In the fast-paced world of supply chain management, every second matters. Delays, inefficiencies, and disconnected data can have massive financial and operational consequences. The COO’s focus is on maintaining a durable, sustainable supply chain, from the design of the product all the way to the delivery and operation at a customer location.

This broad view can yield more efficient operations and greater flexibility, productivity, and sustainability because decision-making is based on data, and resources are used more strategically. Lately, however, the supply chain has given COOs more than their fair share of headaches, especially when it comes to anticipating and mitigating disruptions: pandemics, trade wars and tariffs, social unrest, and even extreme weather. You know the routine. That’s why the introduction of agentic AI—intelligent, sometimes autonomous agents that can understand natural language, bridge information gaps, integrate across systems, and even take action—represents a huge opportunity and even a business imperative for operational improvements for manufacturers and their supply chain partners.

There’s a major difference between generative AI and agentic AI. While generative AI is great for creating content, making predictions or answering questions, agentic AI takes things further. It doesn’t just generate insights—it has the ability to act upon them. AI agents can collaborate, make decisions, and initiate actions across different business functions. In other words, agentic AI moves AI from suggestion to execution.

AI agents can give COOs oversight and control over all design-to-operate processes in near-real time—simply by giving the agent a goal to achieve, phrased as a natural language query. Let’s say the COO at technology manufacturer asks the AI agent, “How can we improve efficiency and reduce the cost of manufacturing laptops by five per cent?” This orchestration agent interprets the request and has the latitude to determine and apply the best actions to take to find the answers. The orchestration agent is the COO’s proxy across all the different systems involved in this process. It will harness information from production, logistics, suppliers and business partners—sometimes interacting with other AI agents in those areas—and bring it all together to determine the most efficient and least costly path. That could mean switching from gold- to copper-based connectors, changing to suppliers with better prices, improving maintenance to reduce manufacturing and assembly downtime, or just moving to a more efficient assembly location.

It’s easy to see why agents can be exponentially faster at finding a way to improve production, compared to manually gathering information from each department.

For COOs, agentic AI presents both a challenge and a unique opportunity. AI agents have the potential to transform supply chains from reactive and isolated operations into intelligent, continuously improving networks.

But as with most new technology, you need to separate the hype from the reality. At present, we are beginning to see the first signs of real value from AI agents in the supply chain, with a handful of forecasting, manufacturing, and warehousing agents operating efficiently within their domains. But the technology is advancing at warp speed, and true integration of these AI agents isn’t as far off as you might think. When agents can orchestrate data sharing and reconciliation across departments and functions, not only do broad processes become more efficient, but the COO and CFO can work better together. That’s why COOs need to consider first steps towards agentic AI now, to be ready as capabilities grow.

What can AI agents realistically do today?

For the moment, cross-functional agents are hard to find in the wild, but the largest companies—whose supply chain is their lifeblood—already deploy specialised AI agents. In 2025, nearly half (44%) of U.S. retail e-commerce sales flow through Amazon and Walmart. Walmart employs AI agents to forecast demand and adjust stock levels across its vast network of shops. The agents use historical sales data and external factors (such as community events or local weather) to predict demand, allowing the company to stock the right products at the right time and reduce overstock. Amazon integrates AI agents in its fulfilment centres to streamline warehouse operations. The agents manage stock, optimise shelf space, and automate order picking. As another example, logistics provider DHL uses AI-powered agents to monitor and optimise logistics in real time. The agents track shipments, identify potential disruptions such as delays or inventory shortages, and suggest alternative routes to minimise disruptions.

But increasingly, AI agents aren’t limited to the largest companies. Supply chains of all sizes can start exploring AI agents as the AI ecosystem quickly develops.

For most supply chain operations, fully autonomous and interconnected AI agents are still very conceptual in nature, but the technology is moving quickly, says Sree Mangalampalli, vice president of digital transformation solutions at supply chain AI company FourKites. Of the 33 different types of supply chain AI agents that Mangalampalli identified in a LinkedIn article, “I would say that 25% are actual reality in supply chain operations today (February 2025).”

AI agents ride a steep learning curve

AI agents today are just beginning to expand their capabilities. Here are four phases of agentic development; each new phase will represent a big step up in complexity.

1. Initial insights. Each AI agent takes a natural language query and analyses structured and unstructured data within its area of the supply chain to deliver relevant information and suggest a course of action.

