Augmenting human capabilities and driving future supply chains with Agentic AI

By Vinay Sharma and Ahish Bindumadhavan

The wave of AI technologies is driving both excitement and skepticism in the supply chain professional community. The promise of catching the excitement of a rogue wave fueled by numerous possibilities of democratization and autonomy from new knowledge, solving complex supply chain problems in seconds, driving decision-making, and enabling process and technology adoption at scale. However, it must, at the same time, address the skepticism arising from governance and regulatory challenges, increased AI skill gaps, security, data availability, and sustainable use of technology.

Among the many variants of artificial intelligence (AI), we will examine how Agentic AI promises to augment human capabilities and create a hybrid workforce of the future (human and machine) to drive the accelerated supply chain transformation journey. We will explain the nature of the agent, the architecture of an agent, and the evolution of the Agentic AI supply chain, and compare and contrast the impact of agents on the activities and point of view of various supply chain stakeholders.

What is an intelligent agent?

An intelligent agent can be thought of as a “stateful” software entity capable of achieving complex tasks with advanced learning and limited direct supervision. It can decompose complex tasks, reason based on its progress and adjust its solution approaches over multiple steps. The defining characteristic of an intelligent agent is its probabilistic and highly adaptive behavior in response to dynamically changing environments and goals.

Components of an agent within an Agentic AI system

Orchestrator. The primary task of the orchestrator is to coordinate the activities across the various components that make up the agent. The orchestrator not only receives inputs from other agents or users but also facilitates interactions among the planner, executor, and memory components of the agent.

Planner. The primary task of the planner component is to decompose the request received from the orchestrator and reduce it into a sequence of manageable tasks and agent components. The planner has intimate knowledge of the various authorized systems that can be used to accomplish the tasks and leverages a Large Language Model (stateful) to identify the sequence. It then associates the tasks with the respective systems.

Executor. The executor agent component is designed to interact with other systems and executes the tasks sequentially as specified by the planner. The systems it interacts with could be other agents, enterprise databases, LLMs, etc. that comprise the supply ecosystem commerce domain.

Memory. The memory component provides memory services to other components. It acts as a store of past decisions and as a cache for the current activities of the other various components/agents.

(Photo: TCS)

Supply chain as a network of agents

Traditionally, the supply chain has been viewed as a network of sequential, autonomous nodes connected transactionally. Absent digitalization, the physical nodes in the network typically operate in silos and interface with each other locally. Significant information asymmetry results in locally optimal decisions. The next stage of supply network maturity is the development of digital twins operating in an ecosystem network of digitally connected nodes. The digital twins enable the interaction of the physical and the digital dimensions across the entire supply chain network capturing the essential dynamics of the supply network’s ecosystem commerce data and processes. However, these digital twins struggle with real-time data sharing across systems, and the resultant supply chain, while supported on a strong technology backbone, is still largely human-driven.

New-age Agentic AI technologies leverage and augment the development of digital twins and robotic process automation (RPA). Unlike traditional AI learning models that simply respond to inputs based on predefined patterns, Agentic AI systems can react autonomously to plan actions, initiate responses, and offer resilience to the dynamics of ecosystem supply networks based on intended and unintended causes arising from changes to the market environment.

Evolution of Agentic AI supply network management

Future agents trained in enterprise supply chain network knowledge will support business stakeholders in their decision-making. As agents scale in their capabilities, the emergent needs of the stakeholders can only be met by enabling inter-agent communication. Agentic AI technology enables the transition from a human-driven supply chain to a human-centered autonomous supply chain network.

(Photo: TCS)

Architecting a future-ready network of supply chain agents

The digital supply chain network of tomorrow can be viewed as an ecosystem network of supply chain agents. These autonomous agents can sense and learn from the external environment, collaborate with each other to make decisions and act upon them. These Agentic AI supply chain networks will be flexible, scalable, robust, and fault-tolerant. Such multi-agent systems will have the following architectural components.

