(So you can skip ahead, if you want.)

00:00 Introduction
01:51 Industry Evolution and Speed
03:43 Infrastructure Automation Challenges
06:56 Defining AI Agents
10:20 Agent Orchestration at Scale
16:02 Operationalizing Copilot vs Autopilot
23:09 MCP Protocol Impact
29:32 Legacy Tech and Brownfield
38:05 Deterministic Outcomes and Wrap Up

William Collins • 00:12

Welcome to the show, everyone. First off… An update on the ground here. So last night in Kentucky and Southern Indiana where I’m at, it was an absolute snowpocalypse. When I woke up, there was snow everywhere. Crazy. But the most that I had to really do as far as planning was figure out what hoodie I was going to wear, get the kids to school, etc. I was going to think, as I was thinking this morning, Yvonne must have had it worse than I did.

William Collins • 00:44

She must have trekked through waist-deep snow to the she-shed, like some sort of cloud architect, Paul Bunyan. But then Yvonne disappointed me a few minutes ago because she’s working in her house. Right, Yvonne?

Evyonne Sharp • 00:56

That’s right. So the heater is still broken in the shed. For those of you who’ve been following that saga, we’re waiting to get folks out to fix it. We got some snow last night, but it’s gone now, but my kids are still home from school because I think they were concerned about the weather, and so it’s the worst of both worlds. Kids are home, no snow, working from the house. So I apologize in advance for any background noise you all hear today.

William Collins • 01:25

No worries. Same here. A lot of background noise, but yeah, good old technology. You can just kind of, you know, take it out. So joining us today is Chris Wade, someone who knows, I think, a thing or two about building companies, scaling technology and, you know, automation, orchestration, among many other things. How are you doing today, Chris? Awesome. Thanks for having me.

Chris Wade • 01:51

We don’t have any. There’s been no snow. No, no snow. It’s probably 50 degrees outside. So we’ll get ours soon enough, but not yet.

William Collins • 02:05

I’m kind of jealous, I gotta say. I was salting everything last night and it didn’t really make a dent in anything, so, yeah. So I guess the, sort of the reason we wanted to have you on the show, like the industry’s going through, as long as I’ve been working in technology, I think some of the biggest changes I’ve ever seen, and it’s not just the changes being big, it’s actually the speed in which things seem to be moving. You know, things always go faster and there’s always changes, but usually it doesn’t seem so daggone fast. You know, if you think 10 years ago, the conversation was, you know, a lot of your pedigree is automation and orchestration, but it was scripting 10 years ago. Like, how do I automate this one task? You know, people were writing expect scripts, you know, maybe some early Python, Perl, and the goal was just really simple, actually. Like, stop doing this one repetitive CLI thing by hand, and then it kind of shifted to automation at scale, and then orchestration where, you know, platforms and such kind of, you know, started emerging. But now this conversation’s shifting again in a major way. It’s not just how do I automate this task thing, it’s, or even how do I orchestrate some tasks? It’s how do I build a true system that can reason about what needs to happen? What is your take on all this, or how do you see that, Chris?

William Collins • 03:41

Just kind of the evolution.

Chris Wade • 03:43

Yeah, I mean, I think we’ve, you know, we’ve all been working towards improving how we operate our infrastructure. And even when you talk about scripting, I’m still alive and well today. You know, we’re doing that in my opinion because the infrastructure that we’re automating kind of requires us to do that. So, you know, infrastructure was built originally for humans to integrate with and we’ve had to use kind of machine technologies to integrate with that human interface. And I think we’ve had a variety of strategies over the years to deal with that, right? We’ve invented DevOps tools to make, quote unquote, scripting easier. We’ve built controllers.

Chris Wade • 04:30

We’ve built intent-based systems. We’ve tried to put REST APIs on things. We’ve adopted NetConf. In my opinion, these are all attempts to make automating infrastructure better. And as we start looking at adopting AI as a technology, I’m most interested in thinking about what changes do we need to make and how we automate infrastructure to adopt kind of that AI technology. And I think most people, we think about it the opposite way, right? It’s like, how do I integrate AI into my tooling?

Chris Wade • 05:04

And I think this is an opportunity where we’re almost going to have to think about it the opposite way. It’s like, what changes do I need to make, you know, in support so that I can have this reasoning be adopted because it’s going to be so transformational for kind of all of the automation and scripting and coding we’ve been doing for the past 20 years.

