AI and the Autonomous Supply Chain | Chris Burchett, Blue Yonder

Chris Burchett, Blue Yonder

In this interview from Blue Yonder AI and the Autonomous Supply Chain, Chris Burchett, senior vice president of AI at Blue Yonder, joins theCUBE's Dave Vellante to discuss how AI is transforming supply chains from reactive, fragmented operations into predictive, autonomous workflows. Burchett leads with Blue Yonder's scale — 27 billion machine learning predictions per day, roughly double Google's global search volume — before explaining how four years of cloud-native development now delivers one common data model across planning, warehousing and logistics. He details how a growing portfolio of AI agents is tackling decision latency head-on, from resolving warehouse allocation shorts to optimizing backhaul routing and automating freight auditing, elevating human workers to focus on judgment and strategy. The conversation also explores Blue Yonder's NVIDIA partnership and the strategic rationale behind building a proprietary model training factory. Burchett explains why frontier models alone create token bloat and reliability gaps in high-frequency supply chain operations, and how training 30B-parameter Nemotron open-weight models — roughly 1/60th the size of leading foundation models — delivers faster, cheaper "owned intelligence" for inner-loop agentic tasks. He details Blue Yonder's responsible AI framework, which embeds data sovereignty, synthetic-only model training and EU AI Act compliance into its development lifecycle, and describes how trust in autonomous agents is earned incrementally through threshold-based approvals. From compressing customer onboarding from months to 72 hours to a roadmap for multi-agent collaboration across warehouse, logistics and inventory operations within the next 12 to 36 months, Burchett outlines how Blue Yonder is positioning itself at the center of the shift from manual planning to machine-speed supply chain autonomy.

Share this session

Clips
More from AI and the Autonomous Supply Chain

Chris Burchett

SVP of AI

Blue Yonder

play_circle_outline Supply chain complexity and fragility due to fragmented data and dynamic global factors

play_circle_outline AI agents (warehouse, logistics, inventory) moving from monitoring to autonomous actions

play_circle_outline Cloud-native architecture enabling common data model and end-to-end supply chain platform

play_circle_outline Closing insight-to-action gap: faster decision-making and reduced decision latency

play_circle_outline Owned intelligence vs rented intelligence — hybrid model strategy to optimize cost and latency

play_circle_outline Focus use cases: intraday warehouse operations, short allocations, exceptions, trailer delays

play_circle_outline Responsible AI: governance, explainability, data sovereignty, synthetic training data

play_circle_outline End-to-End Agent Orchestration: Interoperability, Rising Autonomy, New Jobs, and Enterprise Integration

Info
Transcript

Chris Burchett, Blue Yonder

Chris Burchett

SVP of AI Blue Yonder

In this interview from Blue Yonder AI and the Autonomous Supply Chain, Chris Burchett, senior vice president of AI at Blue Yonder, joins theCUBE's Dave Vellante to discuss how AI is transforming supply chains from reactive, fragmented operations into predictive, autonomous workflows. Burchett leads with Blue Yonder's scale — 27 billion machine learning predictions per day, roughly double Google's global search volume — before explaining how four years of cloud-native development now delivers one common data model across planning, warehousing and logistics. He... Read more

explore Keep Exploring

What makes supply chain management so challenging, and why is AI considered potentially transformative for addressing those challenges? add

What recent updates did you announce at ICON regarding your agentic/cognitive platform, and how are these agents being used in your operational workflows? add

What does it mean for a supply chain solution to be "cloud native," and why is that beneficial? add

How do agentic AI systems benefit customers in terms of decision-making speed and operational resilience in warehouse and logistics environments? add

What do you mean by "owned intelligence," and how do your agents, model training factory, foundation models, and partnerships (e.g., with NVIDIA) work together to deliver value to customers and mitigate the limitations and costs of third‑party models? add

How do you manage intraday warehouse operations to prevent and handle inventory shorts and ensure on-time, in-full fulfillment? add

