Shawn Cordner (00:13.806)
Hey everyone and welcome to the Amplitude of Tech Podcast. I’m Shawn Cordner, Chief Marketing Officer of Amplix, and this is an episode that you have to listen to. There are two things in this episode that you need to know about that you need to understand because it’s going to impact your budget next year. I’m not going to tell you what it is, just kidding. We’re going to talk about the AI costs and how the token use is exploding and boardrooms are starting to ask for ROI. We’re also going to talk about the VMware Cliff that we’re walking towards in October of 2027. Stay tuned. This was a great conversation with Bryan Smith, CEO of Expedient.

Shawn Cordner (00:53.934)
All right, Bryan Smith, welcome to the podcast.

Bryan Smith (00:56.654)
Thanks for having me.

Shawn Cordner (00:58.104)
Happy to have you. maybe just take a second, talk about who you are, what you do and where you come from.

Bryan Smith (01:04.302)
Sure. Bryan Smith, I’m the CEO with Expedient. We’re a intelligent infrastructure provider. So we do private cloud services, physical data center services, and then AI as a service as well.

Shawn Cordner (01:17.23)
All right. That was the most concise explanation of what a company does that I’ve heard on the podcast yet. So good job to you and your marketing department. so maybe just start off you know, telling me about what you’re seeing in terms of AI spending right now.

Bryan Smith (01:33.262)
Yeah, the the it’s one of the biggest topics people want to talk about because there’s this explosion of of spend as adoption really increases. And obviously there’s lots of different factors and reasons that come into that that are different based on different situation where different companies are.

Shawn Cordner (01:49.806)
Now, I’m hearing a l a lot about companies that are maybe not a lot, but I’m hearing in the news about companies that have blown through a year’s worth of AI budget in two or three months. Are you seeing that as like an outlier or is that a trend or is the truth somewhere in between?

Bryan Smith (02:05.854)
I I think you’ve seen that from a couple of big brands like Uber and Visa that there’s been stories out there are probably the biggest that that people reference. And we definitely see people that get surprised by AI spend and it’s mainly the way that they’re where they’re at in the cycle. Because when you think of a lot of organizations, they’re in a build phase right now. Excuse me. They’re in a build phase right now. And you think about it like if you were building a house

Think how much more construction material and how much more you would need to purchase from a Home Depot versus if you were just owning a house and doing some minor updates along the way. And so most companies are thinking that they’re starting AI in a run phase and they’re completely forgetting about the build phase. And it’s similar if you’re doing like a huge software deployment, you have an implementation fee that goes with that. If it’s a Service Now or a Salesforce or anything else. And

that’s something that seems to be overlooked when they’re approaching AI because it’s not really just a software purchase, it’s not a hardware purchase, it’s something in between that they have not really done before. And it’s almost a new utility that they’re bringing into their business.

Shawn Cordner (03:19.544)
So there’s like a total cost of ownership that people are are not really evaluating, right? It’s they’re they’re looking at a specific project, they’re not thinking about the infrastructure. I’ve seen that with software as a service personally in my career where I made the mistake of buying a marketing automation software that, you know, just barely fit the budget given the price tag for the the licenses. I had no idea how expensive it was going to be to hire developers to get in there and actually do the work to build it the way that we needed it.

built out. And worse was, you know, we made a decision that it was a small company that we went with and so the developer pool was a lot smaller. And so the cost for development was a lot higher than it was if you were using something like WordPress. So is that a good analogy for what’s going on right now?

Bryan Smith (04:06.52)
Yeah, it’s the upfront cost that people don’t see, but it’s also the use case that you know you have all these people that are trying to build things that they’ve wanted for years and didn’t view that they they could do it. They couldn’t build it themselves. And so it’s changed the dynamic. And when somebody’s just using like an AI chat compared to somebody that’s doing software development, you you’re talking twenty to a hundred times more usage from the person doing development.

than your average employee. So you really need to segment who are the users, the daily users of AI to help them do their work versus who’s leveraging AI to write new software, write new applications, because those are incredibly different profiles.

Shawn Cordner (04:49.676)
Yeah. Are we also seeing ’cause it sounds like what you’re saying is people just didn’t see the cost, but I feel like we’re also seeing an increase in costs. Like if you took a like for like project from last year to this year, would that project have costed more this year?

Bryan Smith (05:08.192)
It doesn’t really work that same way because in one aspect, a frontier model, like the best model that’s available a year ago compared to the best model available now is about three times more expensive. So the way that you would say it, well, price has gone up. But that model is so much more efficient. If you look at the same c capability, the model from 18 months ago, think of that like a a Chat GPT four.

That same capability now is over 200 times less expensive than it was. You know, so it’s more the combination that the model capability has gone up dramatically, so that it’s more expensive for per use, but it’s actually a lot more about the number of users and the volume of consumption has greatly outpaced actually the price increase.

Shawn Cordner (05:56.92)
So for example though, I think we’ve seen that Anthropic had changed the tokenization the way that they measure how many tokens are being used for a specific prompt. And I think that was ending up costing people like thirty, thirty five percent more. So are you seeing indications? Another example of that might be that AWS this year, they’ve increased prices on GPUs twice, one in January and one in July. so that seems to be a trend where

You know, it’s it’s kind of like the the social media model to me. It used to be that Facebook would give you a Facebook page if you were a small business. It was free. You could get a thousand people to follow you. You make a post, a thousand people follow it, they get you hooked on the business model, and then they say, Wait, now only a hundred people are gonna see it unless you pay for advertising, right? And so, did we get basically sold on the drug dealer model? Did you know these things come in at a much cheaper rate when we started?

And now that we’re hooked, they’re starting to ramp prices up.

Bryan Smith (06:57.75)
Yeah, I’d say there’s there’s three factors and you hit you know two of the three. So one, the the price increase that you mentioned from Anthropic is really a model change. So the price has gone up for their best model, but that model again does a lot more. So I don’t view that one to be in the same bucket as the other two. The GPU going up, absolutely, you know, we’re seeing that in all instances that you know, even non GPU servers, you know, if you look at now versus eighteen months ago.