William Collins • 05:22

That’s usually not, that’s usually not the advice that you hear from a lot of tech companies out there like, hey, you’ve got to like think about what you’re doing on your side, you know, a lot of companies will just sell you anything like, hey, now we’ll sell you something that’ll make everything all better. But a lot of times what you do have is you have this technical debt that’s been stacked on and stacked on and stacked on over the years. And it does, like the bigger the change, it does require a change in just how you do. I mean, really, in this case, I don’t want to say everything because that sounds a little bit exaggerated, but so many things like procurement, lifecycle, cost management, FinOps is huge. Cloud and on-prem, like for a while, I mean, I knew this probably wasn’t going to happen, but many thought everything’s just going to go to the cloud and on-prem doesn’t matter anymore. But you really have hybrid, you have both, you have infrastructure everywhere. So it is complicated. And you made a few really good points there. And I think It makes me think like we should start out with some basics to kind of like an elevator pitch will not them not selling anything but the term. AI agent you know like most things that can i get started around a lot and. I think one thing that would be really helpful just to kind of clear.

William Collins • 06:46

You know clear context and in is it pertains to like network infrastructure. Is like how do you define an agent and what does it do with your network what is it even.

Chris Wade • 06:56

What is it like at the most basic level. So i think we’ve all chatted it up with our favorite lm and that’s been most of our experience of working with a technology so when i think about an agent i’m really thinking about. An autonomous system. that is leveraging the knowledge base in the LLM to have mostly a goal-oriented outcome, right? So we think about the limitations that we can talk about today of the technology, context windows, integrating with third-party systems, MCP, et cetera, and we have to think about how we can constrain the work so that as that agent is running and trying to achieve its outcomes, it can leverage the reasoning available in LLM to achieve an outcome. So just trying to be like super specific, you know, you could have an agent that is doing compliance checks, you could have an agent that does software upgrades, or you could do something very simple like an agent that updates tickets. So, you know, especially in the infrastructure space, I think that’s probably where most people think about where this fits.

Chris Wade • 08:14

We had some people at reInvent last week, and, you know, someone was talking about Blue Origin having 3,600 agents running. So, you know, we talk about these things as small, but there’s already some organizations out there that have vast, vast numbers of agents doing, you know, unique and bespoke things.

William Collins • 08:35

Speaking to that, so you kind of gave a few good examples there and you kind of made it seem like you’re going to have just very goal and like platform or thing specific agents. So like when we’re kind of like framing like, okay, how big can an agent get? Like do you have one giant agent for a whole company or do you have a giant agent for a department or a platform line or a device type or like, is there any rhyme or reason to how you think about that? Or is it just back to the context window and what it is that you’re interacting with?

Chris Wade • 09:15

Yeah, I mean, you know, we also use agents internally in our software processes. And I think, you know, there’s a lot of examples floating around of how agents work together to achieve an outcome. I think in the infrastructure space, most of the agentic stuff I’ve seen so far is very kind of task based. And I think we’re going to follow a similar paradigm to what we went through with automation. I see so many parallels between AI adoption and automation adoption in the sense that, you know, we’re going to start with some read only concepts. We’re going to start with some task based outcomes. And then I think over time, much like we let automation led to orchestration, you know, with the 8a protocol, you’re seeing that, you know, we’re going to start integrating these agents together with agentic orchestration to kind of achieve a similar outcome. So I think as people think about agents, I think, you know, where you started with scripts and task based automation, I think is a really, really good metaphor of how we should think about it.

Evyonne Sharp • 10:20

Well, and you mentioned, you know, 3,600 agents running in an environment. And so when I think about that, I think about, you know, the emergence of microservices and Docker containers and how ultimately we needed an orchestrator of Kubernetes to try and manage and orchestrate all of those different containers. Do you see infrastructure coming that’s going to be like an agent orchestrator? Because ultimately, if we have that many agents, it’s going to be pretty difficult for people to keep their eyes on that many agents to be sure they’re doing what you need them to to be sure that they, you know, aren’t getting stuck. How do you see like the larger agentic framework playing out when you have thousands of agents running in an environment?