1. Agents. Individual intelligent agents trained in supply chain domains such as order fulfillment, demand planning, procurement, supply planning, transportation and distribution, logistics and operations planning, etc.
2. Environment. The ecosystem universe from which the multi-agent system receives data to build models. Agents are trained to scan the environment to sense conditions that may affect the flow of goods and associated responses. The result is a resilient, “sense-and-respond” autonomous ecosystem supply network. While the model determines the actions and decisions to respond to fluctuating market dynamics, the ecosystem environment involves human engagement with agents to collaborate to resolve exceptions.
3. Communication framework. The agents communicate with each other through standard agent communication languages. Supply chain agents’ communication typically involves actions such as informing, requesting, queries, etc. The communication framework also enables interactions between agents. Based on the goals, the interaction between agents involves cooperation, coordination, and/or negotiation. For instance, the demand and supply agents can be involved in negotiations in which the agents collaborate to make and accept demand and supply proposals and can even terminate the negotiation process if the goals defined by the human user are unmet.

Agent network structure/architecture. Depending on their use, the agents can be organized in a hierarchical or decentralized structure. A typical hierarchical layout is a control tower where a master agent controls and drives the activities of multiple other agents. A decentralized agent network/architecture can be effectively leveraged to solve emergent problems in the supply chain network. For example, a network /architecture fosters resilience as it is flexible and adaptive by its nature. Newer agents can be plugged into the existing Agentic AI supply chain without significantly affecting the real-time operations of the existing supply chain. The division of labor in the Agentic AI supply chain mirrors that of the real-world supply chain network. The greater efficiency of the Agentic AI supply chain network vis-à-vis physical supply chain network is due to the elimination of information silos and efficient, faster collaboration between autonomous agents in dynamic environments.

(Photo: TCS)

Role of the supply chain leaders in the multi-agent supply chain system

Future supply chain leaders will have to transform from current supply chain business models to new  Agentic AI models that manage autonomous agents to drive automation, advanced decision-making, and organization structures that foster a culture of collaboration and continuous improvement. The table on Page 16 summarizes the change in the life of supply chain leaders in the future.

(Photo: TCS)

Implementing a supply network with agents and humans in the loop: The future is now!

Future supply chain managers will need to work collaboratively with supply chain agents to accelerate decision-making and eliminate bottlenecks to optimize supply chain performance. They will have to monitor the behavior of the Agentic AI supply network, define guardrails and rewards for agent behavior, and provide active guidance, based on their experience, to agents to achieve defined outcomes.

Such dynamic interactions between human managers and agents in multi-agent supply chain systems will act as iterative feedback loops enabling these systems to refine decisions and adapt to complex business environments. Supply chain managers must also guard against unintended consequences of agent decisions and must implement decision-making within an evaluative framework comprising agent recommendations, tacit knowledge, ethics, and business goals.

Managers must ensure that a multi-agent supply chain architecture is designed to facilitate the explainability of agent actions and audit of inter-agent interactions. For example, in the context of supply chain planning, the manager must be able to evaluate the decision-making process of the planning agent—the number of supply nodes explored to satisfy demand, the trade-offs made between levels of inventory and overall service levels of the network, etc.

For all their promise and abilities, agentic supply chains are human-made systems which embed the knowledge in the form of models. Managers cognizant of their capabilities can augment their decision making to build robust, resilient, adaptive and self-learning supply chains of the future.

About Global Links

Global Links appears in each issue of Supply Chain Management Review. Richard J. Sherman, retired guru of SCM, is the Global Links column editor. If you are interested in participating in the column, he can be reached at [email protected].

About the authors

Vinay Sharma is a partner in TCS’ Global consulting practice, supply chain transformation group, and can be reached at [email protected].

Ahish Bindumadhavan is a business consultant in TCS’ Global consulting practice, supply chain transformation group, and can be reached at [email protected].

SC
MR