That’s a 400% price increase in in servers. And you’re seeing the same thing in in GPUs. so you know, some of those things are driving the increases being passed through on that side. Now, squarely 100% of what you’re referencing, I think is more, you know, where you had different models that you could purchase in a flat rate, somewhat in all you can eat. And that’s more of what I would say is the drug dealer model, know, where they’ve said, okay, for twenty dollars a month or a hundred or two hundred dollars a month, you get this giant

pool. And if you think of like an anthropic, they have their 20 max plan that’s heavily subsidized versus if you bought that same quantity of tokens directly through their API, instead of being two hundred dollars, you’d be you know twenty five hundred or five thousand dollars in expense. So they’re massively subsidizing kind of the flat rate plans and you’re already seeing people start to move away from that. The first really was Microsoft Copilot in the development side.

you know, went away from that flat rate to consumption. So I expect that trend to be really the norm where people have to get better at the observability and measurement and the controls as everything moves into that consumption driven platform.

Shawn Cordner (08:41.346)
How much of this also is maybe just bad habits or people not learning yet exactly the right way to use and to configure these models? Like for example, bad problem to engineering or bloated context windows, not caching. there’s two more that I want to talk about, but I want to talk about them separately. So what do you think about that kind of like user behavior type stuff?

Bryan Smith (09:01.91)
I think all of those are you know very accurate that you know from those different pieces you can reduce the the consumption dramatically and even more so from picking the wrong model. And I I think that’s probably the one you’re looking to hold for the a completely different topic. But if you for example, if you can do proper prompt caching on the input tokens, you can reduce the consumption by ninety percent. Doesn’t do anything on on the output side.

And then also one that people don’t think about very often is a batch process. You know, so almost all of the different models have an option that if you don’t need a real time response and it can wait a while, you know, it’s a work that’s being done to summarize files or something, a report that’s happening the next day, your cost of token is actually about fifty percent less when you’re using public models. But that’s even before we talk about you know looking at using the right model for the right reason.

overall, or do you use a public model versus a private model?

Shawn Cordner (10:02.966)
Yeah, and you’re right. We’re gonna talk about the the right model for the right reason and and how to manage those costs. But the other one that I didn’t mention yet that I thought should also be discussed separately is agentic workflows or runaway automation. So I think sometimes people are implementing AI in places where maybe it doesn’t belong or they’re building automations and they really didn’t predict the potential I don’t know, the the externalities that could drive up the usage of that workflow and

This is not an AI work workflow usage issue necessarily, but an example is we use, you know, Zoom Info as a a data platform to help us you know for sales and marketing intelligence. And we found out that we used up all of our credits for the whole year in the first six months. And actually it all happened in one month, and actually it was because one user set up an automation without thinking about what that would actually do.

forgot to turn it off and we ran through like fifteen thousand credits in about five minutes. So are you seeing stuff like that?

Bryan Smith (11:05.878)
Yeah, that that’s very much the the automation side and just people understanding APIs as a whole and not understanding what the different limits are, what the the thresholds. Yeah. I’ve done some of those things myself where I set up an automated process again for something that would be hard to do in the past of doing like tens of thousands of web searches on different things and you know burn through you know fifty thousand API calls in an hour. And and so but at the same time

I got work done that I couldn’t have done previously. So there has to be a balance to say, okay, all the work that was done on your side was Zoom info. If it was done and it’s work that you wanted to do, you may have just compressed the time window of consuming those credits, but you also may have compressed the time window of getting the benefit. You know, so there’s combinations of having the right observability so you can see what’s going on, the right type of controls that are that are happening, but there is also a shift in mindset of

If you do the proper scoping on an agent or workflow or whatever it may be, and you’re scoping it correctly, you should understand the value that you’re getting out of the other side. What is the offset of the work of how it was being done previously and building that total cost of ownership to say running this agent cost X every time? And then you can think about even how do you optimize some of that.

Shawn Cordner (12:28.258)
So that leads us into the forecasting problem, right? So maybe this is just such new territory that it’s hard to think of all these things, but why are we seeing so many projects fail to produce an ROI? We’re definitely seeing it in the pilot phase, but then I think we’re also seeing it in the scale phase, maybe even more painfully.

Bryan Smith (12:52.972)
I’d say that there’s a few different factors. The biggest for sure is people think about observability in the ROI tracking last. It’s more can I build this? How do I build this? And very little of should I build this? And so those components and not having them on the front end or not having them part of the the project is a a big impediment. but then I think some of the bigger pieces are they choosing the right type of thing. and

Often they’re going after kind of the magic wand, the thing that they’ve always wanted in this giant project that takes a long time, data’s not ready. And you know, so if they kind of work backwards to say, okay, what data do we have right now? And is it in a clean state? And only green lighting projects out of the gate that actually have data in a good quality state that that’s usable. And then you need to look to say, you know, what are higher volume type of transactions and having metrics on what’s the value of each one of these transactions.

How often does it happen? And then also running that through, is it using external API calls? Is it using external tooling? Does it need a human to do other pieces along the way? And factoring the the cost of that entire workflow, both the AI component and you know any external tooling to really say, here’s my actual cost to run this. Here’s the alternative if I didn’t do it. Is this replacing work that was being done previously? Am I being more efficient and getting more work done?

that I wouldn’t have been otherwise. So I think a a lot of times people are more of can I build it? Should I versus should I build it overall?

Shawn Cordner (14:32.088)
It almost sounds too like there might be a chicken before the egg problem here where maybe total cost of ownership needs to be calculated before ROI can be calculated effectively and maybe that step is being missed.

