Frontline Mobility Edge

AI at the Edge: What's Working for Frontline Workers (and What's Not) | Justin Griffith, StayLinked

• BlueFletch

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 1:06:04

AI is supposed to transform the frontline. But in warehouses, hospitals, and retail stores, the math doesn't always work.

Brett Cooper talks with Justin Griffith, CTO of StayLinked, about what AI at the edge actually looks like when it hits real operations. Justin's team has spent over a decade studying technology adoption in environments that run on terminal emulation built in 1969 and still power 70% of the world's warehouses.

They cover the three main edge AI use cases: vision, voice, and workflow automation, and why each one hits a different wall. Vision AI drains batteries too fast for full-shift use. Voice AI adds inference latency that warehouse workers can't afford. And workflow automation requires WMS integrations that take years and carry existential risk.

Justin frames the core problem: AI right now is a solution in search of a problem. For the middle 80% of the market (i.e. companies that need technology to work and need their investment to see a return) AI hasn't cleared the barrier of viability yet. The top 10% can afford to experiment. The bottom 10% can't afford to move. Everyone else is stuck calculating whether the juice is worth the squeeze.

They also dig into data ownership, the OEM hardware arms race, why memory prices have tripled, and what happens when a robotics company turns out to be a data company in disguise.

Welcome And Edge AI Setup

SPEAKER_00

I'm Brett Cooper and this is the Frontline Mobility Edge, where we discuss the latest in mobile device technologies and how they're shaping the frontline landscape business. Thank you for joining us. Let's get started. Hello, I'm Brett Cooper and joined today by Justin Griffith, CTO of Staylinked. Justin, thanks for joining me.

SPEAKER_02

Thank you for having me.

SPEAKER_00

Um Justin's one of my favorite Photos talk about technology with, and I I think in my my show notes I I might have accused you of taking apart your parents' VCR when you're younger. Is that something you actually did?

SPEAKER_02

Yeah. I actually there is a box that is effectively a graveyard of transformer parts. Where I was like ripping apart transformers because I just had to know how they worked, but I never got them back together.

SPEAKER_00

They were very hard to get back together. I think that too, yeah. The GI Joe's are easy to put back together, but the uh transformers definitely wait, we do any pieces. Yeah, I just can't imagine the assembly lines in China.

SPEAKER_01

Yeah.

Staylinked And Why Telnet Survives

SPEAKER_00

And on the long lines of technology and taking things apart, this is uh want to talk about AI at the edge. So Justin and I had a conversation at one of the conferences, I think it was NRF or PSCO or Modex or HIMS or somewhere where we were talking about the devices and when we're gonna see more AI instead of being in the cloud at the devices. So it really led us down to a thread of a couple subtopics. We'll hop into those, talk through a bit of that, start out with use cases, talk through some of the technology and hardware, and then talk through some of the um, I guess, fringe issues or concerning issues we have on the edges. So that'll be what we cover today. Um I guess before we hop into that, Justin, do you want to talk about what Staylink does for anyone who doesn't know, and then maybe what you do day to day with the organization?

SPEAKER_02

Yeah, yeah, sure. So Staylink is a software company uh that's been around for the last 30 plus years. Um, and for the last 23 of those, we've been specializing in terminal emulation, uh, which is one of the oldest computing protocols still in active use today. Um it's wild. The protocol that terminal emulators speak is like one of the originals. It's called Telnet, and Telnet stands for Teletype Network. It was created in 1969, the same year we put a man on the moon.

SPEAKER_00

69. I thought I thought Neo invented the Matrix.

SPEAKER_02

Yeah, right. Yeah, yeah, that too. Um but it's it's just wild. So we we uh specialize in this technology because it it still runs like 70% of all the warehouses in the world. Because it is old and ugly, but it's also fast and everywhere. And uh so over the last couple of years we have been building more application delivery, uh application interface delivery mechanisms. So we just released a browser um because we are seeing an increase in interest in in sort of our play browser. Um and then uh we also specialize in technology integration into these older technologies that have no context for robots or AI or all of that, because again, they're they're built for an interface that was you know created in 1969. So there's two divisions of the company. We have the core business, which handles all the all the uh application side and what we go to market with, and then the other side is uh Stay Link Labs. And that was founded about 11 years ago with me and my counterpart Porig Regan. And uh it was basically born out of this obsession that uh the first day when I started with the company, my dad sat me down and he's like, Look, do not get too comfortable here. This maybe has two or three years left, speaking about telnet and and terminal emulation. And that was 22 years ago. And he was a great technologist, and I always kind of wondered why why was he so wrong? Uh he you know, like why why was everybody so wrong? Literally every day of my career, people have been talking about how uh how this technology was going to die. But there's more users on it every year. And so that that that just really stuck with me, and it really got me obsessed with this whole idea of technology and technology adoption and why people adopt technology and and what keeps technology from being adopted. What are the headwinds, which is sort of a thing that I don't think it's talked about a lot. We hear a lot of narrative about why technology is so important and why it's so urgent and all of that stuff, but we don't hear a lot about what is keeping people from getting to where they want to be technologically. And so we started Stay Link Labs specifically to research that. So we do a ton of studies about this, we do, we do a lot of our own research. I get a lot of technology, it's all a thinly veiled cover for my gadget habit, right? This is all an excuse to just buy the cool fun stuff and play with it. But a lot of it is this endeavor to try it out, see what works, see what the headwinds are, and then see if there's any way that we can give an approach or a technology or or some sort of uh philosophy that will help the technology be adopted faster.

SPEAKER_00

So your day-to-day is doing a lot of this analysis working with our team on these projects in addition to supporting the core products.

Adoption Headwinds And Tech Graveyards

SPEAKER_02

Right, yeah. So I manage the company's sort of uh general strategy and and technology direction, but then uh between me and my counterpart, Porg, we do a lot of the industry research and uh a lot of a lot of uh stuff like this, just a lot of uh um pushing on technology, see where they fit, see where they don't, which is honestly more important sometimes, just knowing where things aren't gonna find their fit, and uh and then trying our best to give that to the market so it gives the technology the best chance at uh at making it.

