AI agents are becoming the new back office of supply chains
AI agents in supply chain operations are specialized software assistants that autonomously execute repetitive procurement, logistics, and supplier management tasks end to end, reading and interpreting unstructured documents, updating core systems, and escalating only exceptions to humans, which reduces manual effort, improves data quality, and accelerates decision-making across procurement workflows. AI agents are no longer a lab experiment; they are quietly taking over high-volume, low-judgment work across real-world supply chains. The question is not whether they work—it is what happens to organizations that still treat them as a side project while their competitors automate thousands of transactions a day. Two recent enterprise AI deployments show how fast the shift is happening and why the impact is too large to ignore.
Lemvigh‑Müller: 100,000 order confirmations no longer need humans
Order confirmation automation used to be the graveyard of digital projects: messy PDFs, inconsistent formats, and countless exceptions. Lemvigh‑Müller decided to stop forcing one monolithic system to do everything and instead broke the work into multiple AI agents built on SAP Business AI. One agent reads incoming emails and attachments, another structures PDF data, and a third compares it against SAP purchase orders in a single procurement workflow automation pipeline.
The payoff is blunt: more than 100,000 supplier order confirmations are now processed automatically, with faster processing, better data quality, and more accurate delivery information for customers. What used to take hours or days to surface—such as delays, quantity changes, or price discrepancies—is now detected almost immediately, so customers get an accurate delivery picture far sooner and issues are fixed before the final invoice. This is AI agents supply chain work in its most practical form: the routine is automated, humans focus on the difficult edge cases.
From failed RPA to agent-based procurement workflow automation
The most important lesson from Lemvigh‑Müller is not the technology—it is the path they took. The company had already tried RPA and traditional automation without the desired effect. The bottleneck was unstructured supplier PDFs where small discrepancies in price, quantity, or delivery dates triggered heavy manual work in procurement. Instead of giving up, an AI‑savvy project manager ran a low-key experiment, matching an order confirmation to a purchase order using a conversational model, and then escalated the idea internally.
From first test to enterprise AI deployment took 10 weeks, not years. The solution, implemented with an external partner and fully integrated into the SAP landscape, is expected to free resources equal to three to four full-time employees and deliver ROI in quarters rather than years. In other words: when AI agents are scoped tightly and wired into existing systems, procurement workflow automation stops being a PowerPoint promise and turns into tangible capacity and speed.
T‑Systems and SupplyOn: AI agents for 140,000 connected companies
While Lemvigh‑Müller shows depth in one process, the partnership between T‑Systems and SupplyOn shows breadth across an entire industrial network. SupplyOn connects around 140,000 companies in more than 100 countries and manages core processes like procurement, supplier management, logistics, and risk management. By linking this platform to a sovereign Industrial AI Cloud, AI agents can now take over key tasks across the entire supply chain while data stays under strict protection and within regional control.
This is not a modest infrastructure play. Since February 2026, the Industrial AI Cloud facility in Munich has been running 10,000 NVIDIA Blackwell GPUs, delivering 0.5 exaflops of computing power and 20 petabytes of storage for enterprise AI deployment at scale. On top of that hardware, companies can automate and accelerate procurement, logistics, and supplier management processes with AI while retaining full control over their data. If AI is becoming “the operating system of modern supply chains”, this stack shows what that OS looks like in practice.
The new competitive line: agentized workflows or manual drag
These deployments point to a clear conclusion: the competitive line in supply chains is shifting from who has the best people to who gives those people the best AI agents. Lemvigh‑Müller now lets AI agents update procurement data almost instantly, while humans handle the most complex and exception-driven orders. On the infrastructure side, the Industrial AI Cloud and SupplyOn partnership is building a shared platform where AI agents will next expand into planning, quality assurance, e‑invoicing, and risk management.
The risk for laggards is not being “behind on AI” in some abstract way; it is carrying a permanent manual drag on procurement efficiency and order processing speed while competitors automate the same tasks for thousands of suppliers. Each year, Lemvigh‑Müller handles about 175,000 purchase orders to more than 2,000 suppliers. At that scale, leaving AI agents on the sidelines is not caution. It is a strategic handicap.