How do you integrate responsible AI principles (human-centered design, transparency/explainability), data security and privacy, and regulatory compliance (e.g., the EU AI Act) into your software development lifecycle, and how do you ensure data sovereignty and that customer data is not used for model training? add

How will autonomous AI agents change supply chain operations over the next few years—how will they interact across domains, what is the expected deployment timeline, and what impact will this have on human roles? add

bolt Powered by CUBE AI

Chris Burchett, Blue Yonder

search

Dave Vellante

>> Supply chain has always been one of the most complex and unforgiving disciplines in enterprise technology because it sits at the intersection of physical operations, financial outcomes, of course, human behavior and increasingly volatile global conditions. Every single node in the chain, whether it's suppliers or factories or ports, thick transportation networks, you've got warehouses, you've got retailers and you got end customers as well. Each of these nodes generates massive amounts of fragmented data that often sits in disconnected systems. Even small disruptions through that system can cascade quickly across the entire network, so it creates this domino effect that impacts inventory levels or service quality, end customer satisfaction, and that ultimately relates in lower profitability. So the challenge is made more acute by the fact that supply chains, they are inherently dynamic. So you'll get unexpected shifts in demand, weather, you've got geopolitical events, these can all alter sourcing strategies that can impact the availability of labor. And we've recently seen tariffs and regulatory policy just instantly reshape the economics of supply chain. So traditional planning systems, they were largely designed for relatively stable environments and what we like to call deterministic workflows, which left to organizations struggling to react in real time when things change. This is why AI represents such a potentially transformative force for supply chain operations. Unlike previous generations of analytics and automation, AI has the ability to continuously ingest and harmonize vast streams of unstructured and structured data and identify patterns that maybe humans can't easily detect or certainly not as quickly and then generate recommendations or increasingly autonomous actions at machine speed. So in our view, the opportunity is far bigger than incremental efficiency gains. AI has the potential to create a more adaptive and intelligent and resilient supply chain that can sense disruptions earlier, simulate outcomes, actually do so dynamically and optimize decisions across the entire network in near real time. So as enterprises move from reactive planning toward predictive and eventually autonomous operations, firms like Blue Yonder are positioning themselves at the center of what could become one of the world's most important AI-driven transformations specifically within enterprises and today with us to discuss how AI is changing the dynamics of supply chain is Chris Burchett, who's the senior vice president of AI at Blue Yonder. Chris, great to see you again. Thanks for spending some time with us.

Chris Burchett

>> Yeah, and great to be with you, Dave.

Dave Vellante

>> All right, let's get into it. So Blue Yonder, like a lot of tech firms, you've been talking about AI for quite a long time. In fact, I mean, you've been applying some form of machine intelligence, machine learning for decades. You've built up quite a bit in terms of your development efforts, your R&D, both organic and through M&A. Why don't you give us a high level overview of the work that you guys have done to maybe put forth your broader vision for an AI-driven supply chain?

Chris Burchett

>> Yeah. Yeah. Thanks. Thanks, Dave. We have been doing this a long time. You're right. For decades, we've been driving both optimization as well as predictive machine learning, and to the point where right now we deliver 27 billion machine learning predictions per day, and that's about double what Google does search results for the world. So that's a staggering amount really when you consider all of the machine intelligence that's going into that. And what we've been up to is really, we have spent the last four years moving everything to a cloud native platform, to a cloud native architecture. We've now got all of that technology in a place of infinite scale of data and compute. And so we now, for the first time in the industry's history, we have end-to-end supply chain operations and solutions built on one common data model, one common platform. And as a part of that build out, our agents are a huge part of that now with the warehouse ops agents, logistics ops agent and inventory ops agent. And actually, a growing portfolio of agents beyond that across category management, returns, commerce fulfillment, and really much more. And these agents are now beginning to work alongside of humans. They've grown from monitoring and analytics to actually more autonomous actions as well. And what we announced last week at ICON was basically the next step. We announced our partnership with NVIDIA. We announced additional capabilities in the cognitive solutions. We announced even a lot of updates around frictionless adoption. And what that means is we've now made it possible through agentic technology to onboard customers from their on-prem solutions onto this cognitive platform in really just a fraction of the amount of time. We're talking days instead of weeks or months. And so we even had customers on stage showing the benefit of that in just 72 hours. So that was pretty exciting to see and that just catches you up a little bit on some of the things we've been up to lately.