What Edge AI Means At Work

SPEAKER_00

We need to go through that. We need to do a podcast specifically on the technology graveyard. I've got like 10 right now, like blockchain, NFTs, data. We we can go on and on. Um yeah, you could see a pattern, right? Yeah. That's the next thing, but it never never materializes. And and I think that's a it's almost a better question of not what makes success, but what makes stuff not successful is a great question. Um so I get I get hopping into the topic here, I'm gonna give some context of this in the way you I think Justin framed it up a bit of you know, Stalink is on millions of devices um across manufacturers, we retailers, warehouses, 3PLs. So anybody who's moving things and needs to track things or tying into these legacy systems or even some of the newer systems, uh, you guys support that. And I I typically frame those as the uh the frontline workers, people that actually do shit out there. Um and I think the the thing uh we think about is when I think about AI, it's the LLM is the thing in the cloud, you go make a request to it, it does something, versus AI running on your device. And over the last couple years, there's been a shift of people realizing that it's very, very expensive to run things in a data center. How do we move it down to your computer, you know, Mac Pro on your desk, and now down to some point? So I think really talking about that that evolution, what that looks like, what are the use cases and business cases? So topic A, um you know, I think the um on the on the note of LLMs, I know we all do our our cat videos or cat memes and um have it write our homework assignments for us and then uh um you know file our taxes for us when we need to do, but you know, from a standpoint of the primary use cases you're seeing out there for frontline workers, um, what are you know I I know a couple we talked about for so vision, voice, and just like general workflow, but maybe you can talk about some of these and what you've seen in them. So maybe we'll start with like vision. What are you seeing around vision at the edge or vision vision AI on devices?

Vision AI For Compliance And Identity

SPEAKER_02

Yeah, so uh a lot of a lot of the edge uh computing use cases, again, I think they're it's still finding its balance point, right? There's there's a lot of question around uh capability, right? Like, can it do it? Yeah, absolutely, it can do it. Is it practical to do it, right? Like is it is it practical to have an AI model that can do complex computer vision on a system that runs a battery, right? And that the the resource utilization that that uses. So um what we're finding is that in a lot of these use cases, it can't be something that it has to do that all the time. Right? It can't be an always-on AI that is doing inference on an image, even with a very tight model, even with very tight constraints. Um, it's just a very resource-heavy, like you can see the news, you know. It's a very, very resource-intensive technology. Um, but it's also been around for a long time. And, you know, the definition of AI is pretty broad. Like lately, it's become synonymous with LLMs, right? And so a lot of the use cases tend to be LLM type use cases or or content generation kind of use cases or content uh content inference kind of stuff. Like I'll upload my my work manual or something. So this is this is a good use case we've seen. I I'm gonna have a chat bot that runs on my device and I'll upload my procedure manual into that, my training manual. And if the user gets stuck, they can pop a chat window and ask the as opposed to calling a trainer or walkie talking a manager or something like that, if they get stuck, um, they can they can get some help when they need it, right? Using that. So that's that's an interesting one. Um kind of anytime you can consolidate data, we're now seeing that the AI is powerful enough on uh consumer devices, and then that, you know, you've you've seen in the rugged space, it takes a while for that then to get ruggedized down to where it's used in a warehouse. Um, so for vision, we've seen a lot of interesting uh use cases for things like compliance. Compliance is a very good one because it's like, okay, I'm gonna point it at a shelf and make sure everything is where it should be. And that is that is relatively uh simple to do. But I think one of the potentials that I see uh is around uh visual validation beyond beyond compliance, things like, okay, I want to validate who is who is going into a gun cage. You know, I'm I'm at a sports retailer, I've got a gun cage or something, I want to validate who is the user walking in, guarantee that it it matches the username signed into the system, right? Yeah, it's great use case, yeah. You know, and so then you have this sort of trusted chain of custody through the thing you validated that the person who's signed in is the person holding the device. Um I do love that about identity. It's one of the things I've loved about what you guys do, right? Is that uh the WMS knows the username, right? And the device uh up to now hasn't really had any indication of who that is, who that actual person is. So being able to tie those two things together, I think is really important. So those are some of the interesting things, I think.

Voice Workflows Demand Pure Speed

SPEAKER_00

That's actually a use case. We've had I have one of the customers we work with. We have, for those of you who don't know what context Blue Fletch does, single sign-on login. One of the things we added in is biometric login. So you can log in with your face. And we had a customer, and the guy's like, I love it. Every time we load a truck, I want you to revalidate. I'm like, what? He goes, Yeah, we have these guys, they'll come in the internet shift, log in, and then we'll have their cousin work for them all day loading the truck, and then something goes missing, and we don't know who who who stole it. And I'm like, Oh, that that is a spectacular use guys exactly aligned to what you just said there. What do you what are your thoughts on voice, like voice use cases and voice edges? I know like vocal ed, all these things have been like doing voice picks, and it's I I still had an experience, I can't remember who it was last week. I called somebody and it was like it was like say what you want. I like said the thing and they're like, we didn't understand you and I'm like how to say it five times, and I'm like, I just gave it.

SPEAKER_02

Well, and this is part of the thing that um I don't think a lot of people appreciate about voice. So we used to have our own voice software, and I killed it when I became product manager. I killed it because one of the things that um people don't appreciate about voice is that you can't dabble in voice. Voice is one of those technologies that's so core to the human experience that if you have to repeat yourself, you're annoyed. But if you have to repeat yourself twice, you're ready to throw whatever you're holding across the room, right? And this was part of the reason call centers and stuff uh kind of got away from that experience. Um voice in the warehouse comes down to brevity, right? The number of syllables spoken to you can have a huge impact on the number of seconds it takes to get through a workflow. And when we did a study in Stalink Labs on the priorities that warehouse workers had and that that warehouse managers and and businesses had, speed was far and away. The most important thing. If you weren't talking and aren't talking about speed in the warehouse, you are missing something significant there. Um the next most accurate thing, or the next the next most uh important thing is accuracy for them, right? So, but if they had to pick between accuracy and speed, they'll take speed.

SPEAKER_00

So one of the things with voice, I I feel like and the reason I think it's going to be more relevant is the thing I've experienced around context with AI. So AI is starting to do a much better job of having a large window of context. So if I have a discussion with Claude, and Claude now has like permanent memory or GP Chat GP has permanent memory, um, and you know, you start having this experience where I'll ask it something and it will be able to correlate that to a question I had asked like five days ago. And I think for me, that especially around voice makes a big difference. Like when you when I say something, there is contact context to what I had said, you know, five phrases ago or 20 phrases ago that a lot of the systems that we've used in the last 10 years just don't have that. It's like here's the here's the change true tree, go from this. It doesn't have any context to either what you're doing, where you are, who you are, what you've done in the past, you know, anything else.

