From Managing Supply Chains to Executing Them

Why the next generation of enterprise software may change not only how supply chains are managed, but how operational work gets done.

For decades, enterprise software has transformed how businesses organize their operations. It has replaced paper-based processes, connected departments, standardized workflows, and made increasingly complex operations visible and manageable. Yet one fundamental characteristic has remained remarkably consistent: software provides the systems through which people do their work. Even as automation has expanded, humans continue to bridge gaps between systems, interpret changing circumstances, coordinate activities, and resolve situations that fall outside predefined processes.

Artificial intelligence is beginning to challenge that division of labor. Software can now interpret unstructured information, reason about operational situations, and increasingly initiate actions based on context rather than predefined instructions alone. This opens up a different possibility for enterprise technology: systems that do not merely help people manage operational work, but progressively take on parts of that work themselves.

For global supply chains, the implications could be substantial.

Supply Chains Run on More Than Data

Modern supply chains generate enormous amounts of information. Orders, shipments, containers, inventory positions, transport schedules, delivery commitments, and countless operational events are recorded across enterprise systems.

Over the past two decades, substantial investments have gone into making this information more accessible. ERP systems manage commercial and operational transactions. Transportation Management Systems coordinate transport planning and execution. Visibility platforms track shipments and provide updates on their progress. Planning applications help companies anticipate demand, allocate resources, and optimize decisions.

These capabilities are essential. But information alone does not make a supply chain operate.

Consider a container shipment whose estimated arrival is delayed by five days.

A visibility platform may identify the delay immediately. A transportation system may update the expected arrival. A planning system may indicate potential consequences for downstream activities.

Yet the operational work has only just begun.

Someone must determine which customer orders are affected, whether existing commitments are at risk, and whether alternative arrangements are necessary. Information may need to be requested from a carrier, discussed with a supplier, or communicated to a warehouse. Delivery appointments might need to be adjusted. Customers may require updated information. Several systems may need to reflect the resulting decisions.

None of these activities is particularly remarkable. They are part of everyday supply chain operations.

But together, they illustrate an important distinction: knowing what is happening is not the same as making the necessary things happen.

Much of the work between those two states still depends on people.

The Limits of Managing Work Through Software

Traditional enterprise software is exceptionally good at certain things.

It can maintain structured records, enforce business rules, calculate requirements, execute transactions, and coordinate processes that can be defined in advance. Modern enterprise applications already automate significant parts of procurement, production, transportation, and financial operations.

The limitation is therefore not that conventional software cannot execute processes.

It is that real-world operations frequently extend beyond what has been explicitly modeled.

A shipment arrives late, but the consequences depend on the products inside it, the customers waiting for them, available stock, commercial priorities, and commitments made in previous conversations. A supplier confirms a production delay through an email rather than a structured interface. Two systems contain conflicting information. A customer changes delivery requirements after the transportation plan has been finalized.

Each situation may require a different response.

Traditional automation can address many such cases through predefined conditions, rules, and workflows. But as the number of possible situations and dependencies increases, defining every relevant combination becomes increasingly difficult.

The result is familiar across many organizations: sophisticated enterprise systems coexist with extensive manual coordination.

Employees move between applications, interpret emails, reconcile spreadsheets, request missing information, and decide what should happen next. They provide the connective tissue between systems, organizations, and business processes.

This work is not necessarily a failure of the software. Much of it exists because global supply chains are dynamic, distributed environments in which information and decisions are rarely contained within a single system.

The question is whether software can now assume a greater share of that connective work.

From Information to Action

Recent advances in AI introduce capabilities that differ fundamentally from traditional rules-based automation.

Large language models can interpret natural language, extract meaning from documents, summarize complex situations, and work with information that does not conform to a fixed data structure.

Connected to appropriate tools and business systems, AI agents can also participate in operational workflows: retrieving relevant records, evaluating conditions, preparing decisions, initiating permitted actions, and checking their results.

This does not eliminate the need for conventional software. It changes what can be built on top of — and between — existing operational capabilities.

The shift is already becoming visible. In its Logistics Trend Radar 8.0, DHL identifies agentic AI as a major emerging technology, highlighting its potential to move beyond assistance toward planning, coordination, and execution. Early applications include customer communications, warehouse coordination, and operational follow-ups.

Return to the delayed shipment.

Instead of merely presenting an updated arrival time, an AI-native operational system could investigate the event in its business context.

It could identify the affected shipment, connect it to its purchase orders and cargo, assess relevant delivery commitments, and determine whether further information is needed. It could prepare a communication to a carrier, request clarification, or propose changes to downstream arrangements.

With suitable integrations, permissions, and business controls, the system could execute certain low-risk actions directly, such as updating an internal status or sending an approved notification.

Other actions, such as accepting additional transportation costs, changing contractual commitments, or selecting an expensive alternative route, might require explicit human authorization.

The important difference is not that every decision becomes autonomous.

It is that the system can participate in the operational process rather than merely describe it.

The transition is from software that informs action to software that can carry out action.

And that changes the role enterprise software can play.

Execution Requires More Than an Intelligent Model

It is tempting to view this transition primarily as a consequence of increasingly capable AI models.

Better reasoning, larger context windows, and improved tool use certainly expand what AI systems can accomplish.

But operational execution introduces requirements that cannot be addressed through language-model capabilities alone.