Bryan Smith (14:43.862)
Yeah, I think that’s just part of the overall governance miss right now because you have this world that the majority of the AI work that’s happening, I would consider shadow AI. That, you know, most companies don’t have the breadth of AI tooling to give to their overall users. So people are picking and choosing their own capabilities. And that could be their own subscription that they swipe a credit card. That could be new development platforms that are running applications in a business that like a CIO would have never approved. And

It’s not that they’re do doing things maliciously. It’s more that the business isn’t keeping up and isn’t giving them kind of that platform and the capability of what’s the right thing to use or the education, the enablement. So they’re using what they see and find along the way and what they get comfortable with.

Shawn Cordner (15:31.862)
Yeah. That sounds like a a similar problem to software as well. So it maybe just take the analogy a little further. Like is AI harder to budget for than traditional software, or is it the same thing and we’re just missing it the way that we missed it for software?

Bryan Smith (15:48.662)
I think it it’s very different that you need to think of AI as a completely new infrastructure platform, which has so if you go back to and compare it to infrastructure in your software, there’s a data center it runs in, you know, there’s hardware it runs on, there’s you know, people that are managing that, there’s network, and then there’s the software layer. And people are only really thinking the comparison I buy a software license, but all the other decisions that went into that.

And AI is using all those different components and you have to think about not just the software license to have access to use it, but it’s how much do you use. So it’s it’s a lot closer to more of an infrastructure as a service or when people did on demand licensing or on demand usage from hyperscale cloud because they didn’t know exactly how many compute cycles and you know so you had this runaway cost.

So I think there’s a lot of things in AI that you can go back and think about of what were the phases we went through with cloud and what did we learn. And if you are a liter little retrospect, you can avoid a lot of those challenges because think about the phases of cloud. It started out with cloud could never run production. We just that has to all always be on premises. And then you know it quickly went to everything was cloud first. And there was this exuberance that everything was going to be there. And then it was.

my gosh, our cloud spend is going through the roof. We need to dial that back. And it ended up that hybrid was the right solution. And that same trend is happening with AI. It started with block everything. And then it’s like the board said, give me some of that AI, and then they’re getting a runaway cloud spend. And so if they kind of just step back and say, Well, what did we have to solve for to get cloud and infrastructure under control correctly? It’s almost the exact same things that they need to solve for so they can have the proper governance and controls.

From an AI perspective as well.

Shawn Cordner (17:48.278)
One of the areas where I think we see cost overruns for software as a service, licensing. And similarly with cloud is setting something up and then forgetting about it, right? So you’ve got licenses for users that leave the company and you never turn those licenses off, or for people that are not using it actively and you could have turned those off as well. On the cloud side, you’ve got spot instances instances that you forget to turn down and so they’re just running away.

you know, eating up money. are you seeing the same thing with AI? Is it you know, just a failure to manage closely?

Bryan Smith (18:29.45)
Absolutely. And I think you need to kind of split those apart on ideally you’re looking for a way that you can kind of pool usage across different people versus just a pure license you pay for it if you use it or lose it. versus if you can find different AI services that allow you to pool that gives you a big advantage on the user side.

On the agent side, I think the biggest miss is people n I would recommend quit calling them agents and think of them more as a digital employee or a virtual employee. And you know, employees need management, they need oversight, they need directions, they need coaching along the way. And to provide that, you also need to have observability that if you had an employee that you never talked to, didn’t inspect their work, didn’t watch along, ride along with them.

it’d be difficult for you to c to coach them well. And that’s the piece that I think a lot of people are missing is that mindset shift that they need to manage their agent fleet very similar to the way they would manage a team of employees, which requires the right observability. And you need to be able to not just see the cost of what happened, but actually understand the logic of what steps went through in the thought process, what models did they use, how big were the prompts.

So that then you can actually start to do extra true management to say, well, what if I changed the model to something that was much less? What if the response came back in ten minutes versus ten seconds? how does that change the business? Or what if these three steps aren’t even needed, how does that that removal of data does it really change the quality? But you need to have that observability and that level of control of each of the different steps to really properly manage and

get to something that’s giving you the value while spending the l least amount in resources.

Shawn Cordner (20:27.518)
Once you have the observability, obviously you need to have some sort of process and governance around that process in place to look at it and understand it and take action against it or using that information. what does that actually look like? How like I I hear everyone talking about observability and how important it is for this and for that. And I just I I don’t I don’t know who owns that.

Who’s supposed to be the person watching that? How closely do you look at it? Is that a manual process or can I just export a a bunch of data out and run it through my cloud and it can tell me that we’re doing just great or tell me where to fix it? I mean, what what does it actually look like?

Bryan Smith (21:08.77)
Yeah, that I think a good analogy would be like a speedometer in your car doesn’t limit your speed. It actually just tells you how fast you’re going. And that’s what most people are doing is have a dashboard of here’s what my spend was. It’s not actually being predictive or or real time. So the governance that you’re talking about is like we the way that we’ve deployed it ourselves and what we do inside of our AI control platform is having that governance in place. So it’s not just a view of what happened.

But you can set controls and it could be by individual, it could be by team, you can use role-based access control, it can be by agent, and you can have different notifications. So it could message you in Teams or Slack or open a ticket, or you can have it set up where it just throttles usage. So performance just degrades so you don’t have outages, or you could even have where it defaults from an expensive model to a less expensive model. So having the ability to actually put controls in place and not just see what’s happening.

is really that next layer that it’s difficult for somebody to build themselves and that’s really where leveraging a service provider to help give you th that platform that has the that observability, the controls and the optionality to to kind of leverage different tools and bring that visibility all in one place, I think adds a a lot of value.

Shawn Cordner (22:30.094)
gonna strain your analogy just a little bit if I can. I if I am looking at my speedometer and I’m going too fast, I can make the decision immediately to take my foot off the gas or hit the brake. I’ve got recourse that I can take myself immediately, and there’s very little lag in time between the moment I see that I’m going too fast and the moment that I can take action. So what does that look like in an enterprise?