Connectivity Drag And Hidden Costs

SPEAKER_02

Well, yeah, and the use cases for voice, again, don't usually use that, right? They they it's a very specific workflow that they're doing. And so the ability to have a memory about this person and this person's quirks or this person's uh where they are in their in their career and all of those things, none of that uh kind of matters to say like a vocal act workflow, right? Where it's a very repeated task that they need to perform at speed, and the point is the speed. Voice is about not having to type, it's not having to look for anything, it's being told what what to do, which is 80% of the benefit of voice. It's the guidance of the job. And then validation, literally, if you can get that down, you'd have a long string, and instead of saying the long string, you'd do a check digit, which just validates you okay, we're talking about the same thing. But it's all about compacting that down to faster and faster things. So AI in voice, it really depends on the use case. But for productivity, again, waiting on the inference to happen, waiting on it to interpret what you're saying, um, even that drag is probably too much for the workflows that rely on, you know, if they could take half a second off a workflow, that times this transaction 10,000 times a day is a huge impact per year for them. You know, like economy of scale here. We ran we ran a study about a year and a half ago on connectivity because I was I was dead curious. This is one of the things we specialize in. Terminal emulation was not built for wireless. And so we built uh a fix for that, an architecture that works better for that. But I was curious about how big a problem is this still. You know, Wi-Fi's come a long way, uh, there's new connectivity modes, you know, you can do 5G now and and all of that. And so is connectivity still an issue? Yeah, it's still like the biggest issue. And what we found was that it happens all the time, every day to every worker, and it's not a huge interruption, but multiplied out times so we what we found was it happens often enough that if you have a worker with fifty you you have a warehouse with fifty workers in it, that those fifty workers having the number of interruptions they have today, it could cost the business four hundred thousand dollars a year, or a WMS upgrade every year, or a device refresh every year. Yeah, or a complete network overhaul.

SPEAKER_00

Stacks up fast, yeah.

SPEAKER_02

Absolutely. And that's if you have 50. If you have a hundred, double that. It like the numbers get to be unbelievable. So when you think about the drag of a technology on something, you've got to make up with that in in equal and and opposite benefit or or better benefit than what you're losing. So if you lose half a second waiting on an AI to interpret, then what what are you making up for with that? Are you gonna get the same benefit in in the boxes out the door kind of benefit, right?

Can AI Replace Task Management

SPEAKER_00

No, it's it definitely adds up fast. Little things add up fast. One of the things in there, uh just want to dive into more. Uh so talking to somebody this week around the looking at, you know, we've all seen the workforce management software, work workdays, Mondays, Salesforces. What is your thoughts on that being replaced with AI to go figure out what people are doing? And I know this is sort of a a thing everybody says was gonna happen, um, whether it's um, you know, the sales forces, the SAPs, Microsoft, they're all gonna give you this AI that can go replace your um your thing that's tasking your workers, telling workers what to do. How closer to that is that something's even gonna happen. It just seems very, very complex.

SPEAKER_02

I think there is a massive underappreciation for what a unique and special Snowflake every customer is. And their particular workflows and their particular uh their particular business, right? Like even if you the what we found being in this industry is that the most obscure components, the most obscure raw piece um of a thing, like we have a company that specializes in making bolts for bridges. And what you'll find is that there's a huge competition in that market. There's companies that are are competing going, those guys make crummy bolts. We make the best bolts for bridges, right? Like um so you have all of this uniqueness in how people deploy their environments. And we saw this with uh people who would kind of make the assumption of like, okay, I'm going to take uh a WMS and I'm gonna build for the standard version of a WMS, a standard version of SAP or an out-of-box version of Manhattan or something. And I'm gonna build a solution that relies on the out-of-box mode that that comes with. And then that should get me 80% of the way there, and then I'll fill in the gaps of the of the other 20% around customization with services, right? And I've seen that again and again and again.

SPEAKER_00

The entire industry was built around SAP consulting for the last 30 years. Yeah.

SPEAKER_02

Yeah. Yeah, but that's that's part of the reason is that there is no out-of-box anything. The first thing you do when you deploy one of these environments is you customize it around what it is that you make and what it is that you do. So even in very specific industries, if you if you know grow and distribute strawberries or something, the process by which you get that to the shelves could look totally different from your competitor. Um so kind of having an AI that can automate your business workflows for you, right? And give that work to users in these spot solutions and stuff, and for very um kind of repeated or structured tasks. Like I see that in retail where there's a very set kind of maintenance cadence out there. Um that that's not that's not a huge lift. In fact, a lot of times that doesn't even take AI, right, to to do. And I do think we get caught in this trap where it's like, okay, AI right now is kind of a solution in search of a problem. And so when we see a problem, we think AI, because it's sort of top of mind. But there's a lot of things you could do uh outside of AI that get the same effect, get the same solution, right?

SPEAKER_00

Or get do you think there'll do you think there will be any winners in this? You know, just and not to um harp on SAP, but I feel like that my my typical experience with SAP implementation is you pay a million dollars for the software, um, expect to have it 90% working, and then you pay$100,000 for consulting, and then you end up spending another four million for consulting. Like, do you think that that margin is going to go faster and smaller down with some of these AI tools to go figure out those workflows or those edge cases?

SPEAKER_02

Yeah, I mean it it will it will help in kind of interpreting um there there was a mild panic in in the marketplace going like, okay, AI can. Read code now. And so it can modernize all these legacy applications. And one of the struggles that people have had when trying to modernize their applications, because typically what people do is they will install one of these apps, which takes a very long time. Most people don't appreciate that. If you want to jump from, say, you know, one ERP to another, right? You want to go from Oracle, you know, over to something else. That's a three to five year project, if managed well, and if it succeeds. There are a number of companies that try and then fail and have to pull back to what system that they were using. So a company that is running, you know, a 15-year-old system, sort of the assumption is that it never got it never got paid attention to, it got ignored. But a lot of times that company might have tried two, three, four times to roll onto something different and discovered that uh one of my favorite phrases in the world, the juice was not worth the squeeze on it, right? And it it could have tanked the company to keep trying, because you could keep throwing money at it and never succeed. So I think we might see a higher success rate a bit, but but again, the most important thing to a company behind speed is in risk mitigation. Right? They have to be able to keep hitting the numbers that they're hitting today and do better the next year, and whatever solution they put in place has to meet this barrier of viability where whatever I put in has to do as good a job as what I have today or better. And if I can do that, okay, then it's viable, I can budget for it, right? But if that's not proven, then I have this whole risk mitigation conversation that I have to have. And I think that's where it's stuck right now. That's that's a good frame.

