By Prof Avijit Raychaudhuri, Faculty member of Supply Chain & Operations Management, Indian Institute of Management, Udaipur
Picture a hypothetical. Sales of your company’s products have suddenly experienced a steep fall within an area, for no obvious reason. It now falls on you to get to the bottom of the issue and take corrective actions. Most likely, identifying the root cause(s) would involve significant time and effort spent traveling across your zone and interacting with various stakeholders in person. Sounds like a lot of work, right?
Well, fret you not, Agentic AI is here, needs no rubbing, and grants infinite wishes! It is like having your own personal detectives to whom you can just describe the task, and they will do their AI black (-box) magic and tell you the root cause of the problem. If you provide enough autonomy to the agents, they will even take relevant corrective actions on their own without your intervention!
In a nutshell, Agentic AI can provide autonomous assistance to you for mundane tasks (such as managing your personal calendar) to complex tasks (such as seen in the hypothetical) without active prompting from you every step of the way. Its core appeal is that the business applications of AI no longer need to be constrained by human choice, thereby helping AI assist humans at blitzkrieg pace!
Or so it seems!
Let us revisit the hypothetical. When it is imperative to travel for business, would an AI agent be able to physically travel, identify key people, converse with them, and eventually zero in on the root cause(s)? Surely not! Moreover, how you identify the key people and interview them depends a lot on your personality and your personal rapport (if any) with them, among other non-quantifiable attributes. Agentic AI cannot replace such attributes until the day that they become digital clones of people.
Okay, but what if Agentic AI could indeed “do everything”? Then, at the very least, the AI agent should be able to identify the appropriate first points of contact from your contact list, call them, extract contact details of people who might not be connected to you but are relevant to solving the problem, contact them, interview them, and collect all necessary information.
Obviously, this also implies that the AI agent is intelligent enough to mirror you with respect to the non-quantifiable attributes. Let us also assume that to make up for the agent’s lack of physical mobility, you have all the relevant physical locations covered by cameras either having edge-computing capabilities or connected to some centralized system. All these would have to simultaneously hold for an AI agent to address the hypothetical.
Notice the issue here?
The issue is that of a digital bottleneck. Since an AI agent is inherently digital, its reach is also limited by the extent to which your company and your entire supply chain are digitally covered. For the AI agent to be able to do everything, you must ensure that not only your AI agent has access to all the latest data being generated throughout your company at all times, but also your supply chain partners are completely digitalized and willing to share data generated by their company (within the scope of your supply chain, of course) at all times!
Simply put, for companies to experience the true promise of Agentic AI, all companies in all industries must be digitalized to the fullest extent possible, and all supply chain partners must share all data amongst themselves at all times. Essentially, all supply chain partners within every industry would need to get vertically integrated first!
Hopefully by now, the prospect of Agentic AI upending your business sounds ludicrous enough to force you to pause and reflect.
So, is Agentic AI a farce? Certainly not. It’s very real.
Case in point: until very recently, global media was abuzz with discussions on prompt engineering, whereby one tries to “engineer” appropriate prompts for Generative AI platforms to produce desired outcomes. Today, the discourse has shifted to Agentic AI which is supposed to require only rudimentary prompting, on the lines of how you would interact with your team for performing a complicated task.
So, what should you keep in mind while investing in Agentic AI?
If you follow an AI-first mindset and invest heavily in Agentic AI first, then you will most likely find your company’s digital architecture to be the bottleneck for information flow. This will not only prohibit your investments from fetching commensurate ROIs, but likely spell disaster for any further digital transformation.
First and foremost, data cannot be the new oil unless it has adequate pipelines to ensure smooth flows. With this in mind, create a digital architecture mapping the network of data pipelines within your company. Furthermore, negotiate data-sharing agreements with your supply chain partners, and then integrate the data pipelines generated from the successful negotiations within your digital architecture.
This integrated digital supply chain architecture can now act as the constraint (at the supply chain level) towards driving ROI-driven investments in Agentic AI. Post deployment, the architecture can be continuously updated to reflect the operational constraints that will become visible only during actual implementation.
Second, temper your expectations. Instead of taking a FOMO-based plunge, take an ROI-based approach.
Third, identify the use cases of automation within your organization. In the short to medium term, ROIs from Agentic AI will largely be realized within the scope of automation, especially of low-level operations.
Fourth, understand how the different supply chain flows of materials, money, and information could be impacted by agentic capabilities. Creating such a map will help you identify appropriate use-cases for Agentic AI and estimate the long-term ROIs.
Finally, develop in-house analytics and AI talent, and foster an experimentation-based culture. While core agentic capabilities will be mostly built by the likes of OpenAI and xAI etc., more and more IT services companies will pivot toward using the AI tools from such companies into configurable solutions for their clients. Complete reliance for your agentic (and AI, in general) capabilities on vendors might reduce short-term costs but will prohibit you from realizing long-term ROIs resulting from unconstrained experimentation.
– Prof Avijit Raychaudhuri is a faculty member of Supply Chain & Operations Management at the Indian Institute of Management Udaipur. He was the founding Chairperson of the Doctoral Program Committee at IIM Udaipur. He is passionate about solving business problems using digital technologies and analytical methods, and in general about a wide gamut of issues related to global supply chain management and strategy, interface of operations and marketing, and corporate sustainability.