Dave Vellante

>> Yeah, thanks for that. And I want to get into some of the themes at ICON, and I definitely want to get into the NVIDIA announcements, but before we do, I want to double back to something you said. You said now you're cloud native. And when people hear that, you know our industry, we go from like fashion. Every so often, we change the narrative. And so when people hear cloud native, they might think, "Oh, but you have to now be AI native." I'd like you to explain why being cloud native is actually compatible with being AI native and actually can accelerate AI nativity, if you will.

Chris Burchett

>> Yes, yes. No, it's a great point. And the reality is the supply chain has been a world where it's been a system of fragmented applications that have driven the supply chain. I've heard it said that many customers have over 28 different systems. And so being cloud native means, first of all, that you have on place where all of that data and intelligence resides. And that being cloud native also allows you, because you have that one place, you can have one common data model. And all of those applications, now they don't have the problems of syncing up data discontinuities and timings and all of that because they all drive off this one platform. And being cloud native also means you stay current. So it's like any cloud native application, it's like our email, right? Google Mail or whatever. It is an application that stays up to date all the time. We don't know when they update and bring new features, because the applications never go down. And so that's what we mean by being cloud native. It means always available. It means that it's continuous operation and continuous innovation cycles that drive these end-to-end systems that are fully connected and solving problems that don't stop at the planning boundary or at the warehouse boundary. It allows you now to do things that are unified, have decisions that are global optimum instead of individual siloed optimums. And this is really changing the game for customers.

Dave Vellante

>> Right. So you could traverse those silos with a common data model and then it allows you to scale. I mean, you referenced 27 billion machine learning predictions a day. You can't do that with the legacy traditional technologies. All right. One of the themes across the solutions that you guys talked about at ICON is closing the gap between insight and action. So when you talk about cognitive solutions, AI agents, scenario modeling, we've talked about all those things together. What are the practical shifts that you see for customers? I mean, what does it mean for them? They have obviously better visibility, they can make faster decisions, they can be more responsive, but what did you hear at the conference? Tie it all back to customer value.

Chris Burchett

>> Yeah. So customers are benefiting from these systems in several ways. The first one is they really, and I think a lot of it all boils down to speed of decision making. It's not only visibility the customers need, they need to ... I think a big challenge today is decision latency. How long does it take them to make a decision and to respond? And there's a lot of things that have to line up in order for customers to make a decision. And what the agentic AI is doing is as those disruptions that you talked about in your opening happen and as inevitable chaos that ensues when you start your day in the warehouse and a truck is late or a pallet falls over, as those things happen, how do you maintain the decision speed? How do you keep things flowing in such a way that you avoid the congestion, that you're able to still get all the orders out that you need to on time and full? And this is something that our cognitive warehousing and our cognitive planning or transportation management, these systems are now built to be inherently automated and with the human in the loop, but the agents now are able to take more and more actions. And it's things that a human can't really keep in our head. If I have allocation shorts in a warehouse, how do I know all the possible decision trees and locations to look for inventory? I can't keep all that in my head. I need an agent to be able to help me and prioritize things for me and bring me some solutions that I can then approve or change. And this is also happening not only in the warehouse, it's happening in, like I said, in planning with scenario comparison and automatically run that solve for me and bring it back with a comparison to the baseline so I know immediately how to make this call. And it's things also like in the transportation world where you might think about backhaul opportunities and deadhead miles and what are my best options. And is it a two hop or a three hop? And what about the unroutable shipments that I had after my planning run? Can you help me identify why those were and how I can resolve them quickly? All of these are places or freight auditing. Freight auditing is huge, right? And it's one of those things that takes as very time intensive, but it's really important for the business. And so in all of these areas and many, many more, there are opportunities for us to automate more and more of the mechanics of the software and elevate the human to being able to deal with judgments and to be able to have more time to build relationships, think strategically, serve their customers. And this is really where we see the opportunity for our customers.