Legacy Modernization And Risk Mitigation

SPEAKER_00

That's a good framework to think about it. Yeah. If if you put it in, it better be faster, or at least the same, yeah. Um other use cases at the edge. So I I know I, you know, coming out of Modex from last week that we were at, there was a lot of um, a lot of different use cases for AI, and it wasn't people actually, I was very impressed of the how people have toned it down where they're not saying AI anymore, they're just being they're just saying, you know, automated fork truck or you know, inventory function. Use cases. Yeah, use cases, which I I loved that everybody started doing that. And um, I guess your thoughts, the things I I I saw that was interesting. So the automated like yard trucks and forklifts, we that seems like a great use of you know putting uh you know, putting an NVIDIA uh chipset at on a device in a microcomputer or NUC and then having it actually do vision and and do a lot of these things that you'd normally have a driver do that are you know mindless or kind of boring. Like what are your thoughts around where you see more of that? Is that actually a real thing?

Self-Driving Forklifts Need Integration

SPEAKER_02

Yeah, we've done it. We actually worked with a company that was doing uh distribution for uh uh alcohol, and uh they had a uh they're called VNA trucks, very narrow aisle trucks, and they had a self-driving system. And the driving is the AI bit, but telling it where to go does not take AI, right? We knew because we could see it on the screen. So we needed to be able to interpret that that's where they were supposed to go. So instead of the user typing in something or having to do this big expensive open heart surgery on the back-end system, right? Again, the risk mitigation thing. Usually when you're talking any one of these technologies, uh if they're a if they're an individual solution, right? So say you want to get self-driving forklifts in there, how that would typically go is uh you get a pitch for how great these self-driving forklifts are, what it will do for you in accuracy and what it will do in productivity. And it sounds like it'll be a thousand percent better than what you're doing today. And all you have to do is hook it up. All you have to do is just just hook those two systems up. And that gets said like it's the easiest part. But to a customer, what they hear is you're about to touch the most expensive, most risky piece of software in the entire enterprise. And if you screw it up, you could tank a Toyota for doing that, right? You could you could kill the company for doing that wrong. And so if you don't go in with the deference of that, of that with the recognition that that's what you're saying to a customer, you're not likely to get much airtime for that, right? You need to go in and appreciate what you're asking them to risk if you're going to touch that. So that was one of the reasons we came up with labs and we came up with our our technology integration uh evolve, was that we wanted to be able to integrate these technologies without having to basically make the WMS aware that anything was going on. Right? If we could get self-driving forklifts to activate because we could see it in the stream, we could see it in the terminal emulation screen where they were supposed to go next, extract that and feed that to the self-driving system, the WMS had to have no idea that they didn't care. Um so yes, you will see more of those companies out there, but until the integration part, which is the absolute hardest, riskiest part to do, gets simpler, um, you're not likely to see more deployments. And we actually ran a back-to-back study on this uh two years we did in 2022, and then uh just a little ways into 2024, we ran a technology adoption study. And we ran through, you know, what are your what are your priorities, you know, and what are the what are your technology ambitions, drones and and AMRs and and self-driving forklifts and all of this stuff. And they all came back and said, Yeah, we're desperately interested in those things. And what we found was that the age of the WMS had a huge impact on whether people had any success in deploying these things, right? And whether they had any success in even testing them, because in many cases the cost to even try these technologies was the same as the cost to buy them. To even try self-driving forklifts, you have to do the integration into the system. And uh so even that was very, very expensive, and it took a lot of buying to even get that going. Um, so next year, right, there was all this rush toward the end of COVID about robotics and automation because you couldn't get workers into the warehouse safely and you needed to guarantee that your warehouse would stay open. And we'll fix that with automation. And then they try and deploy it, and their system is too old, or there's literally no plug to in the in the WMS to plug that in, right? So then fast forward about a year and three months later, and we run the same study, and a huge number of people had now jumped their software version up. They the belief being, okay, we're gonna have more success because it's easier now on the new version of WMS. They say so there's new API sets, there's now a plug to actually plug those things in. Um but what that study uncovered was that we had a ton more people able to try the solutions, but there was no more success in actually buying, no more success in actually deploying these new technologies. And what I attribute that to is that there is a huge difference between something being easier and something being easy. Right? If you're going to deploy a very intricate automation system, again, that could be a two-year project. If you're on a legacy system that takes integration and all of that. And if you're on the new version, it might shave 25% of that off. Now you go from two years to a year and a half. That's objectively easier, but it's not easy, right? That's not overnight. And I think that's what a lot of people underestimate about the environments that we go into. They're they are enormously complex and enormously important to the people running them that they stay running.

SPEAKER_00

Do you think that interface layer is going to change? So I when you said that, it made me think about the thing I've seen the big LLMs do well in the last six to twelve months is you know, either I'm gonna call it MTP, like model context protocol, where you have all these different systems. It can, you know, Claude can plug into Notion, it can plug into my Google Drive, it can plug into my email, and it's not having to do API calls. It's doing it's actually has a model that's been exposed by that provider. So do you see like a thing where the WMSs and other enterprises ERPs are actually gonna have like a MCP type interface sometime in the near future?