Bryan Smith (22:59.17)
Yeah, th I mean the way that you can set something like that up is actually building a agent that would be working within your governance dashboard so that you could set up a Slack message that if a team hits you know fifty percent of their consumption and they’re twenty percent through the month, or you could set up whatever variable that the line of business owner can be notified and say, Do you want to continue this? Do you wanna make edits? Do you want to switch models? And you know, you could also have that same agent have you know tools available in suggestions to say,

about switching models or it could say that you have an employee that is, you know, writing software with a model that that’s not what their their intended role really was in the business. So that is it one rogue user versus the entire team. So having those not just the visibility but the controls in place and then having agents that can do that notification proactively that gives you a lot more control versus if you’re just looking in the very mirror.

Shawn Cordner (23:57.836)
Is there a gap between having the capability to do that and enterprises actually doing that? Or are you finding that enterprises that are actually spending real money on AI right now are engaged at some level in those practices?

Bryan Smith (24:11.456)
it’s something that’s becoming incredibly important to them quickly. I I don’t see many that that have a lot of maturity, because many people have tried to build their own. And so it’s you end up with very different results versus it it’s the reason that I buy a car, I want reliable transportation versus building my own. And that’s the the world that we’re in right now with a lot of AI is you people are building should be building things that are unique to their business.

Things that are common and needed across everybody, like an observability, like a control air, is something that they should likely purchase and get a better outcome.

Shawn Cordner (24:48.95)
Yeah. You had mentioned earlier in the podcast here that boardrooms used to be, you know, screaming out, give me some of that AI. Right. And that that was definitely the climate at the time was we just we need I we need AI. We don’t know what that means, but we need it. We got to get it, go get it for me. I feel like we’re not hearing much of that anymore. I think what we’re hearing is boards and C suites are starting to put some pressure on the investment in AI to generate.

ROI, so do you think the honeymoon phase is over?

Bryan Smith (25:21.378)
I think that what we heard last year was from the early adopters. And you know, that’s where it was giving some of that AI. I would say for the masses, you’re still not even there yet. And that you just still have a lot of people testing and you know, but you are starting to include questions on how do we ensure the ROI? Or they even they’re starting to have enough experience that people have tried different things and they understand the AI project is gonna cost X.

Or if they plan it’s gonna cost X, they need to plan it to really be for X by the time they actually move it into production. Because everything was so new, I don’t think there was any malice. It was just new and they didn’t have and folks didn’t have an experience in that yet. And collectively people are getting enough experience to come with better use cases right out of the gate, things that are better scoped and more proper expectations. So that if they can pick a use case that is the right size.

and actually has an an observable ROI, they are getting approved as fast as they can. When you just look at the demand for hardware, the backlog for availability of things, you know AI for sure isn’t slowing down.

Shawn Cordner (26:33.516)
Yeah. It’s been a few episodes since I mentioned the famous or infamous MIT study about whatever it is, ninety-five percent of pilots. Sure thing. So I’m gonna dust it off and and bring it up here. But it one of the things I it’s it’s controversial f I think for a few good reasons and data can tell any story you want it to tell if you try hard enough. So I’ll leave that aside for a second. But one thing that I think it did point out that I

usually hear reinforcement of is the the pilot to scale gap. So from an ROI and from a cost projection perspective, where do people e people usually go wrong from looking at the cost of a pilot to looking at what it looks like in production?

Bryan Smith (27:20.802)
I took the biggest is that they’re only looking at the AI kind of initial build cost. And they’re not taking into account that you know building is what’s become very easy in software. Hosting it, running it in production, running it at scale has become is much harder than it than it’s ever been. And so they’re not taking into account an entire platform that can enable those capabilities at the front end that they’re just

hacking away on the software side and not thinking more holistically of everything that’s involved to move from you know building to production.

Shawn Cordner (27:59.214)
I think you’ve called that the the great inversion. Was that your language? Yeah. So what does that mean? What do you mean by the great inversion?

Bryan Smith (28:07.48)
Yeah, so the the the great AI inversion or great inversion is you know what was hard is now easy. What was easy is now hard. And what I mean by that is it used to be hard to build software. You had to have specialized skills, specialized people, but you know, that small team of developers worked very closely with the CIO and they knew what infrastructure it was going to be deployed in. They knew what storage systems and what network s systems and everything was in place. So moving from software development to deployment.

was almost automatic. That there was very little that was involved from between the teams because there was no friction because they knew each other and they had a a great process in place. And now building software, anybody can do it. I mean, you know folks could you know pick up a twenty dollar subscription and start building an application later that day. And but they would have

no awareness of what is the security controls for a company, you know, what is the data storage platforms, what APIs do they have access to? Who do they even talk to in the company to get access to those things, let alone when they want to move it into production, is it written in in a format that they can even support? So, you know, what was hard was writing software and what’s now hard is writing or making it move into production. And all of that is really shifting the day to day life and work of a CIO.

Shawn Cordner (29:29.644)
Yeah. I think you used a dog park analogy. Maybe give us that.

Bryan Smith (29:34.552)
Yeah, the well, if you think about you know taking your your dog out and you just let loose, you have no no idea exactly what they’re gonna do. But you know they’re inside the dog park, it’s a safe environment, it’s controlled you know for them. That’s like having a platform that’s the things that the security controls are in place, you know what things are accessible, versus if you just you know open the door and let your dog out in the woods, you have no idea what other animals out there, you know, what other things are are going on and

Or what other bad actors may be out there. And so having yeah, enough freedom so your dog so your dog can go have fun, that’s like the life of a a s citizen developer that has a proper platform to build on, versus right now the shadow AI is just letting all of your teams free to kind of do anything they want without those proper controls. And

Shawn Cordner (30:25.814)
dog park you have to clean up after your dog. So just like in AI, you have to clean up behind yourself. So make sure you’re not wasting money and letting things run that shouldn’t be running. But so let’s go back to you know kind of the temperature with boardrooms right now. So what percentage do you think are still in the give me that AI phase versus you know people that are actually taking a more methodological approach to this?