Dave Vellante

>> Excellent. Let's get into the NVIDIA relationship. I mean, NVIDIA is the mainspring of AI and this AI era. Of course, we're all talking about agents. Your announcement with NVIDIA is specific. Why did you decide that you had a partner on a model training factory with NVIDIA? I'm curious as to how you're using frontier models from some of the leading providers to build supply chain. How does that support it? But why did you feel like you had to do something with NVIDIA that was more specific?

Chris Burchett

>> Right. It really comes out of our last several years and months of working with this technology. And so we started very early on in '23 and as we've built agents and worked with customers, we found that the foundation models, they're really good for a lot of ... They have huge broad knowledge and they can reason in amazing ways and they write code very proficiently, but we also started to see that there were some failure modes. And some of these failure modes were around deep understanding of the supply chain problem itself right there immediately. So you could make them understand, but they didn't get it immediately when you ask the questions sometimes. And they also didn't understand, of course, our APIs and our data model and the surface area for the applications. And finally, they didn't really understand very well sometimes how to present information back to the supply chain expert in a way that a supply chain domain expert would want to see sorted by this column and prioritized in this way and filtered and whatnot. And so all of those things were things that we realized, we were doing a whole lot of context engineering. That's the way you manage the data that you give to the LLM is a huge software engineering problem and people call it context engineering. And all of that context would build up over time and we could get the agents to be reliable, but it was creating a lot of token bloat, if you will, a lot of context and that context that had to be managed, it drove a little latency because, hey, the models have to process every token and it also made it kind of cumbersome at times. So what we realized is if we train a model, we can have a much smaller model that will be faster, better, cheaper at these three dimensions that I laid out where we saw failure modes and have more of a hybrid architecture with the model, with a large model. So we very much believe that the frontier models are key for our success and that's a big part of what we want to continue to get good at and evolve with them as they evolve. But we believe that for the types of operations that are happening in the inner loop of the agent and have to be performed over and over and over millions of times as we make decisions and run supply chains, for those kinds of decisions, it became increasingly important that we had a model that was small, light and efficient and really, really expert at the supply chain problem.

Dave Vellante

>> So I'm probably going to keep you longer than you had expected, because this is opening up some really interesting conversations that we've been having related to some interesting conversations we're having in theCUBE. And we've talked a lot about, you look at the SaaSpocalypse and say, okay, the companies that are going to do well, the SaaS companies are the ones that can take their deterministic software and blend it with the non-deterministic intelligence from the LLMs and specialized models as you're talking about. But in many ways, as you kind of laid out before, you said on average, 28 different systems. So determinism has largely been a myth. You're deterministic within your little silos. What you've done is you've broken down those silos and now you're able to bring, it sounds like the best of determinism. You've made that enhanced specificity with the NVIDIA relationship and you're still tapping the cognitive power of the frontier models who are moving at breakneck speed. So that's kind of a setup. I know in your announcement, you talked about owned intelligence versus rented intelligence. Tie that all together. What does this mean for customers? Did I get it right in terms of bringing together the deterministic world with the probabilistic world and across transportation networks and planning? Take it from there if you would.