An AI model may correctly understand that a delayed shipment threatens a customer commitment. That does not establish which information is authoritative, which actions are permitted, or whether a proposed change is consistent with the current state of the business.

To execute reliably, a system needs several complementary capabilities.

Operational context. The system must understand relationships between business objects such as orders, order lines, cargo, containers, shipments, locations, and delivery commitments. An event only becomes operationally meaningful when it can be related to the business activities it affects.

Domain-specific logic. Supply chains follow commercial, logistical, and organizational rules. Quantities must remain consistent. Documents and events have specific meanings. Decisions may depend on contractual arrangements, transport conditions, or operational constraints that a general-purpose AI model cannot simply assume.

Reliable state management. Operational actions change business records and sometimes trigger further processes. Systems need to maintain consistent information, distinguish proposed changes from committed ones, and prevent conflicting or duplicate transactions.

Controlled execution. An agent needs access to tools, applications, and external services, but that access must be bounded. Actions should be evaluated against permissions, business constraints, and their potential consequences.

Verification and traceability. The system must establish whether an action was successfully completed, what changed as a result, and when human intervention is required.

Together, these capabilities suggest a different architectural emphasis.

In many conventional applications, software functions are organized around the activities users perform through an interface. In an AI-native operational system, another organizing principle becomes possible: the business outcome to be achieved, supported by agents, domain models, integrations, and controlled execution mechanisms.

The user interface remains important. But it is no longer necessarily the place where every individual operational step must originate.

This is also why adding an AI assistant to an existing application is not equivalent to building an AI-native execution architecture.

An assistant may explain information or recommend actions. An execution system must also have a dependable way to perform those actions and account for their consequences.

Autonomy Is Not the Same as Reliability

If software is going to perform operational work, its ability to act must be matched by its ability to act within acceptable boundaries.

Supply chains involve financial commitments, contractual obligations, physical assets, and dependencies between companies. A minor data error can propagate into multiple downstream activities. An unauthorized action can have consequences that are difficult or costly to reverse.

AI models introduce additional uncertainty because their outputs are probabilistic. They can generate plausible but incorrect interpretations, overlook relevant constraints, or select inappropriate actions.

This does not make AI unsuitable for operational environments. But it makes the design of the surrounding system critical.

A reliable architecture should distinguish between interpreting a situation, determining which actions are permitted, and committing a change to an operational system.

Not every action requires the same level of control.

Retrieving shipment information may be straightforward. Sending a routine status update could be permissible under defined conditions. Changing a delivery commitment or authorizing additional expenditure may require human approval.

Some decisions may be too ambiguous, consequential, or poorly supported by available information to automate at all.

An effective operational system should recognize these boundaries and escalate appropriately.

These principles are consistent with the broader emphasis on governance, risk management, and trustworthiness reflected in the NIST AI Risk Management Framework.

Human oversight, therefore, should not be understood simply as a temporary limitation that will disappear as AI models improve. It can be a deliberate component of the operating model.

The objective is not maximum autonomy at any cost.

It is the greatest useful degree of automation consistent with reliable, accountable execution.

A Different Operating Model for Supply Chains

As operational software begins to perform more of the work previously handled by employees, the implications extend beyond technology architecture.

The structure of operational roles may gradually change.

Today, many supply chain professionals spend substantial time collecting information, following up with external parties, reconciling inconsistencies, updating systems, and coordinating next steps.

These activities are necessary, but they are not always where human expertise creates the greatest value.

If software can reliably handle a growing share of routine coordination, people can focus more attention on situations requiring judgment, negotiation, commercial understanding, and strategic decisions.

Rather than processing every operational event themselves, teams may increasingly define objectives, establish decision boundaries, supervise automated processes, and intervene when exceptions require human expertise.

This shift will not happen uniformly.

Highly structured processes are already extensively automated in many organizations. Other activities depend on fragmented data, informal communication, complex commercial relationships, or external partners who cannot easily be integrated into a common workflow.

The opportunities will therefore vary substantially between companies and processes.

Nor should the economic consequences be assumed.

AI-based execution introduces its own costs: model inference, integration, monitoring, governance, exception handling, and ongoing maintenance. A technically feasible automation is not necessarily an economically attractive one.

The meaningful question is not how many agents an organization deploys.

It is whether those agents can reliably complete useful operational work, at an acceptable cost, with a measurable improvement in business performance.

That suggests a different set of metrics for enterprise software: not just information coverage, user activity, or workflow adoption, but successfully completed tasks, processing time, intervention rates, execution quality, and the cost of achieving an operational outcome.

From Managing Work to Getting Work Done

For decades, enterprise software has become progressively better at representing and organizing the world of business operations.

It records transactions, makes information accessible, supports planning, and automates well-defined processes. These capabilities will remain indispensable.

AI introduces an opportunity to extend that foundation.

By combining contextual interpretation with domain-specific logic, reliable business state, and controlled access to operational systems, software can begin to take on work that previously required continuous human coordination.

The transition will be gradual. It will depend as much on architecture, integration, organizational trust, and economics as on advances in AI itself.

And it will not eliminate the need for experienced supply chain professionals. It may, however, fundamentally change how their expertise is applied.

The next frontier of supply chain software is not simply better visibility into operational work.

It is enabling software to take on more of the work required to move operations forward — reliably, within defined boundaries, and under human accountability.

That is the shift from managing supply chains to executing them.