Easier Versus Easy Adoption Reality

SPEAKER_02

There's there's sort of two competing uh philosophies about that, right? One is the belief that the inertia of things like MCP are gonna see rapid adoption, right? And that the MCP will allow you to then integrate or speak to an AI system that could do coordination. Um But the the other side of the coin is to say that, okay, well, we need to create something that doesn't rely on MCP, we need something that can do computer use, right? So I can have something literally monitoring a computer that can interface with the same apps that we all use today. And again, those are all those are all great options. The issue is in right now the accuracy of those things. If if you need an automation system to hook into your WMS, that needs to work 100% of the time. Right? And right now, if you look at all the benchmarks for a lot of AI, they're like, Yeah, I you know, it's a score of 83% or 70%, or it gets a it's a great first draft. It's a brilliant sometimes, it can one-shot things, sometimes, right? But you need something that works all the time. It needs to be math, it needs to not be art, right? And uh so yes, there will be more ties. I think that's more useful in the extraction and interpretation of things than it is in the functional execution, particularly of productivity and and you know, kind of these integrations. I think it also again, we kind of come back to the age of the systems in that a lot of these new MCPs will be features provided by warehouse management systems that are on their latest version. And so it's to drive you to upgrade to the newest system. But again, to upgrade, that's a three to five year project, right? See in 2030, yep. Right. You know, and again, that the three to five years is going to pass anyway, so you might as well try, right? And some systems, like once you're on them, uh, there's a couple of systems that are cloud hosted and versionless. And so there's some some systems where the promise is you get on that system and then you wake up one weekend and all of a sudden you've got a TMS system. You woke up and all of a sudden all your transportation management is capable of being onloaded onto the system. To get there can be open heart surgery. Right. And so are you willing to take the time and are you willing to risk it? Most people are not most businesses are not logistics companies, right? If you're a logistics company, every second counts and speed counts for everything, and so every dollar you spend on that is worth it, right? Because that's your business. But if you make lumber, right, that's not necessarily your business. That's not necessarily your concern. If you make bolts for bridges, you don't care about any of that. Logistics is a function of what you do, right? It's it's just a portion of what you have to spend money on. So are you more uh are you do you benefit more by buying a new, you know, saw in your sawmill versus making an investment in the WMS? And for many customers, then it becomes sort of this push me pull you of okay, they they want a great logistics system, but does the one serve today? Yeah, absolutely.

SPEAKER_00

Yeah, I think there's other industries too that are like that. So we were talking recently at a healthcare conference of you know, should I go invest in a new uh you know, this new healthcare AI platform, or should I just go buy a new MRI machine? It's a lot easier for the CFO of a hospital to look at we we got the new MRI machine. That's you know, we can do this many scans, and that's this many dollars for us. And so yeah, I it is not um not just manufacturing. There are a lot of industries that have that same like where do we put our capital? What hardware we invest in versus what software invest in.

MCP Interfaces And Accuracy Limits

SPEAKER_02

Right. Yeah, like I remember this talk during COVID, and uh, you know, I was I was saying that there was so much dialogue around uh automation, and a lot of that was around picking, which was around uh retail. And I was trying to kind of communicate that that is an industry, it is not the industry at large, right? Um, and so I think you're right. One of those things about AI is that it's very difficult to define the benefit. There's sort of all this promise that it could be, you know, kind of the business's equivalent of the cure to cancer, but you don't know for sure. And you could spend a massive amount of money trying to get there. And we see companies doing that now. A lot of the layoffs that we're seeing reported around AI are not because AI is replacing the workflow that those workers did. It's that the companies that are making giant investments in AI, because they're a, you know, a frontier model maker or they are a hardware supplier, you know, that that that is working with AI and frontier models or something, they have to make a a choice here between do I keep people on to keep building sort of my core business, or I'm out of cash. I I'm gonna pay to keep those workers on, or I'm going to pay for more infrastructure to keep pushing on AI. And uh I think that's kind of the pressure that people are under. For those companies, being a front runner is so important that the benefit is always there. Even being a front runner right now is very, very advantageous to them. But for most companies, they still have to keep making what they're making because that's that's what they make, that's how they stay in business, that's how they keep the lights on. So I like to think of the model where in in any technology conversation, you have a market as a whole at the lower end of the market, say the say the first 10%, they have no money to do anything, right? They have to keep using whatever they're using, and if it's not broke, don't break it. Like we're just gonna keep rolling. And then you have the upper 10% here on the other end that can afford to do whatever they want. In fact, it's part of their brand that they will go and try everything. They have a warehouse specifically built to burn down trying new technologies because they want to stay on the cutting edge of this stuff, and that's great for them. But then you have this whole middle market, the whole middle 80% between those two extremes, where they need the technology and they need it to work, right? It it they they need the money uh to be well spent and to guarantee that their investment is going to see a return. So exactly what you're saying. When you're a medical company, you have to decide between a new MRI machine, which has a defined benefit, um, or this AI thing, which has this kind of nebulous uh description that you you'd have trouble justifying to the CFO. Um, that that's that's a very real thing. And so that's kind of what we see in technology generally, especially in our industry. Um, we see these incremental improvements. People are willing to take baby steps in a direction towards a thing, which is partly why you see this struggle with adoption or all of these technologies that you just listed, you know, blockchain and all of these things, you know, voice and drones and all of that. It's because the barrier to entry is kind of an all-in. You gotta go all in on it to see the benefit, right?

SPEAKER_00

Yeah, I mean, I I think the a lot of those are we're like you use the phrase technology looking for a problem. I think there's a lot of problems out there that it that I've seen AI do a great job of solving. So I think there's probably a middle path, and we'll hop into those towards the end. But you know, we we're we're seeing a lot of those, but it's like you said, it is it's not an easy button. There's no easy button of you press a button and it works tomorrow. It's the same thing, same thing we've had for the last 20, 30 years. You've got to do a third you know, three-year plan, build models, move things, and then be able to support it. So a lot of pieces there. Um the hardware, you you're talking about hardware. I know what that was one of the topics we had, uh, and actually thinking about hardware at the edge. And um, more specifically, uh, you know, when I was doing research for this discussion, I was reading about Moore's Law, and Moore was a um engineer at Intel, a senior engineer at Intel came up with this theory that every two years the number of transistors on their chips doubled. Yeah, people keep saying, you know, that's you're gonna have limitations, it's gonna stop. I feel like since the 1960s, we've continued to effectively double the speed of our compute and hardware every two years, um, be it through more transistors or through smaller transistors or through better power management or more memory or other things we keep doing. But um on the mobile devices, I feel like there are still a lot of limitations related to AI. And when you think of, I guess the first thing around hardware, um, you know, the on-device AI versus cloud and the trade-offs for that, what do you see though, as for these mobile frontline workers? It goes back to your point earlier of you can't wait for the latency of you know a five-second network call.

SPEAKER_02

Yeah, well, there's there's a lot of uh it's it's an interesting thing. On device versus versus cloud. A lot of people that don't realize that even the cloud solutions are using a lot of on-device AI. They do a lot of processing locally, what they can. So if you open, you know, Claude, for example, in a chat window and you you type in a question, it's actually doing some inference locally, and you will see different performance on a brand new MacBook versus, you know, your your kind of run-of-the-mill Windows machine. So there is kind of a middle road there too, right? Where you could leverage the on-device stuff for the simple things, which is actually how Siri has been working for years. A lot of the voice agents that have been running locally on the phone for years do that. The stuff that can be processed uh locally is, you know, like uh text.

