Agentic AI: From Chatbots to Autonomous Co‑Workers in the Supply Chain
Introduction
For the past few years, executives have embraced generative AI (gen AI) with unmatched enthusiasm. Surveys show that nearly eight in ten companies report using gen AI, yet roughly the same share see no material impact on earnings. This “gen AI paradox” stems from an imbalance: horizontal tools such as chatbots and copilots scale quickly but deliver diffuse, hard‑to‑measure benefits, while the higher‑impact, function‑specific use cases often get stuck in pilot mode. As boardrooms push for more tangible value, a new wave of technology is emerging to close the gap: agentic AI—autonomous agents capable of planning, acting and collaborating with humans and machines.
Agentic AI goes beyond responding to prompts. It integrates natural‑language understanding, memory, planning and execution to perform multistep tasks on behalf of humans. Where a chatbot might summarise an order history, an agent could reorder parts, reschedule production and notify stakeholders. Analysts believe this shift could transform supply chains, making them smarter, more resilient and more efficient than ever before. This article explores what agentic AI is, how it’s being applied today, and what supply‑chain leaders should do to prepare for an agent‑driven future.
The rise of agentic AI: solving the gen AI paradox
McKinsey’s 2025 report Seizing the agentic AI advantage identifies the core of the gen‑AI paradox: widespread deployment with limited returns. The main issue is that horizontal use cases (e.g., corporate copilots, chatbots) have proliferated but deliver only incremental efficiency, while vertical applications in functions like supply chain, finance and manufacturing struggle to scale. Agentic AI offers a pathway out of this stalemate by turning gen AI models into goal‑driven collaborators that can plan, act and learn across systems.
The momentum behind agentic AI is building quickly. Gartner projects that by 2028, one‑third of enterprise software platforms will include agentic AI, up from just 1 % in 2024. The same forecast suggests that 15 % of day‑to‑day work decisions will be made autonomously, compared with essentially zero today. MHI and Deloitte’s 2025 industry report highlights similar enthusiasm: 28 % of supply‑chain leaders use AI today, and 54 % plan to adopt it within five years, bringing adoption to 82 % by 2029. Over half of executives are increasing tech investment, and 60 % plan to spend over US$1 million on supply‑chain technologies. These numbers suggest that the conditions are ripe for the move from pilot projects to large‑scale agentic deployments.
What agentic AI can do today
From specialised agents to emerging ecosystems
Although the concept of agentic AI is still nascent, large companies are already deploying specialised agents to tackle specific supply‑chain tasks. In 2025, nearly half of U.S. retail e‑commerce sales flow through Amazon and Walmart, and both giants leverage AI agents to stay ahead. At Walmart, AI agents analyse historical sales data and external factors—such as community events and local weather—to forecast demand and adjust inventory levels across thousands of stores. Amazon integrates agents in its fulfilment centres to manage inventory, optimise shelf space and automate order picking, ensuring products are picked and packed with minimal human intervention. Logistics provider DHL uses agents to monitor shipments in real time, identify potential disruptions and suggest alternative routes to minimise delays.
These examples illustrate what SAP calls the first insights phase of agentic development. An agent can analyse structured and unstructured data, deliver a recommendation and allow humans to decide. As systems mature, agents will progress through parallel insights, spotting additional issues in the data and asking whether to act on them. In the suggested execution stage, the agent learns from prior actions and offers to take the same step on behalf of the user. The endgame—autonomous execution—sees agents act on their own with minimal human oversight.
According to FourKites’ Sree Mangalampalli, a supply‑chain AI expert quoted by SAP, about 25 % of the 33 types of supply‑chain AI agents he has identified are already in active use. Forecasting and manufacturing scheduling are early targets: agents combine external market factors with historical data to continually update demand projections, while on the factory floor they adjust production schedules in real time to reduce idle time. In warehouses, agents ensure inbound inventory matches outbound shipments, optimising storage and distribution.
Applications: from warehouses to supplier management
Warehouse optimisation and inventory management
Warehouse operations involve countless decisions about slotting, picking, replenishment and layout. Agentic AI excels at gleaning insights from “oceans of data” and can orchestrate physical movements. MHI member sSy.AI’s SCOTi platform uses agentic AI to identify bottlenecks in warehouse operations; during a live demonstration, the software pinpointed delays between picking and shipping, offering actionable insights. Dexory’s digital‑twin platform DexoryView pairs autonomous mobile robots (AMRs) with agentic AI to continuously update warehouse layouts. CEO Andrei Danescu predicts that future versions will automatically re‑slot inventory or regenerate warehouse layouts based on real‑time data. Verity’s drone‑based system is already deployed in more than 150 sites worldwide and uses computer vision to find misplaced inventory; CEO Raffaello D’Andrea envisions agents collaborating to fix errors without human instruction.
