Explaining Agentic AI
Hello, and welcome to this week’s edition of Straight Talk. Inside, we discuss:
If this was forwarded to you and you wish to receive it each Sunday morning in your email, please go here to sign up for SCMR newsletters.
Understanding Agentic AI
The world of artificial intelligence is changing fast—too fast for some to keep up. Businesses that just three years ago were exploring how to implement AI saw a quick pivot in 2023-24 to learn all about Generative AI. Welcome to 2025, and while Gen AI is still in its infancy, it is quickly being overwhelmed by the excitement of Agentic AI.
So, I decided to seek out an authoritative source on Agentic AI. Naturally, I asked AI. Here is what ChatGPT says:
“Agentic AI refers to artificial intelligence systems that can operate autonomously, make decisions, and take actions to achieve specific goals without constant human intervention. These AI agents can perceive their environment, plan and reason through complex tasks, and adapt based on new information.”
Sounds scary, right? AI systems making their own decisions. How many are saying to themselves, ‘yes, sign me up for that.’ I needed a better source, and that led me to Kevin Frechette, co-founder and CEO of Fairmarkit.
I first became acquainted with Fairmarkit when I listened to a presentation by Fairmarkit’s?Erin McFarlane, vice president of operations, and ?Tiffany Andrews, senior manager at?Boeing’s?Indirect Procurement Center of Excellence, at last year’s ISM World Conference in Las Vegas. The pair were discussing Boeing’s use of Fairmarkit’s global autonomous sourcing platform. (You can read about that here). So, naturally, Fairmarkit seemed like a good place to start to decipher the Agentic AI vs. Gen AI debate.
The buzzwords
Gen AI. Agentic AI. AI Agents. What does this all mean?? “The words will get tossed around way too much over the next 6 to 12 months,” Frechette told me. “You’ll hear people say Agent AI, AI agents, Agent Networks, Agent Layers, and Agent Architecture. So just get ready for a bunch of buzzwords. I think one way to do it is to start to break down.”
Frechette made a comparison to a three-layer birthday cake. The first layer is the data, the second layer is the AI algorithms that produce predictable, consistent outputs. The top layer is the Agentic AI layer, which is running tasks with minimal human supervision.
“Think about a difference between an agentic workflow and just a normal workflow with Gen AI,” Frechette explained. “[In] a normal workflow with Gen AI, you might go to chat GPT and say, write a paper on this topic and whatever the topic is, it’ll write the paper. It’ll take a little bit of time, and then you can interact with it. So then you’re giving it further directions of how you want it refined. Where in an agentic workflow, an example would be to write an outline on this topic, do some research, write a first draft, consider what needs to be revised, research that, and then rewrite it and continue that process. You’re giving it the instructions to then go take different actions.”
?
Use cases
Frechette says there are many possible use cases for Agentic AI, but he explained one related to the world of procurement, which is where Fairmarkit specializes. An AI would be able to draft a scope of work, for instance. By giving Gen AI access to different data or documents, it is able to put together that scope of work. But, it can’t make revisions based off policies and procedures. That is the work of Agentic AI, he says.
“Say you want to set up a sourcing event, and right now Gen AI could say, alright, we think based off of a lot of historical data that we have that’s proprietary, we think that there’s not enough information for a supplier to bid,” Frechette says. “So, give it a very specific example and it can identify that, but what it can’t do is then take action on that. From an Agentic AI standpoint, it can understand there’s maybe not enough information, and then what does it do? It can actually go back and talk to a request or someone in the field, someone at a facility [that can] clarify and make suggestions … to get to better data quality. Is that correct? Can you give us your feedback? And it can operate in natural language. Then once that end user at that factory confirms it, it can go back to that sourcing event, update it, and then pass it on to the next agent.”
Frechette says Agentic AI will excel in areas where there are tasks that can be carried out over time or multiple tasks that require multiple updates. “It’s a little less of just a tool, which I think is more of when you’re thinking Gen AI, it is as a tool. The agent is the multi-step process that you can run on top of it.”
领英推荐
?
Humans still involved
Even when Agentic AI is deployed, Frechette says humans are still a critical part of the equation. Humans are helping training the models. The AI is using millions of data points and looking at historical events, but humans are still guiding the models.
“Say a company wants to negotiate with a supplier, you can actually give instructions to a negotiations agent to say, this is what I’m trying to optimize for. That could be price; that could be lead time; payment terms. There’s a lot different options,” Frechette says. “You can also give instructions on how I want [the AI] to operate. So it could be, I want you to operate with this tonality in this way. You can also give context about the supplier you’re looking to negotiate with. So you essentially give this agent a bunch of information and then it will go and it can engage with those suppliers. Maybe just not one. You can do 50 at once.”
The Agentic AI is then able to collect the information, and assess whether information is missing that prevents a fair evaluation, and automatically generate follow-up messages seeking the remainder of the information. Risk mitigation is an area where this approach can be very effective, Frechette notes.
There are many more possible use cases for Agentic AI. You can listen to the full conversation, including more of those use cases, and a more detailed explanation of Agentic AI’s differences with traditional AI models, in my Talking Supply Chain podcast episode with Frechette. Listen here.
No DeJoy at USPS
Postmaster General Louis DeJoy has informed the U.S. Postal Service board of directors last week that he would be stepping down. “While there remains much critical work to be done to ensure that the Postal Service can be financially viable as we continue to serve the nation in our essential public service mission, I have decided it is time to start the process of identifying my successor and of preparing the Postal Service for this change,” DeJoy said in a statement. DeJoy’s tenure (he was appointed during the first Trump administration) has been marked by controversy, including slowing postal delivery, raising rates, and cost-cutting including service cutbacks that critics argue have hampered service. USPS continues to lose money. DeJoy initiated a 10-year turnaround plan that targeted a return to profitability in 2024, but instead the agency posted widening losses of nearly $10 billion in 2024, up from $6.5 billion in 2023. A timeframe for DeJoy’s departure was not announced.
Risk awareness
Everstream Analytics has analyzed its trove of data on risk events and come up with its top 5 for 2025, assigning each a risk score—the higher the score, the more priority it believes an organization should assign to mitigation efforts. So what are the top 5? Topping the risk is climate change (90%), followed by geopolitical instability (80%), cybercrime (75%), access to rare metals and minerals (65%), and crackdown on forced labor (60%). You can download the report here and get more insights on each risk, including strategies for mitigation efforts.
?What I read this week
Chief supply chain officers are increasingly being tasked with contributing to the company’s growth, and that means more focus on customer experience. … Consultant Norman Katz has penned an article series on how businesses can achieve the holy grail of shipping—delivering the Perfect Order. … According to date from Akeneo, 65% of holiday shoppers returned or exchanged at least one item this year, adding pressure to stressed reverse logistics supply chains. … MIT researchers have unveiled the MiFly drone, which relies on radio frequency to maneuver, potentially making it more useful in often dimly lit warehouses. … Blue Yonder has launched a new supply chain program with the University of Arkansas. … The Cass Freight Index indicated a drop in freight shipments in January.
Thank you for reading,
Brian