SPEAKER_00

I I don't know if I would say she's been working for years. I would uh Yeah, right. She's she's been trying.

SPEAKER_02

Yeah, yeah, giving it her best, right?

SPEAKER_00

Yeah, to go with my thousand requests to her, I'm guessing um 75% returned with, hmm, I can't help you right now.

On-Device Versus Cloud Tradeoffs

SPEAKER_02

Yeah, right. Yeah. Try again later. Um yeah, but that's uh that's a perfect example, right? Imagine that in an environment that's depending on it, right? That that's sort of the uh one of the realities where if you walked in to an environment and you had a machine vision system that said, all right, it's gonna knock out rotten potatoes out of your automation line. We're gonna spot the rotten potatoes and just swat them out of the out of the line, they'll go into a bin or something. And we'll catch it 75% of the time. But that's not good enough. If you make syringes, right, and you you the difference between success and failure is a massive lawsuit or something, you need that right 100% of the time.

SPEAKER_00

97% is not acceptable there. Yeah.

SPEAKER_02

The stakes that different companies have could be wildly different. Their risk tolerance could be wildly different. You know, if you make, if you make bounce houses or something, your risk tolerance is different than if you make, you know, heart valves. And so uh kind of the the technology conversation, even when we're having this one about uh kind of on device versus cloud, that has huge implications to a business where we're saying, okay, if it's cloud, am I relying on their cloud or my cloud? Am I relying on the infrastructure and the service level agreements of their software? And it's not good enough to say you'll pay a fee if your system goes down or I'll get my subscription money back for that month or something, because again, it could tank my company if it goes past a certain point. So what is your backup plan? And then what are the five backup plans for that backup plan? Right. Um, and most vendors are just not there yet. It's not an established technology.

SPEAKER_00

So you're if I'm inferring what you're saying, the holy grail will be on device where you don't have to have that network trip back to somebody else's infrastructure hardware where you're relying on them.

SPEAKER_02

The the holy grail is when it addresses the risk mitigation in full. So when on device can be 100% accurate to the workflow, right? Like if you have an on-device AI agent that you've uploaded your procedure manual with, that it will get it right 100% of the time when it says, okay, you know, the procedure in this moment is this, and it's not going to hallucinate, right? It's not, it's it's going to be accurate 100% of the time in a way that doesn't drag on the battery, right? Or the battery is so expansive and the processors have gotten so efficient, the memory is so expansive that you're not going to drag on the the worker's ability to get through a shift, right? That sometimes these things die for very practical reasons. Like, great, you have this hyper accurate, brilliant performing AI, but the battery is only going to last for an hour. Well, that's not viable, right? Because I don't have eight times the devices to get, or eight times the batteries to get them through that. And I'm losing a lot of productivity in the shift. So am I getting it back in whatever I've built in this AI? Um, so I think I think the industry is still uncovering what the holy grail is for that, right? The the holy grail is kind of customization. I think I say this all the time. The the holy grail of technology is when it can customize entirely around what you do specifically as rapidly as possible. And that's what we're trending toward.

Battery Life And Compute Constraints

SPEAKER_00

On that battery note, I know one of the things that I had noted by research is just the battery life. Um, and I've run local models just testing things out here on devices. The you know, you run a 2 billion parameter local phi model on your device, it's gonna go through your battery in an hour. Um to your point, you have to if you're running this and people are actually doing high-intensive shifts, is and I know some of the you know Zebra and some of these other um Samsung are doing a really good job of they're they're putting hardware specific coprocessors onto these devices, but I feel like they still chug a lot of battery. Is battery tech going to get better? The model's gonna get more efficient. Where do you think that's gonna go over the next three or four years?

SPEAKER_02

Yeah, yeah, there's you know, Morse Law technically, uh, but that's about you know, Morse Law applies to an industry where people are constantly pushing on uh the capability and where we have the so unlimited raw material, for example, uh all all the investment in the world, you probably see batteries you know expanding at the same rate uh in mobile devices, but you don't have batteries that last three days, right? In in mobile devices. Um and so some of that is kind of like does it does it hit a threshold and then it's kind of diminishing returns after that? So we had that with like ruggedization, where uh you would have a uh conversation with a device manufacturer who'd put all of their money into ruggedization, they go, Yeah, this thing is bomb proof. You can drop it from 70 feet, right? Yeah, you drop it off a skyscraper and pick it up and play hockey with it, and then you still scan with it. And at some point, you know, yeah, but it's a thousand dollars more per device to get that. And so the customer is doing the calculation going like, well, we don't get bombed a lot, and so maybe we don't need to spend the extra thousand dollars per device on this and go with something that can drop from six feet, right? Um batteries have kind of had the same had the same conversation, and um so I think especially with the crunch on resources right now, like if you try and build a PC with raw components, it will cost you three times what it cost you a year ago. Um so I think I think we're going to see kind of a drag in that in that these components are difficult to get your hands on right now. And um there's so much focus at the enterprise and putting this in other places that it will be a while before we see that in the rugged space.

SPEAKER_00

Just for context, this is uh Q2 2026 when we're recording this, and that's I'm gonna call it the uh memory has double tripled in price from what uh what it was a year ago because all these companies and data centers, I think tier one, are investing in building models, they're buying up all the memory, and they're building out huge racks of high-power systems to go you know train all their AI models or or run their compute. Um so it is it has driven up the prices and reduced availability of RAM. So if you have uh teenage kids that play video games, they're very angry right now.

SPEAKER_02

Yeah. Yeah, I do. I have that. Yeah. Justin's angry. He's like, Yeah, can't buy it. I'm fine. No, it's fine. I'm fine. But yeah, no, it's it's one of these things where you have if you have a frontier model, right? Like OpenAI walks into a memory company and says, we'll take all of it for the next three years, right? That has a cascading impact on the memory market. So you have people who are making other memory uh offerings, say an SD card or something, that they're bowing out because there's more money in building this type of memory for an AI company, right? And uh guaranteed revenue.

SPEAKER_00

Right. Yeah, and scale. Um on the the companies, I know one of the things I wanted to get your thoughts on are is the um, I'm gonna call it the A the OEM arms race, AA arms race. So I know Zebra released their AI models last year. They started releasing devices that have the AI co-processors, like the TC53Es, and a lot of the newer devices include those, and it's it they're substantially faster if we're running models in the edge. I've I've played around with them, they're really good. And then Stam Slung has done it, I know um Honeywell, Data Logic, some of the others are looking at this as well. Like, and then they're also buying different pieces or partner different pieces. Um, what are your thoughts around what we're gonna see for the rugged handheld devices over the next, you have been a call, like two to three years, in regards to either those companies providing the models andor pre-built things into the systems, and then also just the co-processors, memory, and that type of things getting more powerful on those devices?