Bryan Smith (30:54.082)
Yeah, the we’re still s you know seeing, you know, probably twenty, twenty five percent of companies are getting a more mature approach on how they’re going through that. And those are the people that have spent a good amount of money. And that you know, you’re still in the sixty, sixty five on the trying to really get AI introduced into the business, understand what’s its potential, what the impact is, and then the remainder are still very much you know, this isn’t gonna be a thing. And the

still in that denial phase as far as the level of impact that it could have.

Shawn Cordner (31:28.95)
would you recommend people get from being in the sixty or sixty five to being part of the twenty five?

Bryan Smith (31:35.534)
Yeah. I do think that the the fastest path forward is is a couple layers. So one, I mentioned that the world of the CIO is changing. And so that means that they need to make room for working in AI. So they need to offload a lot of the day-to-day things that they’re doing right now. So one of the first things that they could do is just in the core operations of the business, the the keep the lights on things, leveraging a service provider, leveraging a partner to help do those components frees a lot of the time up.

You know, so now the thing that’s most unique to your business and the most value is the things that are your own internal knowledge about the business. So you can apply that to the go forward. And then the second phase is really putting a using a platform that has all the different layers that you may need. And that could be if you want to be able to self-host models, if you want to have the observability layer, if you want to have the security layer, those are all components that are just standard across anyone that’s leveraging AI.

So there’s not much value that you generate of picking those things yourselves or building your own. So if you can leverage a platform that has those capabilities, you’re gonna be in a much better place because now you’re focused on actually having impact for your business versus spending the first nine to eighteen months just figuring out the platform and you can move more much more quickly to actually getting output and outcomes for the business.

Shawn Cordner (33:01.432)
Do you think it’s important to show ROI on a pilot? Because I think that was one of the, for me, one of the fundamental problems that I had with that MIT study. Not that I read the study, but I read the articles about the study. And I couldn’t help but think pilots are experimentation, and maybe not all of them are meant to have ROI. So what are your what’s your position on you know spending money to learn as opposed to spending money to show ROI? And is it more important to just have a

path to ROI once you’re scaling that project if it succeeds.

Bryan Smith (33:36.482)
Well, I think that at the time AI was even more new when that article came out. So that applies some to the percentage. But if you look at the the rest of of the the data there, five percent was really when companies try to roll it themselves. The success rate was sixty percent when they actually leveraged a partner to do and somebody that had experience and had a a framework at platform on how to do it. So there was a huge gap as far as what the result was. So I think that the

Having a pilot, you need to define the purpose. If the pilot is to just create familiarity and understand the capabilities of the tooling stack, it doesn’t have to have an ROI because it is just bringing somebody up to speed, but it has to have a short time window. But we would do most pilots and design most pilots where the first project would actually pay for the entire platform. You know, so that very quickly you’ve got an ROI and you try to match a platform.

pro cost profile that the cost goes up as your consumption goes up, and you would expect that the value you’re getting from that consumption continues to go up as well. So you need to really keep those in alignment and always be looking at the return you’re getting from the agents or the usage from that. And be sure that you’re reclaiming either the cost savings that were predicted or the increase in productivity that was said to be possible because it’s one of the two levers.

Shawn Cordner (35:07.628)
Yeah, do you think that have that having a platform that you bought through a partner like Expedient is a more predictable cost than trying to build that infrastructure yourself?

Bryan Smith (35:25.344)
Absolutely. There’s gonna be multiple reasons there. You know, one, you know, we know what all the components are. so you’re not gonna have to rebuy technology because it doesn’t work together. So that’s gonna be a a very easy one. Two, as a service provider, almost everything we have is consumption based licensing so that you can buy it as you start scaling into it, versus if you look at different SaaS platforms like a ServiceNow.

You either have to turn it on for all of your employees or none of your employees. And so having a platform that you can license for the number of people that are really leveraging it helps align cost. But then also having speed speed is a little different. We’re kind of like a middleware between external AI platforms like an anthropic and an open AI and your data. So that you have controls in place, you have observ observability in place.

And you can pick different models for different reasons. So from a cost management perspective, you can use any of the different public models that are out there and you can understand the different costs of those. We have least cost model routing built in for similar to what you used to do for bandwidth capabilities in there. But then you can also do p private models so that you can say I want to have a fixed amount of expense for this amount of compute. And so you could say that, okay, I’m gonna get this many tokens out of this set of hardware.

But I’m gonna have a fixed cost. And so you can have controls on that. Or I’d like to move from being an operating expense to a capital expense. So if somebody could buy their own GPUs co-located inside one of our data centers and then plug it into the management stack of the AI control platform, leverage it together. So it gives them the ability to choose OpEx versus CapEx. They can cap off performance, or by using the governance layer, they can still cap performance or cap cost.

on any of the different models and layers in there as as well. So you can know exactly what your maximum spend would be and have visibility of what’s going on.

Shawn Cordner (37:29.262)
I’m gonna talk about the routing for a second, but this is just my ignorance on the subject. The CapEx versus OpEx, throughout my career, the conversation has always been trying to shift from CapEx to OpEx. It seemed like CapEx was always where the constraint was and OpEx was a little easier to to manage. are we going the other direction for any reason right now?

Bryan Smith (37:50.326)
It really varies by industry. Yeah, so that for example, people that are owned by a private equity company, they want to have the highest EBITDA possible, you know, so they don’t like OpEx. So they would rather spend CapEx. Or you have certain industries such as utilities that prefer or manufacturing that are used to often investing in a capex over time. So there’s certain areas that they prefer CapEx, other people prefer the OpEx.