Chris Burchett

>> Yeah. Yeah. No, I think you got it perfectly right. And really, the way we think of it is it is that 27 billion intelligence operations, the optimization, the predictive machine learning, the API surface area, all of these things are the tools that empower our agents to be true operators. They really allow us to differentiate the capabilities of the agent, because if we can give the agent all of those tools and it can rerun solves when it needs to and it can rerun some of these optimizations, that's tremendous power that enables the agent to be more free to solve broader and broader problems over time. And so that's a key dimension is the investment. In fact, we have over 400 data scientists and they are always innovating in predictive ML, in optimization technologies to the tune of about 14 patents a month, in fact, that we're continuing to file. And so it's an exciting area of innovation even beyond the agents. And when I take that and now I add it to what you were talking about with NVIDIA and the model training factory, we now have the first fruits and the first instance of a model in the warehouse that is enabling us to apply this intelligence, this owned intelligence in a very specific way to warehouse problems so that we really truly do have faster, better, cheaper as a part of our strategy from a product and an intellectual property perspective. And that enables the customers then to get the benefit of that speed, that reliability and the problem solving capabilities that really just aren't there. But then the foundation models, they provide their own reasoning and they provide really rich interactions with humans to understand intent deeply and to be able to speak in human vernacular and language and even cross language. And they're multimodal for speech as well as text and video. And so there's all of this rich capability that we get at that layer. And then we have our own intelligence layer that is the model training factory. We've got the first model, but we've got more coming over time and then we've got the foundation of the API service, the common data model and the predictive and optimization technology. So all of that together we think provides customers with the ability to be completely future-proof, because you can go end to end and you have this rich innovation history with Blue Yonder as well as the future where we're working with the biggest company in the world and the most advanced in driving all this AI. So very exciting time. And that's what we mean by owned intelligence. In some sense, the foundation models are getting such demand right now with all of the coding success that they've had. And what's happening is they're having to quantize their model some and that's changing behaviors a little bit, which makes it hard, of course, for us to keep our agents working properly. They've had to increase token prices from some of the different vendors. And so what we see happening is our strategy of owned intelligence is coming at the perfect time, because it lets us take some pressure off of that with models that we own and can host at the right time when the market is exploding and all this demand is driving costs up. So that allows us to have the best of both worlds using the big models where they're good, but also offloading some of the cost onto these smaller models.

Dave Vellante

>> Well, for so many years, months, whatever, the whole chasing AGI was the big thing. We actually coined the term enterprise AGI. Others call it enterprise singularity. This gets to the differentiation that enterprises can actually affect you guys as well. And of course, you're working with NVIDIA. NVIDIA brings so many advantages, not only the pace of innovation, they're in every cloud. They've got massive presence in every industry. They've got the Pareto set and deepest set of libraries. Obviously, CUDA is going to attract more and more developers, so there's this gravity there. But one of the things that you said that just kind of caught my attention was one of the catalysts was the economics or sometimes people talk about tokenomics. We've observed that the cost per token dropped probably 90% last year. Jeff Clarke last week at Dell Tech World said that the consumption of tokens increased 320x. So you got Jevons paradox at play there and that's not going to change anytime soon. So you got frontier models, super powerful, not necessarily the best fit for the things that are happening in supply chain. You've got to compliment that. So when do you apply? When and how do you apply to decide rather to apply the generic AI models versus the specific AI models? Do you take care of that for customers? Is it all sort of hidden from them and do they get involved? How does that all work?

Chris Burchett

>> Yeah, great question. All of this is built into our cognitive solutions. So customers don't have to worry about what model is the right model, and will it be applied in the right way? It's all done for them automatically. It's built into the way we engineer our agents and you might say the agent harness and the context management depends on using the big models for where they're good and using the smaller models where they're needed. And we do that based on the failure modes that we've observed, when are the big models struggling a little bit more because the reliability of calling some of these APIs to answer certain problems is not as easy. And we do use skills as well and we allow extensibility of the agents so that customers can extend those and can change the behaviors within reason. Of course, we have to be careful with that because we want them to stay very, very reliable, but this is all a part of ... The tooling to do that is a part of our agent marketplace and it's all kind of built into our platform, but customers really, they get the full advantage of this without really having to do any extensions on their own. It's available to them, but right out of the box they're able to deploy and use the agents and see the benefit of the models right from the start. And what we've done in selecting the models internally, we've selected the NVIDIA Nemotron models. And we did this because NVIDIA has made a huge investment in these open weights models that allows us to have a partner who we know is going to build enterprise ready, enterprise grade models. And when we tested them, we found them to be on par with all of the other models in the same weight class, so to speak. The 30B models are all the same. And so that's something that not exactly the same, but they all operate at the same level. And what we've been able to do is train those 30B models and that 30 billion is 1/60 roughly the size of maybe one of the bigger models. And so what you get there is you get so much faster, lower latency, you get lower cost. And so we also find that by training that model, we can get much higher performance. And so these things happen in the labs and we prove them out and then we just build them in by default as a part of the solution. So there's nothing for customers to worry about in terms of being able to leverage them.

