Jeff Jones, senior customer manager at Made4Net, points out that agentic artificial intelligence with actionable, decision-making capabilities is becoming reality in today's warehouse. Jones characterizes agentic AI as a "self-healing system" that takes over non-value-adding human interaction in handling exceptions and alleviating challenges to ensure shipments depart on time. From a supply chain perspective, Made4Net delivers integrated solutions across WMS, YMS (Yard Management System), TMS, LMS (Labor Management System), and delivery applications through its SCExpert platform. HighJump and Tecsys are competing players offering similarly comprehensive solutions in the mid-market segment.
This does not mean human oversight is absent. Jones states, "The use of agentic AI is built on policy, on the rule set employed and entered by a human as to how they want to operate their system." Jones adds, "Agentic AI can actually make adjustments or handle these exceptions as the day-to-day volatility of orders comes into the warehouse, and it can handle that more efficiently than a human can." From a supply chain perspective, the policy-as-code approach requires the use of frameworks like Open Policy Agent (OPA) in WMS integration. Exception management encompasses real-time handling of daily operational deviations such as order short, damaged inventory, stockout, carrier no-show, and dock door blockage. Automatically addressing special cause variations by AI according to Six Sigma and DMAIC methodologies is critical for maintaining high OEE (Overall Equipment Effectiveness) scores.
It is unfortunate that more companies are not using agentic AI now; Jones states, "We're on the journey now, but most companies have only moved from predictive to generative AI, not agentic." However, bad data is a barrier even to agentic systems. Jones says, "Bad data is notoriously common inside of warehouse systems. So until we find a way to maximize the cleanliness and purity of the data, I think that's going to prevent us from seeing the full benefit of agentic AI." From a supply chain perspective, master data management (MDM) requires maintaining consistent and clean hierarchies of SKU, UoM (Unit of Measure), HU (Handling Unit), location, and vendor. Data lineage, data quality scoring, and data observability implementations are managed through platforms such as Informatica, Talend, Collibra, and Monte Carlo, and serve as prerequisites for the success of AI projects.
Jones notes that unlike stories of malicious actions taken by AI in the consumer space, such issues do not occur in supply chains. Jones states, "I think we're seeing it operate in the confines we define it to operate in. It doesn't have freedom to roam about however it wants to. It's just multiple agents sitting on top of different policies, procedures and processes inside the warehouse to make exception decisions faster than a human can." Advances in AI and much of warehouse automation are enabling shorter training periods. Jones notes that 30-45 day training programs are expected to be reduced to a single day. From a supply chain perspective, workforce onboarding is a critical bottleneck for 3PLs and retail distribution centers during seasonal peak periods (Black Friday, Cyber Monday, Prime Day). Vision picking with AR (Augmented Reality) glasses and voice picking systems reduce the learning curve and lower error rates. Vuzix, RealWear, and Honeywell Vocollect are leading hardware providers in this segment. In conclusion, Jones's vision signals that agentic AI will rapidly proliferate in warehouse operations, contingent on data quality conditions.
Key Takeaways:
1. Jones defines agentic AI as a "self-healing system".
2. AI operates according to policies and rule sets entered by humans.
3. Most companies are transitioning from predictive to generative AI, not agentic.
4. Bad data prevents the full benefit of agentic AI from being realized.
5. AI can reduce training periods from 30-45 days to as little as 1 day.