Shawn Cordner (38:15.832)
Side. Got it. Thanks for telling me. so the reason that we’re having this conversation to begin with is because of the the least cost routing model. I I come from the telecom side of the business and least cost routing in long distance and international, especially, was a big thing back in the you know in the 90s when I started. And I was having a conversation with someone else in the industry. We were at an event and I ran into her and I was like, I’ve why don’t we do least cost routing for these models? And I thought I had the

the next billion dollar business idea. Turns out that it someone’s already beat me to it and you’re one of the people that have beat me to it. So talk to me a little bit more about the y you know, how it works and also what’s the utility of it and how does it impact budget.

Bryan Smith (39:01.26)
Yeah, so built into the ag control platform is our model router and it works has multiple different layers in it. So first it’s gonna look at the question that you’re asking and determine somewhat the complexity and also the subject. So is this something we’re trying to generate an image? Is this something that we need to do a lot of deep research, or is this asking what the weather’s gonna be tomorrow? And it’s just a web search. And based on those different criteria, you know, the model router understands.

the the complexity, i the availability, and also the underlying cost for the token sets for different types of models. So it can leverage all those different components to come back and either give you a single answer. You can ask at have multiple answers that then you’re voting on to give it feedback over time. or you can also do model routing that includes things like role based access control. So you could say my marketing team

can have access to a model that’s great at image generation and content creation, creative writing. And but my developers are going to be buying a model that is a private hosted model that just crushes it at s soft at software coding, but isn’t great at generating images, but is a tenth of the cost. So you know the model router is both from helping select the the the correct one

But then also giving you the controls and stuff to who has access to what models based on their function.

Shawn Cordner (40:34.87)
I is that kind of pre built? You’ve already built most of that logic and it just has to be fine tuned or is it something that has to be trained for each use case?

Bryan Smith (40:45.864)
Yeah, it’s something that that we’ve pre-built and for your numer normal chat type of user or someone’s interacting through a user interface, a desktop application, or a chat interface, a web interface, it’s gonna automatically, if you leave it in the default mode, it will pick the different model, it tells you which one is picking, and then it’ll deliver the response. You can also override anytime you’d like and pick a model of your choosing, and but we can also provide reporting back on that to say.

Shawn is always overriding and using the most expensive model versus his peers are saving 80% token usage because they’re using the model router. So you can also use it from a employee enablement and training perspective, and you could set up those type of automated notifications of hey, FYI, and that’s part of the observability and how it ties together because we could show you how many tokens or or credits you’re allocated, if that’s how your company would like to do it.

And as you’re using a lot, we can also give you tool tips and stuff on how you could re reduce that usage. Now if you flip over to the agent side, that’s more when like when we were talking about how you manage people that inside the AI control platform, an agent run can be replayed like a DVR. And you can actually go step by step and see every single decision that was made and what the model returned. And then you could actually take that that entire flow.

And put it back to one of the AI agents to say, investigate this, look at the model choices, look at the output. W how could I optimize this to reduce my total token consumption? And it’ll give you incredible feedback to say, well, you’re using a OPUS model from Anthropic, where you could be using a haiku, which is multiple tiers less and would be a thirtieth of the expense for this stuff, but you still want to use Opus for this stuff. So

That’s where you can really get granular on the management of that and you can have that type of agent run even automatically in certain cycles as new models get released and new capabilities come out so that it’s constantly looking at that and helping you optimize your agents along the way.

Shawn Cordner (43:03.31)
Do you typically find people routing between established models or are you often seeing, you know, people building their own models and working that into the workflow? And if so, how does that get onboarded for lack of a better term?

Bryan Smith (43:18.296)
I would say that’s probably one of the fastest trends we’re seeing is people using a combination of public models and I wouldn’t say building their own. I would just say running privately hosted models. So an open source model that they can run on our infrastructure or you know infrastructure that they bring. And they’re often smaller models for specific use cases and that could be from a transcription of phone calls or that could be from a coding perspective.

Or that could be mid-level work that would be equivalent to like a a Chat GPT five, you know, where I’m just looking to summarize this text or look at this information, but you’re not using really complex reasoning. Those are things where using local models can be very cost effective. And some of the things that people don’t often realize is part of the benefit of using a service provider is also the same reason that cloud exists.

Because we can take a private model or a local hosted model and actually multi-tenant it. so that you could if if you know five people wanted access to the newest version of Quinn, then they could all access that and none of their data is being combined. It’s just a compute process that that’s you know being used. Or you can set up different batch processing and and and different capabilities versus if you’re buying your own hardware, you generally have to run it at north of 70% utilization.

to have a return on investment compared to using a external public model.

Shawn Cordner (44:51.402)
What is adoption like on this type of solution so far?

Bryan Smith (44:56.054)
It’s it’s really building and you’re seeing that push into breaking things apart as people are getting higher consumption on things that are moving from pilot to production. So I was just at the AI four conference a couple of weeks ago and it was probably the biggest topic was related to cost management and the biggest driver was how you leverage the private models or self hosted models.

in partnership with a a public model to get that cost management. And that’s really where people are understanding that this isn’t just an AI conversation. It there’s an AI conversation, there’s an infrastructure conversation, and they really have to work together to get the performance, the cost, and that the total capability.

Shawn Cordner (45:44.27)
And is it able to break out a prompt into multiple pieces and route it and then re-aggregate it? Because I’m I’m thinking like if I were using it to create a a marketing, let’s say I was using it for a blog, right? We use we use it for a blog, we use it very intentionally, we use it as a tool with a human, right? So ideas come from humans exclusively, not from AI. And then the human writes the outline.

Gives it to Claude. We use Claude. Claude does the SEO and AEO research, and then it assists in doing a draft, and then there’s heavy editing back and forth. It’s a very kind of collaborative process. And then right now we hand that off to a designer. The designer creates the art and then that gets passed off to a digital marketing person. That person then posts it on, you know, puts it in WordPress and posts it and

Does the social media promotion, builds the email and HubSpot, and sends it. So that could potentially be one agentic workflow. That could just be one prompt. And each one of those phases actually probably has multiple pieces to it where you know some things require a a more sophisticated model where some things are very, very simple. So is it as easy as just giving it that single prompt, assuming I’ve got that workflow built and it’s able to split out

not just the different steps, but also, you know, the the different needs within each step.