Dave Vellante

>> Well, it's been fascinating watching NVIDIA and what they've done with Nemotron. Open weight models are frankly necessary for enterprises and organizations like yours and ISVs to actually be able to tune the models for themselves and have propriety. We know that China has been delivering open source and open weight models so there's competition there. The big frontier model vendors in the US, they're spending so much on CapEx, they have to sort of protect their base and have been reticent to aggressively put forth these open weight models. They've put them out there, but they're somewhat deprecated, and you've got NVIDIA on the frontier. So this is really fascinating. I want to talk about use cases for a little bit, because initially in the whole GenAI phase, you saw hundreds, if not thousands of use cases coming to the front of organizations. They had to go through a prioritization exercise and then really try to hone in on the ones that were going to make a big difference for them. Your initial proof points, things like warehouse workflows, short allocations, making exceptions for inventory. I know there's been a focus on changing schedules where there's greater urgencies and dealing with incoming trailer inventory. Why do you start here? I presume they have big impacts. Explain that. Maybe you can take us through what success looks like in those environments. How are you measuring that?

Chris Burchett

>> Yeah, this is a great question. And really, it comes down to inside the warehouse, the intraday operations is where it's all at. And typically, things can fall apart very quickly. Like I've said before, pallet falls over or trailer is late to get there with that inventory that you need. So how do you deal with that? And what we've built in our cognitive solution is a whole set of intraday operations. This is a new area for us that includes how we do pre-shift planning in a new way, how we do post-shift evaluation and all the intraday operations. And part of that is you have all these planned allocations of inventory, you're allocating them to orders. And when you do that, sometimes it can come about that there are shorts, there are short orders and your primary metric in the warehouse is on-time in-full at the end of the day. That's what you want to drive, on-time in-full. And so if you have shorts, this is a big problem and many warehouses have two or three people, maybe even if they're a large warehouse, that just chase down what happened to the inventory. "Did it get misplaced? Did it go stale? Is it out in a trailer somewhere in the yard?" And so what we chose to do is focus on a high hitting use case that everybody deals with. It's broadly applicable and it helps to solve one of the major areas of disruption for warehouse operations every day. And by doing that, we have something that's highly reusable and applicable to pretty much any warehouse operation. And the agent then allows us to do that and it can look at congestions. And increasingly, we can add due time and other metrics within the warehouse as a part of this operation in the future. And so what that allows us to do is start from a very solid first use case and then grow over time to do more and more autonomous operations with human of the loop still, but eventually more and more autonomous as the humans develop trust.

Dave Vellante

>> Okay. Well, that sort of answers my question around trust. When you start to let agents loose and you've got this autonomous supply chain, you got to worry about data and AI trust, governance, compliance, got privacy issues, security. So how do you de-risk these systems and models on real supply chain tasks before you go to market, before you deploy it? What's the role of the humans in the loop?

