Last year I held a session at the 2024 Parcel Forum conference. (Watch here.) It was a deep dive into what was real versus hype in August 2024.
My conclusion was that a lot of the AI technology available at the time was simply too early to make a meaningful difference in the shipping industry, and instead the biggest real-world impact was found within incredible optimizations that Machine Learning was giving operators.
ML is best thought of as an optimization tool. Optimization is ultimately a big math problem. Find a decision that requires trade offs, turn those trade offs into models, and pit the models against each other to reach the most optimal decision for your business at that exact moment time.
In 2024, ML was killing it with optimization. I gave several examples:
Meanwhile, the only real evidence of AI making an impact in supply chains was with generative use cases: examples like updating product SKU descriptions, customer support chat bots, and so forth.
Large Language Models (LLMs) hadn't yet matured to a point where workflows were being impacted, in particular machine-to-machine workflows.
It's now August 2025. Wow, have things changed!
There were three mega trends that I observed over the last year.
The speed at which LLMs are innovating is staggering. Since August 2024, the leaders (OpenAI, Anthropic, Google, etc.) released 26 major model updates. That number is astronomical.
Basically, they are doing complete rewrites and updates to the underlying technology. The average is about one a quarter. That's staggering.
The speed of LLM model improvements has meant that the technologies built on top of them can do more. This leads to observation number two.
There is now real evidence of agentic workflows addressing real use cases and automating parts of enterprise supply chains.
Last year most everything I saw was theoretical. Most messages were hype, and those who were building things were butting up against the limits of the maturity of the AI technologies that existed in 2024. Software chaining decisions from one controller to another is not novel. The gap was the absense of actual AI powering the decisions.
I have now seen enough evidence from partners, the market, and our internal team to say that the first applications of agentic workflows out in the wild are very real, and we will see an incredible pace of development and expansion in the years ahead.
The way I would explain the shift is the shift to "always-on" 24/7 decision making that sheds the old ideas of static configurations, and instead makes the decisions based on AI inputs that then acts like an associate on behalf of the operator.
A simple example is with post-order delivery experience: An agent can identify that a shipment is probably going to be late well before we used to be able to do that ("Look at bottleneck at this FedEx Memphis hub!"), and do things like reroute an order to hit a delivery promise or reach out to customers who will likely get a late shipment to offer things like a repurchase option based on the type of SKU.
There are now dozens of examples of very interesting agentic workflows up and down the supply chain that will make it to enterprise supply chains in the next year.
In many ways "adoption" is now the throttle, which brings me to observation number three.
The largest shift in technology, culture, economics, and output over the last year was found in the engineering community. How developers develop has changed forever, and I observed it myself with our team.
The tools have matured so fast and so well that R&D orgs can move at a faster pace with more output.
The trick has been a complete rewiring of development workflows in order to bake AI solutions into the human-driven process of creative production.
On the frontend, we have changed the way in which we scope work and manage product by making those processes AI-first. Engineering teams have changed the way they prototype, build, and test software. Documentation is more automated. And so on.
To me, this is the biggest observation and trend to apply to the supply chain industry. The companies who empower operators to completely change their workflows to an AI-first approach will actualize the benefits I'm seeing in points one and two.
Part of the resistance is going to come from how hard it is to put AI at the center of workflows. It's a cultural change as much as a technical one.
I'm very interested to see how this plays out over the next year. If the last 12 months is any indication, it will be totally different!