has reframed the question every enterprise shipper should also be asking: what did they do, when, and can you prove it?
For most supply chain teams evaluating AI investments, the questions center on capability. How many exceptions can the agent resolve without human intervention? What’s the efficiency gain? How much manual labor does it replace?
Those are the right questions for a proof of concept. They’re not sufficient for enterprise deployment.
Read our full breakdown of the ruling and what it means for carrier selection.
What the Montgomery decision changed
The Supreme Court ruled unanimously in Montgomery v. Caribe Transport II that federal law does not preempt state negligent-hiring claims against freight brokers. State common-law standards of ordinary care now apply to carrier selection. While brokers are the named defendants, the underlying reasoning extends further: if you chose the carrier, you own that choice.
Justice Kavanaugh’s concurrence is the most useful thing to read if you move freight for a living. He articulates a defensible safe harbor: brokers and shippers who selected carriers on a documented basis, and can demonstrate what they knew at the moment of selection, will generally be able to defend negligent-hiring suits. The standard going forward is reasonable care, and reasonable care means being able to answer four questions on demand, sometimes years after the fact: what did you know about this carrier’s safety record? What did you verify before you booked them? What did you do when conditions changed? And can you produce time-stamped, auditable records to support those decisions?
That question has always mattered operationally. Now it matters legally.
For supply chain teams deploying AI agents at scale, this isn’t a compliance footnote. It’s an architectural requirement. And it connects directly to how you build and govern your AI stack.
Three criteria for enterprise-grade agentic AI
For AI agents to be deployable in a modern enterprise, not just demonstrable in a proof of concept, they need to meet three criteria.
Explainability. The logic behind a decision should be traceable. If an agent reroutes a shipment or initiates carrier outreach, the reasoning needs to be visible. A black box that produces the right answer most of the time is not an enterprise-grade system.
Governability. Shippers set the guardrails. Agents execute within them. That’s not a limitation on what agents can do. It’s what makes autonomous execution trustworthy at scale. Customers determine which workflows are active, which carriers agents can contact, and at what thresholds escalation is required.
A system of record, not just a system of action. Every action timestamped. Every decision documented. Every automated communication logged and available through a transparent audit trail.
The record that protects you in a courtroom is the same record that runs your operation well. These aren’t separate concerns.
The context question that doesn’t get asked enough
There’s a related question that doesn’t get asked enough when teams are evaluating agentic AI: what does the agent actually know, and where does that knowledge come from?
An agent operating on shallow, poorly integrated data isn’t just less useful. It’s a liability. The quality of the decisions it makes, and the defensibility of those decisions, depends entirely on the quality of the context it’s drawing from.
project44’s agents operate on 11 years of network context: 1.5 billion shipments tracked annually, over 700 million logistics events processed daily, trillions of data points validated annually, and connectivity into more than 1 million logistics facilities and carriers. That context is what makes the intelligence actionable and what makes the audit trail meaningful. An agent that can document what it did is valuable. An agent that did the right thing, and can document it, is what enterprise deployment actually requires.
Accountability isn’t an afterthought
The teams building toward an agentic supply chain need to design for accountability from the start, not as a compliance afterthought, but as a core architectural requirement.
Because in 2026, « we were automating » is not a legal defense. It shouldn’t be an operational one either.
The shift to AI-driven supply chain execution is real, and the efficiency gains are real. The question isn’t whether to deploy agents. It’s whether you’ve built the foundation that makes deployment defensible when it matters most.
Key takeaways
- The broker liability ruling has made auditability a legal, not just operational, requirement for AI-driven supply chain decisions
- Enterprise-grade agents need to be explainable, governable, and operate as systems of record
- Context quality determines decision quality: agents operating on shallow data produce defendable mistakes, not defensible decisions
- Human-in-the-loop governance isn’t a constraint on AI capability; it’s what makes autonomous execution trustworthy at scale
- Design for accountability from the start, not as a compliance layer added after deployment