Memory Shortages And OEM AI Arms Race

SPEAKER_02

I think again, there's kind of a line where there's diminishing returns to keep investing beyond a certain point, right? So if you create, say you're an OEM and you create a really powerful device that can run a really great onboard AI model, the the lead time between seeing that actually executed and masked on those devices deployed. So say you make one of the most popular devices deployed, and you uh you put in all this AI co-processing and better memory and all that stuff. There is a whole software ecosystem that we're a part of, you and I, that then we have to adopt something in that in that hardware, right? We have to put our our time and money and belief and effort um into that AI model to make some workflow or make some some enhancement that uses that. And then people have to buy those devices and then buy the solutions and implement the solutions for them. So there's there's just a ton of lag time in that, especially before that becomes sort of the predominant technology adopted in mass to the thing. So a lot of what OEMs do is build for potential, right? And they build for future proofing against these possibilities in the marketplace. And you can't be in the hardware conversation and not be talking about what you're building for AI in preparation for AI and not talking about what models you're including and not illustrating some workflows that can be done with these things. But a lot of that is around potential. And right now the conversation and probably will be for some time is AI, but exactly what you were saying, where um they're not saying AI for AI's sake, because that was like saying uh it's like talking about a raw material. It's like saying aluminum, right? Like we have aluminum in our stuff. Great. Well, what did aluminium for aluminium for the yes for the Brits. Yeah. Um, yeah. So, you know, if you're talking about this raw material, great, you know, it's like we've got aluminum. Great. Well, what did you build with the aluminum? Because that's what people are interested in. You know, there's a very specific market that's interested in the raw material. Like we, as software providers, are very interested in kind of the raw material of this. Okay, you've got an onboard AI model. Okay, well, that gives us openings to say what could we build with that that turns into a workflow that somebody might be interested in implementing. Where I'm at on my side is I'm actually on the productivity side. I'm on the tracking and and and uh the actual scanning and execution of workflow, right? And so in a lot of cases, there's again, there's no plug to put that in to the application. There's no plug to work that into the workflow. Um so like I remember when uh one of the WMSs announced that they'd finally had a place to put pictures. Well, up to then there'd been cameras in mobile devices forever at that point. But even if you wanted to take a picture before that WMS had a place to put them, it wouldn't have done you any good. And it wouldn't have done the the independent software vendor community any good to build anything because there was no place to put it anyway. So um I think that's just one of the things that that people sort of forget or or aren't aware of is the number of people that have to buy in to get a technology adopted. And so there's seasons where a lot of the conversation will be around scan range and the wars over scan range, or drop spec and the wars over drop spec, or battery or co-processing and AI models and uh modalities, right? Like we've got you know a smaller wrist wearable one, or we've got a heads-up display, or all of these things. Um these are cyclical, right? So right now the conversation is around AI. That will always sort of be a background thing, uh, I think, because going forward, this is going to be the conversation, but I think it'll be a lot more around solutions.

Ecosystem Lag From Hardware To Value

SPEAKER_00

You know, and so OEMs will, you think next three years at least will be they'll if you build the the they'll come to the field of dreams type model. So they're gonna continue putting things in there, seeing what sticks, and then it's up to guys like you and me to go have our teams figure out what to go actually utilize, and then then that'll really precipitate what's useful in those changes or not. Right. Um last topic, I know we've got a couple minutes left. Wanted to talk through this. Was uh a sticky one, but data and um security and compliance. So I I think the the thing, and you know, going back to the AI is great, but it needs this thing to feed it, and that thing is data. So the training data, the models, um just you know, context of what's happening. Um I heard somebody talk about the you know, Pokemon Go. I think this is on my first million, they talked about they uh sold the game, but they kept all the data, and now they're reselling the data for more than they sold the game for originally, just because all these companies want all the pictures people took in Pokemon Go because it allows them to build these really great maps and uh virtual worlds. But I think the the thing, all these companies have all their scan data, you know, they have a cloud provider that it's going to, and it makes me think of the situation with um with AWS when it came out. I remember a bunch of retailers refused to use AWS, they're like, Amazon's gonna steal our data, and Amazon's like, no, we're not. Um we already have it. They, you know, they um there was a huge scare about this. Like when you think about who owns data and and where it goes, like what are the things that that come to mind for you?

Data Ownership Security And Compliance

SPEAKER_02

Yeah, I think data ownership is one of these hidden brick walls for a lot of customers. Again, this is in the risk mitigation part of it, right? Um especially in the age of AI where you know that you have a frontier model warning customers that they're about to release the most effective network hacking tool on the planet. Okay, you know, even even the least paranoid data person at a company is looking at that and going, you know, the the best way to to get through that is to just not make it accessible, right? Is to just brick wall it and uh not let my data leave my network, right? And so we've seen that with automation, where uh somebody was offering an automation play, and the robotics company was not actually a robotics company. They were a data company in robotics company clothing. And so the offering was okay, well, you uh you get your orders in, and then we will extract all that data into our cloud system, into our black box, and we will build the orders in a way that the robots can interpret as quickly as possible, and we will give you a thousand percent better performance. So beyond the conversation of the open heart surgery and the risk of doing that, now you're also tacking on a data ownership thing because they are saying, okay, once it leaves your system and it's in ours, we have rights to do whatever we want with that data. And if you're a retail customer, sort of the promise is, hey, we can build metrics that give you a baseline so we could tell you whether you're on track or off track with your competitors. Well, if you're the leader, why would you want to give people that information?

SPEAKER_00

I'll just I'll just context this. So an example, not to this is not a real example. I'm gonna pick on Amazon here, but you know, if somebody had was using an AI provided by Amazon for production on their ERP line and had the Amazon AI tied into that, Amazon would now know what their production line looks like for the next 14 months. And they're like, well, they're gonna produce this SKU, this SKU, this SKU. Um, that company who also sells those SKUs can either go co-manufacture them with somebody else and sell them cheaper, or they can do price negotiations because they have visibility into you know, they know how much volume somebody's gonna have and how many they're gonna have to move. So I think that's when when you talk about data ownership at the you know basic level, that's the sort of conflict I always think of when you have these ecosystems of people that you know may not have um you know aligned incentives.