Bryan Smith (47:16.118)
Yeah, the the way I would probably build that as like a small application. And so the application would have a user interface that it could be you interacting with or any of your team where and I would probably do it via voice where I have a microphone where I just describe what I want. And then I would have that application drive the agentec workflow that would go do all those different steps that it’s gonna do the external research, it’s gonna bring it together.

I would have it run probably against you know two or three different models to have and have one that’s an adversarial agent that is disagreeing with things, looking for you know things that that are like AI slop so it doesn’t sound AI written. And then once I have that, I would probably do a human in the middle step where it’s gonna pop to you, have you look at it, give feedback, and then once you approve it, then it’s gonna dispatch a graphics agent to generate art that go with it.

it’s gonna understand your design templates, you know, your brand book and apply it all that together. So you would take that process likely from a few minutes of initial input, the feedback for along the way, and then a a final edit so that it’s something that you could have an automated pipeline. And once you get something that you’re doing very little edits, you get comfortable with taking yourself out of the middle. and then it can just automatically

be something where you have an idea, you can even do it dispatch it off your phone of what you’re trying to do and then it’s it’s automatically writing the post, generating the graphics and implementing it on your behalf. So definitely.

Shawn Cordner (48:52.536)
Can you build that by Tuesday for me? we’ve got some posts we gotta get going here.

Bryan Smith (48:57.262)
Well, that that’s that’s the whole point of using a platform is is the answer is yes. Once you the whole beauty of a platform, I know you were joking, but the the beauty of the of a platform and what it does for a citizen developer and why it gives the controls to IT is you’re putting those enterprise level controls to say, okay, you have access to Salesforce, you have access to HubSpot, you have access to do web research, you have access to use Hicksfield for doing image generation or

a open AI GPT model for doing image generation. And so they can pre-configure those things. They can limit how much you could spend, you know, what your those different controls, and they could say you have access to connect to do LinkedIn posts, but nobody else in your business does. And that’s the big difference of getting the controls on the front end of how you access to different tools and different pieces of data.

Now you as the builder can actually go super fast. And once that’s configured, yes, you could absolutely build that entire workflow and application by tomorrow.

Shawn Cordner (50:00.78)
Wow, that’s crazy. for free, right? You’re gonna do that for free for us?

Bryan Smith (50:07.047)
We can trade.

Shawn Cordner (50:09.379)
I’ll take the trade. all right. So anything else on the least cost model routing that we should touch on here?

Bryan Smith (50:18.53)
Maybe on the model routing, we we kinda skipped over just the huge difference in variance in model costs. And people don’t think about that. Do you want to hit that here or do you wanna so the way a good way to think about this, like if I have a a sore throat, I’m not going to the best surgeon in the in the area to take my temperature and and tell me I have a cold. And but yet people default to the most expensive model.

And all of the common AI platforms default to the most expensive model too, because they’re going to give you the best user experience. You’re going to be the happiest with what they give out the other side. But it’s also going to be the most expensive. And so it’s really on the user to dial back as as many layers as they can while they’re still getting the output that’s acceptable. And you can have a massive reduction in cost.

And these different models are changing prices weekly. So if you don’t if you get locked into a certain technology stack, I’m all anthropic, I’m all open AI, I’m all groc, whatever it may be, and that then you’re really limiting your ability to actually do cost management and also technology feature management. So that’s the other benefit of having something that’s a lot more open across that because you can truly control costs

You can choose to move things onto private models. You can choose to move things into CapEx. So the the the level of control just changes dramatically.

Shawn Cordner (51:56.93)
I hear people talking about not getting you know locked into a a vendor and that vendor lock in is a a huge problem that people need to think about going forward for that exact reason, right? What how does vendor lock in actually occur and how do you avoid it?

Bryan Smith (52:13.592)
I would warn for ecosystem lock in most. Yeah, so that because at some point you’re gonna have to pick a vendor. And and so it’s just a matter of is this vendor married to a single ecosystem? You know, a or can they give you e everything? And it’s similar to food groups. Like you could pick or pick in a specific restaurant, you can only eat that versus a large food distributor or a grocery chain that you could get things from. So the

That would be my closest, I think just off the top of the head analogy. So if you get married to a certain ecosystem, you only have access to what they release. And you’re completely married to their pricing. So that the biggest challenges are change of your user interface, change of your application API, or change of your workflow. And all of those are really disruptive to your business. If you have to retrain your employees how to use an application.

If you have to rewrite your application to use a different API schema, or if you need to change how your workflow functions, all of those are really challenging. Excuse me. But if you have something that is a lot more open that can access all of the different models, then you can and it acts that translation layer, it really opens up your ability to get the best technology available at the best price point.

Shawn Cordner (53:33.442)
What role does open weight models have in this and you know, can can the platform or the gateway provide some protections? you know, if it’s a open weight Chinese model, for example?

Bryan Smith (53:46.542)
So it would be something that you’re running on your infrastructure. So like any model that we make available through our model router has gone you know through our compliance department, it’s gone through the checks to make sure that it’s not siphoning off data. people have the still have the ability to bring their own model. And so it’s more that you can choose to run your own copy, your own and put the walled garden around it so that it doesn’t even have access to the internet if that’s what you chose.

Yeah, so that it can only run and work inside your environment. but you still can have biases from the model. So often it’s less about exfiltrating your data, but it’s more is it shaping the answer differently based on potentially the views, the norms, how thing how business may be done from a different area. So you could get a very different outcome.

Shawn Cordner (54:40.544)
Interesting. How do you protect against that?