Chris Wade • 11:13

Yeah, I, I tend to think about it like, uh, like a workforce. I tend to think about it like, um, you know, our traditional automation and orchestration concept in the sense that there’s a lot of unique roles achieving outcomes. And yes, we’re gonna have to figure out how to how to connect, connect all these things together. Um, you know, when I think about infrastructures, code pipelines running thousands of tasks, you know, you have to ask your question. Do you want human in the loop? Do you want human out of the loop? Um, certain activities, maybe more certain activities, maybe less. Um, you know, I tend to think we’re going to start in a very augmentation centric way. Um, I think that’s how most people are thinking about it. Um, so I think it’s, it’s, you know, we deal with infrastructure, um, as an industry. So I think we tend to understand the critical nature of what we’re doing. Um, and I think a lot of those agentic stories that are on a much larger scale tend to be things that are going to be like, uh, you know, help desk agents, things that maybe are a little bit more human centric, which maybe don’t have, uh, you know, dramatic downsides if, if their goals are not achieved in, in such a successful way. You know, when we think about infrastructure, we obviously are thinking about the, the, uh, guardrails that we need to have in place and the security that we need to have in place. So we have the confidence to kind of unleash these agents being monitored or unmonitored in our environments. Right.

William Collins • 12:38

Yeah, that makes a lot of sense. So kind of to build on that a little bit, so we’ve sort of established the what, like what is an agent and kind of what the goal of it is. And one thing I can’t help but think about a lot lately is there’s been some sort of like a tipping point, kind of like an evolution. And I just wonder like why now? Like why are we talking about this now? Like we’ve had machine learning for a long time. We’ve had automation for a long time. Like was there a specific like technological shift recently that you see that kind of made agents viable for like enterprise infrastructure?

Chris Wade • 13:30

I probably think about it a little bit the opposite way, in the sense that like when ChatGBT first came out, you know, it was a pretty dramatic moment in the sense that people started understanding the possibilities. And I always, you know, in our industry, we borrow and consume technology from so many other industries now, right, which is great. And I think it was the moment where it was like, how can we, consume this technology in such a way that we can advance our companies and our infrastructure. Um, so when you say, like, you know, why was machine learning either like not a good enough technology or what we’re missing? I think we’ve been trying to do this, you know, at least in the last, you know, 10 plus years we’ve been doing it at attentional. You know, I remember software defined networking was going to transform how we did this, um, intent based, um, you know, ITF with with new protocols and constructs, um, you know, Arista putting APIs directly on their devices, um, you know, all these things, um, have moved us forward, but I can’t I can’t shake the sum of times you talk to people that are automating at scale and you might have, like, you can’t replicate maybe all of the business logic that you want to take into consideration when you achieve these outcomes. Um, and for all of us that have built pipelines, you know, you start with the happy path, it’s a couple steps in a straight line, you’re happy with it. And then over time, this merge request failed, this environment was offline, you start building all this, all this workarounds.

Chris Wade • 15:13

Um, and before you know it, you have a pretty large technical debt laden, um, you know, orchestrated framework to achieve an outcome. And ultimately, this reasoning logic can replace a lot of that, and I think that’s what’s so, you know, the optimistic side in the sense that we can focus on the happy paths, and we can allow that reasoning, which to some degree replaces me or you staring at it and telling it what to do next, so that we can achieve so much more automation because the technical debt should be so dramatically reduced. So that’s, you know, it’s really how can we leverage kind of this, you know, inspirational technology to move forward, what we’ve all been trying to do for the past, you know, some of us 20 plus years, so.

William Collins • 16:02

Those were some great points. And, you know, one thing I see a lot out there is just confusion about, you know, copilot versus autopilot. I’m giving a workshop in January about copilots. And there’s just a lot of confusion about what is that going to mean for the way that we operate today. So we have a network operations team, network engineering, maybe network architecture for bigger companies. What is, like, in your opinion, like, how do these things actually get operationalized across a larger company? You know, as far as like starting out, you know, dipping your toe in the water, kind of figuring things out to actually having them in production in some capacity.

Chris Wade • 16:45

Yeah, I think we tend to what is it overestimate in the short term and underestimate in the long term, these types of technology waves. So as far as like, you know, there’s there’s a camp of folks, I think, that feel like this is going to, um, completely disrupt maybe how we’ve operated the network and think that it’s gonna be more takeover. And then there’s, I think I’m in the camp where it’s adding reasoning to what we already have. So I think from an operationalization perspective, I’m super interested in thinking and understanding how people have put automated bits in place today. Some people are, as you say, doing read-only activities. Others are doing some more full lifecycle stuff.