2. Parallel insights. AI detects patterns in data beyond the initial enquiry. The agent has the ability to say, “You have asked this question but, while analysing data for your answer, I see additional issues A, B, and C. Do you want me to act on them?” This stage represents an exponential leap in complexity. Expanding the scope of possible actions from the initial enquiry requires significant work to identify and enable all the available processes that could be executed.

3. Suggested implementation. As the agent becomes familiar with your actions, it will start to recommend actions based on prior decision patterns. If you have asked a similar question several times in the past, the AI agent provided similar insights, and nine out of ten times you have taken similar action, the agent will ask if you would like it to take that same action this time. At this stage, the AI agent provides the user with supporting information to make decisions.

4. Autonomous execution. This is agentic AI’s ultimate goal. Now that the AI agent has “learnt” your responses to certain queries or data, it will take action on its own—or with very minimal human intervention.

Regarding the hype versus the reality, agentic AI is definitely growing rapidly;but full autonomy is still a work in progress, says Carlos Romo, senior account manager at CodersLab, which helps automate the automotive industry’s supply chain. “Many companies are experimenting with it, but human oversight is still needed for complex decisions. The tech is evolving fast, though, so we’re getting closer to more self-sufficient systems.”

The majority of supply chain organisations use AI agents for forecasting, he says. Since the pandemic, organisations can’t count on historical data as a basis for future growth. “You have to bring in external market factors with the closest correlation to your business and do your forecasting accordingly,” he says. “Let the AI learn from those market factors to understand how much they should be weighted into future forecasting and how it will affect your demand. It’s a very complex problem, but the agent is constantly learning and updating your forecast as it goes.”

Manufacturing scheduling is another easy target for AI agents. They can analyse data from materials providers, customer changes, and delivery targets to have more efficient scheduling on the manufacturing floor and less idle time. AI agents are in the warehouse, ensuring the inbound inventory is optimised to outbound shipments, so you have limited storage and efficient distribution.

Are these agents interacting with each other across the supply chain? Not yet, but the technology is moving quickly, Mangalampalli says. “Give it six more months.”

The art of the possible: What is agentic AI projected to do tomorrow?

Agentic AI use cases are emerging for every function in the supply chain process, and each has the potential to improve efficiency, save money, and increase automation throughout.

The ultimate goal is to have all these task-specific supply chain agents not only working together but working with AI agents in other areas of the business—such as sales and procurement—to create an end-to-end process with flexible, intelligent automation.

Product design

AI can analyse customer feedback, market trends, and performance data to suggest improvements to existing products or even to inspire new products.

When designing a new skincare product, AI can capture all of the company’s customer service data about how its current products are performing and what is being asked for in the market. Then the designer could ask the AI agent what is missing from its existing products or what customers are asking for. For example, vitamin C and niacinamide are trending online as top skincare ingredients in demand for 2025. The AI agent will delve into R&D and social media data that matches your customer base and come up with alternative formulas, recipes, or models of how the organisation could create new products. It could even suggest new materials that could be used such as replacing petroleum distillates—which are banned in the UK—with shea butter, beeswax, or coconut oil. Or it could avoid certain materials such as parabens or aluminium that many consumers don’t want in their products.

Manufacturing

AI agents can optimise production steps and processes, and there’s growing interest in using them for quality control. Visual inspection tools can already identify defects on a production line in real time, reducing waste and improving quality. But the AI agent wouldn’t stop there; it could also trigger maintenance work orders or adjust production parameters automatically.

In contract manufacturing, AI agents can monitor the progress of production orders and communicate directly with contractors’ AI systems to ensure organisations are meeting their service level agreements (SLAs).

AI agents will also determine the best sequence of steps in the production line and adjust them accordingly.

Warehouse Management

Agentic AI continuously adapts warehouse operations based on demand fluctuations and stock levels. It can also work with warehouse robotics systems already in place to better store materials for more efficient usage. For example, if demand spikes for a particular electronic component, AI agents can adjust warehouse layouts to ensure that the product is positioned near the loading bay for easier accessibility. With further learning, AI agents will automate the entire process for shipping, packing, and improving fulfilment speeds.

Transport & logistics

AI agents can optimise delivery routes, improve delivery times, and reduce fuel costs. They can also use external data, such as weather reports, petrol prices, and news reports of port congestion, to adjust shipments or driving routes in real time.

