The AI discussion in logistics is quickly moving beyond copilots, chatbots, and better search. At the Descartes Innovation Forum today, Descartes demonstrated something considerably more interesting: agents working across multiple logistics applications and organizations to execute parts of a representative end-to-end supply chain process.

That distinction matters.

Descartes described one of the persistent problems in logistics as the “swivel chair tax.” A shipment may move physically from origin to destination, but the information required to manage it still moves through transportation systems, compliance applications, carrier portals, email inboxes, messaging platforms, spreadsheets, and people. The applications themselves are often perfectly capable. The problem is everything that happens between them.

Descartes’ emerging answer is what it calls its Agent Control Plane. In the keynote demonstration, an order moved through four representative companies and 15 Descartes products, with 28 individual process steps occurring behind the scenes. Agents performed tasks ranging from compliance checks and information gathering to disruption detection and replanning, while the underlying Descartes applications continued to perform the operational work they were built to do.

But the most important part of the demonstration may have been what the agents did not do. At several points, humans remained responsible for consequential decisions. When a compliance issue appeared, the agent gathered the relevant information and escalated the decision. Later, when a disruption created an opportunity to rebook transportation at an 8% savings, the alternative carrier carried a trust score of just 32. The agent surfaced the option; the human rejected it.

That is a much more realistic model of logistics automation than the idea of simply turning operations over to autonomous AI. Descartes summarized the philosophy particularly well: autonomy is a dial, not a switch.

This is where agentic AI starts becoming operationally interesting. Consider what normally happens when an international shipment is disrupted. Someone discovers the change, someone determines which shipments are affected, and other people begin checking capacity, appointments, customer commitments, carrier options, and downstream consequences. Emails and phone calls start moving among multiple organizations, and by the time the problem is fully understood, hours may have passed.

In the Descartes demonstration, agents detected the disruption, evaluated its downstream impact, investigated alternatives, and coordinated information across the participating companies. Instead of simply alerting the shipper that something had gone wrong, the system could potentially deliver something much more valuable: the problem and the proposed resolution together. That represents a meaningful change in the role of supply chain software.

For decades, enterprise applications have largely waited for people to operate them. Agentic systems introduce the possibility that applications can increasingly initiate work themselves—within defined permissions and with humans inserted at the appropriate decision points. Descartes also emphasized that this is not merely a future concept. The company said it has already executed approximately 3.25 million agent operations and is opening an early-access program for the Agent Control Plane.

Just as important, Descartes is building governance around the model. Agents have identities, actions are logged and attributable, and activity can be reviewed and replayed. That may ultimately prove as important as the AI itself because enterprises will want different levels of human oversight depending on the decision, risk, and business context. They may be very willing to let agents do the investigative work, coordinate routine activities, react to predefined conditions, and bring humans the relatively small number of decisions that actually require judgment.

That was my biggest takeaway from the keynote.

The next generation of logistics automation may not be about removing humans from the process.

It may be about removing humans from all the work they never needed to be doing in the first place.

The post Descartes Takes Agentic AI Into the Logistics Workflow at the Descartes Innovation Forum appeared first on Logistics Viewpoints.

Real-time logistics data has become a foundational capability for global supply chains, but data alone does not resolve a disruption. The operational value emerges when visibility is connected to prediction, prioritization, and the workflows required to change an outcome.

project44’s Movement platform is moving in that direction. The company combines a large multimodal transportation data network with predictive ETAs, analytics, alerts, and increasingly autonomous workflows. Its role is primarily execution-focused: it sits around existing planning and transportation systems and provides current operational context about what is actually happening across the freight network.

That position gives project44 a useful foundation for decision intelligence. A shipment event can be evaluated against order commitments, customer priorities, facility conditions, and downstream risk before an operator is asked to intervene. The more accurately the platform can distinguish consequential exceptions from routine variability, the more effectively it can automate communications, recommendations, and bounded execution tasks.

The challenge for visibility providers moving into decision and action is proving that the automation layer creates measurable operational improvement. Buyers should test data quality, coverage, prediction accuracy, workflow integration, and governance around autonomous actions rather than assuming that more real-time data automatically produces better decisions.

project44 is represented in the Logistics Viewpoints Supply Chain Decision Intelligence MarketMap and Autonomous Exception Management MarketMap. The pairing captures the company’s evolution from visibility toward a broader real-time intelligence and exception-management role.

The post project44 Pushes Real-Time Logistics Data Toward Execution appeared first on Logistics Viewpoints.

Executive thesis. Agentic AI becomes consequential when software can take action, not merely generate an answer. That makes authority, permissions, observability, and recovery the defining architectural questions.

The important word is agency

An AI assistant can summarize, recommend, and answer questions without changing the state of the business. An agent is more consequential because it can pursue a goal through a sequence of actions—gathering context, calling tools, evaluating results, and deciding what to do next. In logistics, that can mean interacting with orders, shipments, inventory, appointments, suppliers, or enterprise workflows.

Operational autonomy requires explicit boundaries

The more freedom an agent has, the more precisely its authority must be defined. Which systems can it access? Which actions can it take without approval? What financial thresholds apply? Which customers, suppliers, or facilities are in scope? When should it stop and escalate? These are not abstract governance questions. They are the control surface of the operating architecture.

Context has to be authoritative

Agents are only effective if they can distinguish source-of-truth records from unverified or generated information. Retrieval, permissions, identity, timestamps, and system state therefore matter as much as model reasoning. A confident agent acting on stale shipment status or an obsolete policy can create more operational risk than a conventional workflow.

Observability and recovery are first-class requirements

Multi-step agentic workflows can fail in more than one place: a tool may time out, a system may reject an update, the underlying data may change mid-process, or the model may choose an invalid path. Production designs need logging, state, retries, idempotency, escalation, and recovery. The organization has to be able to reconstruct what the agent attempted and why.

Autonomy should expand with evidence

The practical path is controlled progression. Start with narrow workflows, strong observability, limited tool rights, and clear human approval. Measure error rates, overrides, completion, recovery, and business outcomes. Expand autonomy only where the evidence supports it. The objective is not maximum autonomy. It is dependable delegation.

Logistics Viewpoints’ Agentic AI in Logistics: What It Is and How It Works defines the control architecture around agents: goals, context, tools, permissions, approval gates, observability, recovery, and enterprise-system access.

Executive implication

Enterprises should expand agent autonomy only as evidence accumulates that controls, escalation, auditability, and recovery work reliably under real operating conditions.

Go deeper: provides the durable buyer, architecture, and implementation reference for this topic. AI & Advanced Analytics connects this analysis to the broader Logistics Viewpoints research architecture.

Related Logistics Viewpoints research

Supply Chain Decision Intelligence: What It Is and How to Evaluate Platforms
The New Architecture of Logistics

The post Agentic AI Is Moving From Supply Chain Experimentation Into Operational Work appeared first on Logistics Viewpoints.