Chris Wade • 17:37

And I think it’s, we view AI as a tool. So the question is, how do I adopt that tool to kind of advance my operational paradigm? So I think it’s part of the journey. I think that we’re going to be able to leverage most of what we’ve done as a industry in the automation and orchestration landscape and be able to kind of like adopt this additional technology to, I think, overcome some of the limitations we’ve talked about that we just haven’t, some of the prior attempts with what we had at the time advanced, but didn’t get us over the hump for kind of mass operationalization. I mean, we all sit around and talk about, the hyperscalers do this. Maybe in the radio network, we do that. So I think in certain network technologies and certain network domains with certain kind of operational constraints, you can automate and orchestrate extremes.

Chris Wade • 18:32

And then you said, as in the enterprise where most of us live, I have old stuff and new stuff, cloud stuff and on-prem stuff. And I think your mileage varies depending upon what and how we’re operating.

William Collins • 18:46

Do you think, so, I guess in your opinion, was MCP or model context protocol kind of a catalyst to accelerate this a little bit with, you know, making the agentic AI conversation real with some of these big companies as it addresses, you know, it addresses, I think, a few pain points for sure. But what are your thoughts on that?

Chris Wade • 19:11

Yeah, I learned, I learned some stuff from your autocom for presentation. So whenever that’s up, um, but, um, you know, I, you know, the people that invent, I think it’s interesting who invented it, right? Um, you know, unless, unless my understanding is incorrect, you know, anthropic put it out. And I kind of maybe go back to us all being on a chat interface with our LLM. And I think, you know, it wasn’t that we bent the LLM to work with our infrastructure, right? This, this translation translation layer was kind of required to take our existing API driven world of machine to machine and, and have an interface that was more appropriate, uh, you know, for LLMs to be leveraged for those, you know, to understand this thing we’ve built. Um, you know, it’s like, why is, why, why are we building humanoid robots? Because we have a world built for humans. Um, you know, why, why do we need MCP? We have this whole technical, we have all these systems and applications and architectures. Um, I think we’re going to continue to see how do we rethink what we have so we can, you know, best use that. And I think it’s a, it’s a, it’s, it’s a great example, right? I mean, I’d love to hear your thoughts on it. Well, but, but like that kind of unleashed LLMs on, on, on, on machine learning. enterprise software so what’s what’s the next thing we need to be able to apply that technology and other. Either software spaces or or or or anything else.

William Collins • 20:53

Yeah, I think having open standards earlier on in this technology wave is a good thing with as much stuff that is happening. So having open standards, and this is like, in my opinion, A great place where real true open source belongs like these super foundational things that can be a foundation can adopt and you can have like multiple major players contributing to them and such. It’s not just like a little tool that does this or that like it’s MCP is pretty widely used now. I mean if you’re a startup out there you’re coming to market you’re if you don’t know what MCP is it’s probably on your roadmap or something you know it’s just been such a big deal. I mean mainly from the integrations perspective not having to handle and do these one-to-one integration you know with this model and this set of tools and such like it’s almost like I think in my talk I said going from like hub-and-spoke to full mesh as far as like you know the complexity of it but you also that’s what MCP is here to do it lessens the complexity because you have that you know like you said it I mean it’s essentially a translation layer you know at the end of the day so the less things that vendors have to manage piecemeal and that enterprises have to ask for piecemeal on a case-by-case basis like if you can eliminate a lot of that then that way you know companies vendors included and enterprises can focus on doing things that are actually valuable and impact their business. What about legacy tech though that’s something I think everybody always forgets about the old you know the legacy all the brownfield out there where does brownfield fit into this world or does it fit into this world because it’s it’s hard enough to even do basic automation with some of the.

William Collins • 22:55

some of the old stuff out there. A lot of it’s CLI driven, powered by tribal knowledge, etc. We’ve all had that conversation a million times. Can AI agents even work in this arena?