Predictive maintenance and physical AI
Beyond the warehouse, agentic AI can supervise fleets and equipment. The technology’s strength in pattern recognition and planning enables early detection of anomalies, reducing downtime. Predictive maintenance is among the top use cases for agentic AI. By analysing sensor data and maintenance logs, an agent can predict when a machine will fail and schedule repairs before breakdowns occur. As physical AI—robots and drones that can act autonomously—matures, agents will not only schedule maintenance but also dispatch robots to perform simple repairs.
Demand forecasting and process automation
Demand forecasting is another fertile ground. Agents combine historical sales data with real‑time signals such as weather, social media sentiment or economic indicators. Walmart’s agents, for example, adjust inventory levels by considering local events. Amazon uses agents to optimise shelf space and order picking. Future agents could also automate procurement: vendors are developing agentic systems that renegotiate contracts and manage supplier relationships with minimal human supervision.
Compliance, risk and sustainability
Agentic AI will also help companies navigate new regulatory requirements and ESG mandates. AI agents can build knowledge graphs that map relationships among suppliers, hubs and materials, enhancing visibility beyond tier‑1 suppliers—a technique highlighted by risk‑management experts at WTW. Agents could monitor environmental, social and governance (ESG) metrics and automatically adjust sourcing decisions to meet carbon‑reduction or human‑rights obligations. Combining agentic AI with lifecycle‑assessment tools may help reduce carbon footprints across the value chain.
Overcoming barriers: technology, trust and talent
While the potential is clear, agentic AI still faces obstacles. McKinsey cautions that organisations must reimagine workflows around agents, not just bolt them onto existing processes. Achieving scalable impact requires shifting from isolated pilots to strategic programs and embedding agents deep in core workflows. This transformation demands an open, extensible and observable infrastructure, where custom‑built and off‑the‑shelf agents can interact across systems. Governance mechanisms must prevent uncontrolled agent sprawl and ensure that humans remain in control.
Trust is another hurdle. Many technologies showcasing agentic AI at trade shows were still experimental. Executives worry about relinquishing decision authority to software. Building confidence will require transparent agents with explainable decision‑making, robust security to guard against cyber threats and bias mitigation to ensure fair outcomes. Moreover, organisations must upskill their workforce to collaborate with agents. Analysts emphasize that the larger challenge is human—not technical—requiring alignment across goals, tools and people.
Building an agentic AI strategy
As supply‑chain leaders weigh the possibilities, they should take a phased, pragmatic approach:
- Identify high‑value, vertical use cases. Focus on specific processes—inventory forecasting, slotting, maintenance or transportation planning—where autonomous agents can deliver measurable improvements. Avoid scattering efforts across too many pilots.
- Invest in data readiness and integration. Agentic AI thrives on clean, connected data. Modernise data architectures, adopt shared taxonomies and integrate IT and OT (operational technology) systems to provide agents with real‑time information from across the network.
- Adopt digital twins and simulation. Gartner lists decision intelligence and intelligent simulation alongside agentic AI in its top supply‑chain technology trends. Digital twins allow leaders to test agentic policies in a virtual environment before deploying them in the physical world. Combining simulation with agentic agents builds confidence and mitigates risk.
- Set up governance and ethics frameworks. Establish policies for agent autonomy, human override, data privacy and bias mitigation. Embed security by design, as regulators and customers increasingly demand transparency and accountability.
- Upskill and align teams. Encourage cross‑functional collaboration among supply‑chain planners, data scientists, IT and operations. Provide training on agentic AI concepts and foster a culture of experimentation and continuous learning.
Future outlook
Analysts predict that agentic AI will move from hype to reality faster than many expect. By 2028, one‑third of enterprise software platforms will include agentic AI and 15 % of work decisions will be autonomous. Supply‑chain technology spending is accelerating; MHI and Deloitte report that 82 % of leaders plan to adopt AI by 2029. In the near term, adoption will likely concentrate on warehouses, maintenance and demand forecasting—areas where data is plentiful and the ROI is clear. As physical AI matures and robots gain the ability to perceive, plan and act autonomously, agents will orchestrate fleets of machines to pick, pack, transport and even perform on‑the‑fly manufacturing.