Chris Burchett

>> Yeah, it is a great question and it's something we think about an awful lot. Right from the start as we started to look at agents, we knew that reliability, trust, responsible AI policies were going to be so important. And so we actually did put responsible AI governance policies in place right from the start as we started to develop these agents. It's now built into our software development lifecycle. And some of those key principles are around human-centered design, transparency and explainability. It's around data security and privacy. And then generally around governance and compliance to regulations like the EU AI Act and other regulations that are emerging across the globe. And what this allows us to do is have a very high bar for the enterprise so that we can serve our customers in whatever region they're in. And a big part of that is of course data sovereignty. We keep the data in the same region. It never leaves. We also, in our model, training factory. We've been very sensitive about how we treat data. We only train on synthetic data and not on customer data. And when we run these models and run these operations, they're all customer isolated. So we're not ever crossing the streams of one model instance being used by multiple customers. All of this gives us a strong foundation. But then like we said, trust is developed over time really. And it's as humans get comfortable with the AI, it's not unlike getting comfortable with another human. If you interact with somebody and the next time they remember your name and they follow up with you and they're bringing things that are of value to you and you can start to depend on them, well, you start to develop trust. And this is the way trust is developed with AI as well. If it remembers you, if it understands how you like to see things and formats it that way and if it is providing value, it's reliable when you ask it questions, when you ask it to do things and it doesn't run off and do too much, but comes back and asks for guidance and approval, then that's how we think AI will start to earn trust. And eventually, what will happen is people will say, "Okay, I trust you up to this level. If it's a decision under $10,000, go ahead and you do it. And it's the same issue that we saw the last 50 times, then go do it. But if it's different or if it's above that, then come back and ask me." And I think that's how AI will start to become more and more trusted.

Dave Vellante

>> Well, we've been following Blue Yonder for a number of years now. I remember talking to Duncan several years ago about sort of the vision. I think the NVIDIA partnership is very strategic. It gets you on that. Now, you got to move at the speed of Nvidia, but that's a good thing. You're on that innovation curve. Of course, at the same time, you're watching the frontier models, because they're progressing at a very rapid pace. They're raising tons of money, potentially they're going public. So that's going to be interesting to watch. As you look down the road and thinking about the vision that you kind of laid out upfront, what do you want people? How do you envision the outcome for customers? We're going to continue to see lower costs per token, lower inference costs, better models, et cetera. But how does that all translate to your vision? What do you expect the supply chain world to look like in a couple of years?

Chris Burchett

>> It's a great question and it's hard to predict. I think a couple of things that we're seeing. We're seeing agents more and more within the same domain are taking more and more action. They're capable of taking more action on the part of the user. And I think over the next, call it 12 to 18 months, we'll start to see users get comfortable with those agents taking action. And what will also happen during this time is agents will start to interact more with each other to solve end-to-end supply chain problems. So today, we have this working in our labs, but we don't really have a lot of it in customer deployments today. And so what we'll start to see is within a domain, like warehousing or transportation, you'll see agents more autonomous and working and building trust with humans. And you'll also see it across them where there was a network disruption and now the network ops agent informs the logistics ops agent of the need for new rating or lanes or whatever and it'll go to the inventory ops agent to replan things and come back to the warehouse to inform the warehouse of what's happening. And all of this will give you an end to end solve. Still may, at the end of all of that, come back to the human and ask for approval. All of that is appropriate and as it should be. But I think in the 9 to 24, 36-month timeframe, we'll start to see that. And as a part of all of this, it'll be because we have these new emerging domain specific models from the model training factory that help to ensure the reliability, the cost and the latency within spec that's really going to make this possible. So it's an exciting time. I think there will be also a whole new class of jobs for humans. I think as you think about more of these automations, you need also humans to sort of manage the process, manage the agents themselves. Today, we don't think of humans necessarily managing the agents, the data feeds, the systems around, because those systems are changing all the time and external systems. I think a lot of our customers want to connect our agents with their agents. And so I think we'll see that happening over the next two to three years as customers are trying to solve not only the supply chain problem, but how does it then interact with my HR or my finances or my procurement or whatever it is that is the broader enterprise problem for them. So I think it's going to be an exciting time of all of these dimensions playing out and I really think this is probably the most exciting time of my career with this technology that has just been explosive in the capabilities that it provides.

Dave Vellante

>> Well, Chris, this is next level stuff and this is one of the reasons George Gilbert and I have always been so fascinated with supply chain and the potential for what you're doing generally and then specifically Blue Yonder, what you guys have done, your vision and the execution ethos that you brought to the table. So thanks for spending some time with us. Really appreciate it.

Chris Burchett

>> You bet. Thank you, Dave, for having me.

Dave Vellante

>> Absolutely. And thank you for watching AI and the Autonomous Supply Chain with Blue Yonder. This is Dave Vellante on theCUBE and we'll see you next time.