Chris Wade • 23:09

Well, uh, some of my comments earlier were how we tried to do intent based and abstractions and, um, model driven and all that kind of stuff. And, uh, to all my equipment vendor friends listening, um, the actual infrastructure. I don’t, I don’t think we’ve dramatically changed it that much, uh, in support of automation. Um, I think the controller strategy has been adopted and it’s been great, right? Because it gives me software control of my infrastructure. Um, but I, I personally do not expect dramatic changes from the infrastructure in support of this. I think it’s going to have to be done, um, in a very similar way to how we’re thinking about doing it today for, for the foreseeable future. Um, you know, with the exceptions, like we spoke about of very unique network topologies or technologies that, uh, maybe are the same, but I. Maybe I’m underestimating what’s possible, but I’m probably not assuming an NLP interface on a device too soon.

Chris Wade • 24:19

But maybe somebody will prove me wrong.

Evyonne Sharp • 24:23

What I do wonder, though, is because LLMs are extremely adaptive, and because they’re able to discover and learn, a lot of the challenges with automating and integrating with Brownfield are you’ve got variations of devices and device types and configurations that don’t lean themselves well toward very prescriptive automation. But I do wonder if the LLMs are going to give us a little bit more flexibility, where, you know, we talked about intent-based networking, but everything was still very prescriptive under the covers. Sure. LLMs give us a new interface to be more adaptive in describing what we want. Now, the question is whether they’ll be able to effectively execute that. But I do think that there’ll be operational and toil improvement just because the LLMs will be able to understand a broader range than what we could prescriptively automate. I don’t know if either of you see it that way, but I do think that there’s at least potential there that maybe hasn’t been there in the past.

Chris Wade • 25:43

Yeah, maybe one interesting thought on this is that, like, exactly as you said, I just think of the LLMs as kind of been trained on what’s on the Internet, right? So I think it has like extremely good general knowledge. But what’s interesting is, you know, not to pick on Cisco, but like, there’s only so much Cisco, like iOS knowledge on the Internet, like Cisco has most of that just to use them as a particular vendor or AWS. So it’s like, I think we’re going to have to get some specific knowledge from those vendors that have, you know, unique. knowledge that maybe, maybe is not in the LLM to the degree to support what you’re talking about, right? So it’s, it’s, um, it’s really interesting just to see what people are doing. They’re training it on all sorts of PhD level candidate people are feeding stuff into these LLMs to have specific knowledge on things. It’ll be interesting to see if we go the route of like, putting more of that knowledge into these LLMs, or if you’re going to see maybe more, more, maybe specialized LLMs that we’re going to have to piece together over time to have that, that unique knowledge of some of those things. I don’t know if you’ve thought about it that way.

Evyonne Sharp • 26:57

The cynic in me is going to, is, is, is going to suggest a subscription-based service that provides access to a custom LLM that’s trained on all of Cisco documentation and internal knowledge. But I mean, it’s, it’s not a. You know, it’s not that far-fetched that you would take your intellectual property that you know internally that has some very specific value to the customer if you could demonstrate that it would allow them to better operationalize, be more efficient in their offerings. I don’t think we’re too far away from a world where there will be more customized models that folks can subscribe to in some form.

Chris Wade • 27:45

Exactly. And I think today when you’re building a strategy around this, we have to deal with the generic knowledge of the models that we all have access to, understanding that there’s going to be specialized knowledge in the future. You know, specialized knowledge on, you know, in our space we think about infrastructure, but, you know, whether and otherwise, right, there might be all sorts of models doing all sorts of things. And, you know, to get to hundreds and thousands of agents, you probably think that they’re, you know, those agents are all using reasoning from some large language model or other type of AI infrastructure. It’s probably going to be a lot. It’s probably going to be a lot more than what we think of today.

William Collins • 28:29

Great points from both you there. One thing I’m just thinking about now is a network engineer. We like to do something and have the same result every time and our automation, we’re deterministic with all the things, which is why our networks are so reliable and they never go down, of course. So what, as far as like AI agents in this conversation we’re having, we know so much from history about building technology, certain things have boundaries, you have separations, you have demarks, this is what this does, and it never is mingled with what this does over here. So what something you’ve talked, I’ve heard you talk a little bit about is kind of like how do you balance the AI reasoning with this and the deterministic outcomes that you want? These are two different things, but you know, they’re both very important. What are your thoughts there?