Here, it is vital to have access to unstructured data such as texts and news reports. When a ship was blocking the Suez Canal, the information was transmitted by text and it affected supply chains. Given a similar situation, AI agents could detect the incident quickly from internal communications and newsfeeds. They could then proactively calculate the economics of rerouting deliveries around the Cape of Good Hope or by land, and shift orders to alternative suppliers or adjust inventory strategies and pricing.

Planning & forecasting

AI agents will soon be synchronised across manufacturing, planning, and logistics, and be able to tell you the most efficient manufacturing plan to use to meet SLAs. For example, a planner could ask an AI agent whether he should send two production orders to the same manufacturer or to separate manufacturers to meet SLAs. The agent will gather data on the raw materials, manufacturing specifics, and SLA requirements to form a plan of action.

When it comes to forecasting, monitoring structured and unstructured data, from weather to social media to audio news accounts, with AI agents in real time can help businesses quickly adjust their production and logistics plans.

What’s holding the supply chain back from agentic AI?

Tech vendors are betting that businesses are ready to use autonomous AI bots, but companies aren’t so sure. While 61% of attendees at a Wall Street Journal Tech Summit in February said they’re experimenting with AI agents, 21% said they’re not using them at all. The poll found that their most pressing concern with the technology is a lack of reliability, and that includes having reliable data.

Before AI of any kind can be used successfully, supply chains need both structured and unstructured data to be visible, correct, and accessible. And that’s the problem. Companies are drowning in data—from internal systems, supplier collaborations, logistics providers, and even social media and Internet of Things (IoT) sensors—but much of it is siloed, unstructured, or just plain messy.

While a first step towards agentic AI includes getting your data in order, it’s easier to do that when starting with a few agents in specific areas of the supply chain.

Mangalampalli even suggests shifting to more of a continuous improvement and flexible mindset when it comes to data. “It doesn’t have to be perfect before we get started,” he says.

What role will humans play in a supply chain driven by agentic AI?

As mentioned earlier, AI agents still need a human in the process to verify the decisions they are making. And even as AI’s capabilities grow, the human role in supply chain management does not disappear; it evolves. It doesn’t matter how autonomous the agent is, a human is still going to be at the centre of it providing oversight. But the skillset of the human at the centre is going to change.

Today, supply chain management relies on specialised employees—planners, plant managers, logistics experts, and others. AI agents will absorb much of the expertise needed to create new product designs, set manufacturing schedules on the shop floor, or choose the best materials supplier for a new product. This will shift decision-making towards business operations and the COO, and away from the technical experts.

These reskilled "knowledge experts" will instead be tasked with having a holistic view of the AI agent’s particular function and how it interacts across the supply chain. They will oversee the AI agent’s performance and validate its interactions with different data inside and outside of its functional area.

The key to success will be balancing AI automation with human judgement. It’s doubtful we’ll ever get 100% replication of the supply chain in an agent, and highly likely there will always be a human in the loop. AI won’t replace humans, but humans who use AI will replace humans who don’t.

What should COOs do now?

For COOs looking to start integrating agentic AI into their supply chain, the first step isn’t buying the latest AI tool; it’s identifying where the biggest business challenges are.

1. Have a clear business outcome in mind. If you understand the outcome, you can build your agents to achieve that.
2. Check your data. Do you actually have the correct data to solve these problems? If not, where can you obtain it?
3. Collaborate with software providers. Companies including SAP are developing AI agents so COOs don’t have to build them from scratch. Some are building individual agents with particular functions, while others are offering—or planning to offer—software that manages agents or lets the user quickly build their own agents by combining pre-built functions.

Today’s supply chain is rife with problems that can benefit from agentic AI. AI agents are working to master data sharing and integration among all the global supply chain businesses, even if these sources use different formats, standards, and systems. And it can navigate a dynamic and complex supply chain environment. Agentic AI can go beyond traditional, rules-based automation that depends on pre-defined scenarios, and instead learn from historical data to predict potential disruptions and automatically adjust the plan without requiring manual intervention.

Even more promising is when data can be integrated and shared across the organisation, interactions between COOs, CFOs, and the rest of the C-suite are much more collaborative. This ensures that their functions are working together to find the bottlenecks and incorporate the right balance of autonomy and safeguards.

Agentic AI in design-to-operate processes isn’t just an upgrade to generative AI; it can be a giant leap forward. It moves AI from passive insights to active execution. But to make it work, businesses need a strategy. Start with the right data, find the right use cases, and get AI agents to first be working efficiently within their discipline. You’ll then be ready as technology progresses. That’s how you go from AI hype to AI results.

Read more