SPEAKER_02

Right. Everybody kind of has their own interest in that data. The customer obviously wants the data for themselves and to not share it. They want enhancement to the workflows in a way that doesn't expose that that data and their strategy and and their roadmaps. Um right, you have uh everybody is also fighting for relevance. So I remember a conversation that I I was part of where uh there was a robotics company talking to a WMS company, and they were speaking to the WMS company as if they were the dinosaur, and the the robot company would do them the favor of involving them in the process because they were going to be outmoded in six months. And like three months later, the WMS system said, we're putting a hard lock on all integrations into the system. You cannot integrate with the WMS unless you use our data pipes, unless you use our API infrastructure, or you're out of support. Well, okay, that's a WMS exerting its relevance into the conversation. And again, a lot of that was about you know maintaining the data that they had, that they owned, that they were they were part of optimizing and not letting that out into the world. So there was a protection thing as well, but they also had the stance that they could do it better than a random robotics company that had showed up and and were saying that they could be the be-all end-all of that. Um so there are there are just a ton of these influences on technologies, and a good number of them are commercial, right? Where somebody has a commercial interest in that data. And um, just like the Pokemon Go thing, that might not always be totally visible in the fine print. It says we reserve the right to use this for other things not defined necessarily today, and that might mean taking all your Pokemon Go data and selling it to a maps company. Well, if you're a retailer and you're creating a new line of athletic wear that has some hyper great technology on that, that's a very closely guarded secret. And so saying that I've got a vendor who's who's going to have insight into how much of it I'm making and how big an investment I'm really making into that, that's a lot of trust that you're giving to a company that might have only been around for the last two years.

SPEAKER_00

I mean, also it's not even might not even be the company. Like if an AI is tied into it, does the company even know what the AI is doing in certain situations? This goes back to some of these autonomous agents where uh I've I've seen a lot of funny examples of OpenClaw, which is a quasi-autonomous agent that where somebody did a poor job of configuring it and it you know did some things that were not ideal. And it yeah, it's a little scary that you know the company may not be able to troll what the AI is actually doing because they don't know.

SPEAKER_02

Well, like the more autonomy you give it, right, the more likely you're gonna hear about those things. This is the game of odds, right? You'll hear as many success stories about, like, yeah, it it you know revamped my taxes and saved me, you know, a thousand dollars a month on my you know, phone bills and all of that stuff. And then you'll also hear stories of how it carpet bombed a spouse with, you know, 450,000 text messages or something. Um you know, you really can't uh put too many boundaries around it because even the frontier AI models, they're they're still trying to wrap their mind around what this thing would do if given the keys to the kingdom, right? So again, in our industry, risk mitigation being huge, that's risk from all sides, right? Like uh I don't want to trade upside if it means that I could potentially tank the company unless it's absolutely worth it and I'm in the kind of market where that will that will benefit.

unknown

Yeah.

SPEAKER_00

What what are the other things? Anything else around data you've seen that yeah, as you start thinking about AI at the edge, you know, the ownership at the edge versus like the models that are being provided by the OEMs, is there anything you've seen there that's uh that's concerning or that gets gets a little fuzzy?

Final Takeaways And Where To Connect

SPEAKER_02

Yeah, well, you know, again, I I know what a huge concern that is for people. And we operate in a weird space. You know, terminal emulation and telnet, um, many times they operate in dark network. You know, they they are air gapped in some ways from uh a lot of these a lot of these platforms, specifically for that reason. It just solves a lot of problems for them. They don't have to be concerned about that. And so there's a sort of uh there's a purity in saying you don't. Right? There's there's a a a speed that you can operate in when you're saying I'm not collecting this data and I'm not I'm not using it for anything, right? You don't run the risk of compliance. Um but there is genuinely data out there that uh that is valuable to people that they wished they had. And if you could find a way to anonymize that, if you can find a way to guarantee that in a way that uh that customers can be assured that they're they're not releasing key data to bad actors, right? And the bad actor isn't always a hacker who's trying to get at their information or a competitor who's trying to get insight into their roadmap. Sometimes it's just somebody who would take their data today and sell it back to them for$50 a month per user three years from now, right? So I think that's those are the strategies that we're looking at for you know AI and data and data at the edge generally. There is a huge amount of data in these spaces that people traditionally haven't had any insight or visibility into, um, that they'd obviously love to get to get access to. Um but how can we do that in a way that doesn't tickle the risk, that doesn't tickle the the people who are responsible for making sure that you're not uh setting them up for something down the road that they don't want to have happen.

SPEAKER_00

It's a yeah, that's a very, very large topic. I'll I'll I'll have to ask AI to research that for me to start for me.

SPEAKER_02

Yeah, yeah, yeah.

SPEAKER_00

Yeah. Just give me a summary on that, would you? We are uh we are actually at time and uh Justin, I appreciate you going through this. So uh a couple key things I took away. I think the the data one is it's really important as for enterprise systems customers, even as an individual, think about who owns the data, where it is. I always love the uh the the phrase, you know, if you uh you know if you don't know if something seems free and you don't know how they're making money, um you're yeah, you're the money they're making. If you aren't paying for the product, you are the product. Yeah, you are the product. Uh I think it goes to that. I think the the other point on the um OEMs are gonna continue the the field of dreams. If you build it, they'll come. They'll try different things and we'll we'll figure out what happens on the devices. It's gonna be interesting there. And then um, yeah, I think the use cases the the yeah voice will continue to be hard as as it has been. Uh vision's getting better, and we'll we'll see a lot more of the workflow uh at the edge with some of these AI advancements. So um a lot of exciting things. Um, Justin, if people wanted to learn more about you or or find more about what your your labs and your team does, where where do they go find you?

SPEAKER_02

Yeah, just uh reach out to us, staylink.com or uh find me on LinkedIn. Love to have a conversation about this stuff. It's like my favorite thing.

SPEAKER_00

Awesome. Justin, thank you for being on today. I appreciate it. And uh this is awesome. Have a good one.

SPEAKER_02

Yeah, thanks for the time.

SPEAKER_00

All right, cheers. Thank you for tuning in to Frontline Mobility Edge. If you enjoyed this episode, make sure to subscribe for more content every month. If you'd like to learn more about Blue Fletch, check out the link in the description or visit at bluefletch.com. See you next time.