Chris Wade • 29:32

Yeah, I think this is probably one of the I mean, for me, it’s it’s how I think about it mostly in the sense that like, it’s the guardrails, if you can say that. So I think you’ve spoken with MCP, how you can limit what’s being done, you can have, you know, tool consumption, you can kind of tweak what you’re going to expose to the reasoning model. I think most people are concerned with agents kind of, you know, quote, unquote, doing what they want to do. So one way to solve it is to constrain what the agent can do. And that’s what we do with MCP tools. And then the question is, how much of the things as you said, do you want to happen the same every time? And then what things do you not want to happen the same every time? So I would, I would, I would argue that when you’re doing what most people think of as a happy path, and everything goes just the way you want it, you want that to be mostly deterministic. When you have a long tail of error conditions, you probably would like something to reason through that versus having to code up, you know, many, many error conditions, which might not happen very often. And I think, you know, The models are only going to get better, and I think the thought is we have to put probably more deterministic stuff in line now, and then maybe over time, we can adjust the level of determinism that we want kind of within the infrastructure.

Evyonne Sharp • 31:04

Let us also remember, you know, 15 years ago, working in an organization, doing networking, where we had an executive level leader saying we want the network to be self healing. because that was the word at the time. Then we had a networking manager demand that we use static routing so that we could understand exactly what was going on in the network. I do think as the technology evolves, we’re going to be more comfortable once we understand it, with what right now seems very scary, which is that lack of determinism. I think first of all, we’re going to figure out where it needs to be deterministic and how to make sure it is. But I also think we’re going to become more flexible over time with allowing the model to within certain guardrails, as you say, make some changes and adjustments. If we’re in an agentic world where it is the model who has to understand all those things, and that functionally make the changes, it may matter less.

Evyonne Sharp • 32:20

It’s unfathomable that you could have an enterprise network that’s basically built upon static routing today. Because it’s going to be so brittle, it won’t function well. I wouldn’t be surprised if we look back at this era and say similar things about how we operate our environments. Couldn’t agree more.

William Collins • 32:48

Yeah, like I couldn’t agree more with both of you. It’s just funny. My mind’s like swirling now because in this year in 2025. I’ve talked to someone with a large company that was actually using agents already in production. Many that are prototyping MCP. I’ve talked to someone that is removing static routes from their network and they’ve been working on that for years. I’ve talked to someone that is still like I think 10% into deploying SD-WAN and then they’re still doing multiple MPLS providers with the SD-WAN as an overlay.

William Collins • 33:27

And I talked to someone just I want to say in the last two months that is still using a combination of spreadsheets and some PHP tool for IPAM. And like what a what a widespread of technologies.

Evyonne Sharp • 33:42

The universe is large and varied.

William Collins • 33:45

Yeah, but I remember when it’s it’s got to be so hard for just Someone that’s coming out of, you know, a comp sci degree or just trying to learn. Like, when I started learning about networking, you had switches, routers, and firewalls, and BGP, and DNS. The list was pretty short. There wasn’t a ton of different things. Now, I don’t even know where I’d start. I’d probably get scared and become a plumber. Go on, you know, become an electrician or something, I don’t know.

Chris Wade • 34:17

Well, if you think about it, like, networking’s made its way into every nook and cranny of our universe, right? I mean, my kids, when the power goes out and their Wi-Fi doesn’t work, they don’t understand why the Wi-Fi doesn’t work even though we have no power. So, you know, I think it’s also a testament to kind of how impactful the technology’s been to the point where you do have some networks which are maybe more rigid and need to be less robust, and you have others that are, you know, critical infrastructure on a national scale and everything in between. And I think that part’s going to probably continue to expand. Yeah.

William Collins • 34:59

Yeah, so like, what do you, do you have any sort of guidance or advice or anything? Because I know that… I don’t know if you’d be able to talk about it, but the intern program that iTential runs, bringing in interns and just kind of, what does that look like going into the future? Because I imagine that changes with how fast the technology changes and you almost, I think in my mind, I want to view it as like a NFL draft, like you have these folks that you’re bringing out of college, you’re placing them, you’re getting them integrated with companies, like they’re winning, you’re winning a little bit. Maybe they come back and take a full-time role, but what does that look like?

Chris Wade • 35:40

Are you asking how AI has changed the program or how those students think about computer science and the job market?

William Collins • 35:47

Well, just like a leader is a leader of a company, like what advice would you give to those kids that are coming out of college that are confused with all this stuff? Like where do you start? How, because common sense, like if I was, I can tell you right now with my personality type, if I was a young person and I was trying to get into tech right now, I would see, I’d say, okay, AI is the money thing, MCP, the AI, the agents. And I would skip about 10 foundational things to try to jump right on to what’s hot, which probably isn’t the right thing to do, but just, yeah, advice.

