Imagine you have a new intern at the office, a clever youngster who arrives every morning full of enthusiasm but who has never worked in a supply chain job before. While that youngster has plenty of promise, it will still take a lot of training and coaching to help them reach their full potential.
Industry experts say this hypothetical intern exists now, with one important condition: The worker is not a recent college graduate—or even a human being at all—but rather an agentic artificial intelligence (AI) application. Agentic AI is a type of system that can accomplish a specific goal with limited supervision by using “AI agents,” which are machine learning models that mimic human decision-making to solve problems in real time, according to IBM.
Still in the early stages of development, this technology offers great promise for solving problems in the supply chain sector, which relies on professionals to make frequent decisions based on large sets of data—think of all the numbers involved in staffing, inventory, routing, deadlines, tariffs, and the like.
AN EXPANDING UNIVERSE OF AGENTS
Theoretically, agentic AI could accelerate that decision-making process, since computers excel at analyzing data and making quick calculations. But first, the technology faces a crucial hurdle: AI needs to be trained on the complexities of operations like warehousing, fulfillment, and transportation. Far from replacing people with machines, agentic AI is reliant on supply chain professionals to show it the ropes, experts say.
Sanjev Siotia, executive vice president and chief technology officer for supply chain software developer Manhattan Associates, likens the roles played by the various participants in the process to those played by members of a symphony orchestra, saying agentic AI may be able to make decisions and serve as a conductor, but it still relies on skilled—and human—musicians to play the actual instruments.
Earlier this summer, the company launched a collection of AI agents, saying they can take intelligent, autonomous actions to revolutionize supply chain commerce execution, optimization, and the user experience. The launch included five specific digital agents: the Intelligent Store Manager, Labor Optimizer Agent, Wave Inventory Research Agent, Contextual Data Assistant, and Virtual Configuration Consultant. And Siotia says there will be many more agents to come.
The company has also launched a tool that allows users to create their own customized AI agents—ones tailored to their operations’ unique processes and preferences. That tool, the “Manhattan Agent Foundry,” is the same platform that the software developer’s research and development (R&D) team uses to build the commercial AI agents the company offers but has now been opened up for clients to either use themselves or contract with Manhattan or third-party partners to develop new specialized agents.
According to Siotia, such AI agents could be used in a variety of common supply chain applications, such as:
- Managing warehouse flows, handling tasks like balancing inventory reserves, identifying tasks that are running behind schedule, determining whether reassigning warehouse workers would get operations back on track, and then communicating to those workers exactly where they should move.
- Transportation invoicing tasks, such as assigning AI agents to read the PDF files that are frequently emailed by small carriers and then entering that information into a shipper’s software platform and taking steps to solve any problems that are revealed.
- Fulfillment problem resolution, like answering “Where’s my stuff?” questions and tracking down missing inbound shipments.
HOW TO TRAIN YOUR NEW AI ASSISTANT
When it comes to getting the most from AI agents, proper training is the key, says Rachit Lohani, chief product and technology officer at supply chain software developer e2open. In Lohani’s view, a common mistake in the market is to conflate automation—which is deterministic and role-based, with zero intelligence—with AI, which functions more like an aide or associate.
“Agentic AI is about providing you with an assistant. So [to design it], we have to understand your key metrics, how you make decisions, how you rationalize, and what you are optimizing for. Then when we train the model, we tell it, ‘Start making decisions like that,’ and from there, it keeps learning,” Lohani says.
A hallmark of AI technology is that it gets better over time, internalizing lessons on how to improve its performance every time a human mentor corrects it. And that means that supply chain professionals play a critical role in providing oversight and feedback to the agentic AI systems, he says.
“We understand the value of ‘human in the loop’; humans are a paramount linchpin in this cycle,” Lohani says. “If you don’t have governance and human oversight, it will fail.”
RESERVING THE RIGHT TO CLICK “BUY”
That’s also the prevailing wisdom in the retail sector, where a number of big players are using agentic AI to assist shoppers with their online purchases—say, by comparing prices, shipping times, and availability; sending them alerts on price drops for items they’ve viewed or added to their wish lists; and narrowing down options based on historical preferences.
However, there appear to be limits to how much control shoppers are willing to cede. In a report released this summer, megaretailer Walmart found that while consumers are increasingly open to letting AI guide their shopping journeys, they’re not yet willing to trust a digital assistant to choose and purchase products on their behalf.
“There is a strong desire for human-in-the-loop systems where human oversight and control is maintained,” Desiree Gosby, Walmart’s senior vice president, tech strategy and emerging tech, said in a release announcing the findings of the study, Walmart’s Retail Rewired Report 2025: Agentic AI at the Heart of Retail Transformation. “In fact, 46% of respondents said they were either somewhat unlikely or very unlikely to use a digital assistant or agent to handle an entire shopping trip for them. To me, the preference is clear: AI can help guide, but shoppers want to be the ones clicking ‘buy.’”
Walmart concluded that shoppers are looking for ways to “keep the fun” of shopping while simplifying the practical parts with digital assistants.
Striking that balance is a clear goal throughout the industry, whether it’s on websites, in stores, or in warehouses and DCs, tech leaders say. Assigning AI agents to perform menial tasks with speed and accuracy can free up humans for more nuanced work—jobs requiring the kind of creativity, wisdom, and experience that can’t yet be replicated by code. So when your digital assistant shows up for its first day on the job, remember to be patient—it’s going to take a little time to bring the new hire up to speed.