However, full autonomy remains a work in progress. SAP notes that cross‑functional agents are rare in the wild, and experts estimate that only a quarter of identified supply‑chain agents are currently operational. Over the next few years, we should expect a progression from specialised agents with limited scopes to networks of agents collaborating across functions—creating self‑healing, adaptive supply chains. Integrating agentic AI with decision intelligence, digital twins and sustainability initiatives will further amplify its impact. As Jensen Huang of Nvidia puts it, “now is the time for robots”, and agentic AI is the software layer that will give them purposeful autonomy.
Practical example: data, models and outcomes
To illustrate how agentic AI operates, consider a simplified example in table form. A supply‑chain team wants to autonomously manage inbound materials, warehouse operations and outbound deliveries. The table below shows typical data sources, the AI models/agents applied and the actions or outcomes.
| Data sources | AI models/agents | Actions & outcomes |
|---|---|---|
| Inventory levels, purchase orders, production schedules | Inventory‑planning agents using specialised language models & reinforcement learning | Optimise reorder points, adjust procurement plans, allocate stock to meet service targets |
| IoT sensor data from machines, maintenance logs | Predictive maintenance agents with time‑series forecasting and anomaly detection | Identify equipment failures before they occur, schedule repairs proactively, minimise downtime |
| Real‑time order data, transport availability, weather and traffic feeds | Logistics & routing agents using constraint‑solving and route‑optimisation algorithms | Dynamically select carriers, reroute shipments to avoid delays, reduce empty miles |
| Supplier performance metrics, ESG scores, risk indicators | Supplier‑management agents leveraging knowledge graphs and graph neural networks | Evaluate suppliers, flag compliance issues, automate contract renegotiation |
| Social media sentiment, market signals, promotional calendars | Demand‑forecasting agents combining generative models with external data ingestion | Update demand forecasts continuously, inform production planning and marketing strategies |
This is not an exhaustive list, but it demonstrates how diverse data and tailored models converge within agentic systems to deliver concrete actions.
AI in the Chain Insights
Agentic AI promises to turn today’s reactive supply‑chain applications into proactive co‑workers that anticipate disruptions, collaborate across functions and drive continuous improvement. Yet the journey from chatbots to autonomous agents will be messy and iterative. Early adopters must navigate technical integration challenges, design new workflows and cultivate a culture that trusts AI to make decisions.
Our view is that supply‑chain leaders should move deliberately but boldly. Start with narrow, high‑impact use cases where data is available and outcomes are measurable. Use digital twins and simulation to validate agent behaviour before letting it control real‑world processes. As you scale, invest in architecture and governance to ensure agents operate safely and transparently. Above all, remember that agentic AI is a tool to empower people, not replace them: the most successful deployments will pair human judgment with autonomous execution.
The gen‑AI paradox has shown that simply sprinkling AI on top of existing workflows does not deliver value. Agentic AI invites a more radical rethink—one that could usher in supply chains that see, decide and act on their own. Leaders who embrace this shift thoughtfully will not only gain a competitive edge but also shape the future of work in supply chain and beyond.
Sources
- McKinsey, “Seizing the agentic AI advantage” (June 13 2025) – outlines the gen‑AI paradox and explains why agentic AI and new operating models are needed.
- MHI Solutions Magazine, “Agentic AI: The Next Big Thing?” (June 9 2025) – provides real‑world examples of agentic AI in warehouse operations and shares Gartner’s forecasts for adoption.
- Business Wire, “New MHI and Deloitte Report Focuses on Orchestrating End‑to‑End Digital Supply Chain Solutions” (March 19 2025) – summarises investment levels, adoption rates and technology trends from the 2025 MHI & Deloitte industry report.
- SAP, “Agentic AI in the global supply chain” (2025) – explains the differences between generative and agentic AI and describes use cases at Walmart, Amazon and DHL, along with the phases of agentic development.
- WTW, “Transforming supply chain sustainability and risk management using AI” (February 2025) – discusses how AI (including agents) improves risk management, sustainability and compliance via knowledge graphs and lifecycle assessment tools.
- Bluecrux, “Decision intelligence, simulation & agentic AI: Gartner trends” (July 2025) – summarises Gartner’s 2025 supply‑chain technology trends, including decision intelligence, intelligent simulation and agentic AI.
AI in the Chain 
Leave a comment