Chris Wade • 36:25

Yeah. I mean, yeah, culturally we’ve had co-ops here from the beginning. I was a co-op at the university across the street. So it’s kind of like culturally, I think, super significant. I will say they’re very bright. The college students these days are unbelievable. But I would say they do have that, I’ve seen it over the past 10 years, that they like, if big data’s the thing, then they wanna do big data.

Chris Wade • 36:53

So when you go to the career fair and you see the TikTok line has 1,000 students because that’s the biggest big data problem that they could think of solving, they’re definitely swayed by that. I guess I’m super biased in the sense that a lot of computer science folks have been saying, hey, with AI, go study humanities or other types of things. I still think we need a tremendous volume of hardcore computer science education for all the industries we need. But the feedback we get is they love to understand how AI can impact their work. I would say university systems, education systems maybe still teach more classical concepts. Intelligence as a thread, I would say almost all of them take that because that’s the en vogue thing of the day. But they love to mix the practical with what they learn in school.

Chris Wade • 37:50

But yeah, I think. The future is bright for those folks, but yes, they very much are interested in AI intelligence and whatever the latest technology thing of the day is for sure. I love that. Just like we all do, right?

William Collins • 38:05

Yeah. For sure. Yeah. The future. I think the future is bright. I was actually reading, so my wife sent me something last night, like right before bed, which was great. So I started reading this and I couldn’t go to sleep because I was looking up other stuff, but whatever.

William Collins • 38:20

But it was a thing of the age group right now between like, of course, I know we’re you know, we know we’re not supposed to drink alcohol in the U.S. at 21 years old, but the survey was from like 18 to like 29, I think it was under 30. And it’s like the alcohol consumption is just diving with the younger generation completely. Like their tendencies are changing, they’re staying at home a lot more, they’re not going out and getting in trouble or doing different things. I mean, maybe that’s a bad thing, not getting out as much, but I just think about when I was growing up, the tendency was, hey, I never want to be at home. I want to be out. I want to be doing something all the time. And of course, a lot of the, you know, when you’re young and you’re maybe you’re in college or high school or whatever, you’re going to go party and do your thing.

William Collins • 39:12

But yeah, it’s like on the downward spiral, which I thought was kind of interesting. I wouldn’t have expected that. But yeah, kind of a sidebar. I’ll link the statistics. But anyway, is there anything else on your side of home that you can think of that we haven’t covered?

Evyonne Sharp • 39:31

No, I think I’m good. I think as we look toward 2026, it’ll be interesting to look back and see if whether things moved more quickly than we expect, or whether they moved more slowly. I feel like we’re seeing both of those things happen at the same time. The technology is evolving really, really quickly and changing, and we’re seeing so many new things. At the same time, I don’t know that we’re seeing the world around us change with it. I think there’s this interesting lag going on right now. It’ll be interesting to see if all of these developments we’re so excited about, how they make their way into the broader enterprise, the broader culture in the next year, and whether we see really capable functional solutions start being deployed, or if we’re still trying to figure out exactly what the technology looks like and how to apply it.

Evyonne Sharp • 40:35

But I think it’s going to be an interesting year next year for sure.

William Collins • 40:38

Yeah, agreed. Yeah, kind of funny you say that. I started recently jumping into the vibe coding thing, trying to vibe code a little bit, like through the command line and everything, actually trying to do real things with it a little bit. And I… You just have a lot to learn. There’s so many, you know, the thing I was trying to do in particular was getting it to do code commenting, actually fill out the pull request with GitHub CLI and like go through the whole flow and you gotta have that stuff locked down pretty tightly or you get a lot of stuff up there in your version control that you wouldn’t want to have in my little lonely test repo up there. It didn’t work out as slimmingly as I thought it would in my mind. There’s still gaps, for sure. Anyhow, where can the folks find you, Chris?

William Collins • 41:29

Are you all over the internet?

Chris Wade • 41:31

I would say LinkedIn would be my primary contact for anybody listening who wants to reach out, talk AI or automation.

William Collins • 41:40

So, you didn’t put the TikTok up with the dancing videos or anything? No. Okay. Yeah. Too bad. All right. Well, thank you for joining the show.

William Collins • 41:50

And yeah, thanks everybody for tuning in. Cheers, guys.