Bryan Smith (54:43.412)
It’s somewhat it’s the same way you choose where do you go on vacation, where do you feel safe. And the the way you protect is just making choice in areas that would be good for you and your family. And so it’s a similar of picking models that you would trust and model providers that you would trust for your business in your industry.

Shawn Cordner (55:02.318)
Yeah. So at the very beginning of this, you talked about AI as infrastructure and how it’s actually one decision making process. And I feel like that’s this is a good transition point to talk about that because there are so many decisions that we just talked about and so many considerations for how you architect your your infrastructure. and there’s a lot of moving parts to it. So

What do you mean by AI and infrastructure are one decision and are people treating it that way?

Bryan Smith (55:38.636)
I think that’s probably one of the most important things that people can really watch out for. And so there there’s three huge things that are driving change right now inside of IT. You have folks that are looking at at the cloud rebalancing, which breaks into what do I do with my own internal stack like a VMw stack versus what is my public cloud spend? And so those two things come into cloud rebalancing. And then the other one is what does AI mean for my business?

But what they don’t realize is every one of th those decisions essentially pre-make your other decisions. Because for example, if you say, I’m gonna use all hyperscale cloud, I’m gonna do everything there, and I’m gonna use all public models, that’s gonna drive a lot of your decisions and what you know what you’re locked into. Or if you said, you know what, I’m gonna refresh all of my hardware and buy all my hardware and I’m gonna keep it on premises. Now when you want to use AI, you have

questions about latency, are you gonna have to pay for data twice versus that’s where you’re seeing this big resurgence in in private cloud because people are seeing that when you’re doing AI, having you know one set of data versus keeping two copies, or having the ability to have your data directly adjacent where AI is being processed, or paying for your security layer once versus paying for it multiple different times, we’re not having to pay for data transfer.

You know, so those different components are really coming into play. And because what we talked about before of writing software become is becoming so much easier and so much cheaper, you know, people are going from often SaaS was one of the biggest competitors for a cloud provider. And now people are writing their own software again. So you’re seeing this big swing back to the private cloud. So having all those things run together and then

As people are writing more software, they need a place to run it and host it. So if you can have a platform that helps them write the software, but gives them like a simple publish button that automatically puts it into production in a secure environment, in something that the the company has already approved, leveraging storage platforms that they already have in place. Now you’re just scaling the resources that they have versus having creating more technical debt that they’re gonna have to solve down the road.

Bryan Smith (57:59.244)
So that’s really when they start thinking of how those different things work together, it’s really taking the things you build in AI and how are you going to scale those and how are you going to have that on a reliable infrastructure that has DR, has backups, has all those other protections, just like any other production system you run in your business.

Shawn Cordner (58:19.948)
I feel like the answer to this question is going to be this is why you need to talk to Expedient or Amplix because we can, you know, kinda help find the right provider like yourselves, depending on your use case. But what would you say to someone that’s starting this process now and this all sounds very daunting to get right the first time? Is there a minimum effective dose to this? And then what if you’re somebody that is a little further down the road and

maybe didn’t think of all these things the first time. How and when should they try to escape the technical debt that they’re inevitably building and what are the actions that you think they should take to to do it?

Bryan Smith (59:01.11)
Yeah, yeah, I think that both are are very similar. You know, so it’s definitely, you know, why you engage an expert. It’s you know, to your s question earlier on the MI MIT study of at that time going from five percent to a sixty percent success rate, huge difference, a twelve X increase in your outcome. And that’s continued to get better as people have got like ourselves have gotten more expertise on that. And expedient as an example, we’re not

creating and writing our own technologies. We’re a curator of technology so that we’ve taken the different components that we know work together, that we know function, but they’re done in a modular fashion so that if something’s better out there, we have the ability to plug it in and take something else. Or for people that have already implemented their own things and they like certain pieces but don’t like others, and having something that’s very modular so that you may want to bring your observability platform or your storage platform and have it work with the the rest of the

AI control platform or others that that have similar type of offerings, those things really matter. And and Amflix can help provide the guide to say, here’s the type of partner, here’s where you are, but it’s very much about how can we meet you where you are in your cycle and get you the most value as fast as possible. So that’s like when we’ve looked at the platform, I think that we’ve somewhat underpriced it compared to most of the market because we’re an infrastructure company that also does AI. So we’re not doing

You know, we don’t have to s do all this margin stacking. Whereas if you were buying something that’s software as a service only, they’re not making anything on the co-location, they’re not making anything on the infrastructure that runs it, all these other components, because you still have to buy the the all those other things. So having that vertical integration can also be incredibly cost effective. So so those are just some of those different things that you want to think about. And probably the last is just the time to value is the biggest thing.

Because you know, working with a partner that understands this, that has these things together, that you’re not spending the first twelve to eighteen months just getting your your house in order before you can actually do something. It’s more that you’re moving into a fully furnished condo.

Shawn Cordner (01:01:09.004)
Yeah, that makes sense. You had mentioned VMware earlier. So for those that that have not been following along the VMware Broadcom saga, maybe give us a crash course and then let’s talk about this cliff that’s coming up in October twenty twenty seven.

Bryan Smith (01:01:27.778)
Yeah, th it’s interesting that I don’t think people realize that VMware could be one of the biggest blockers for some AI adoption if if people aren’t paying attention to what’s already happening. For others, it could be one of their biggest enablement. so in you know anyone that’s not aware, you know, VMware is a hypervisor that you know hosts is a way that you can multi-tenant your environment to get to build your own private cloud environments. And

Yeah, VMware was the dominant market leader with you know roughly ninety percent market share. They were purchased by Broadcom a few years ago and there was a big change in the how the licensing was done is what most people think about when you talk about the VMware saga. But the big change that’s happening is what Broadcom’s vision was was taking all of these different VMware softwares it that had different SKUs and different capabilities and making them one unified stack. That is, you know,

called VCF, so VMware Cloud Foundation. And so that stack becomes the default as of October of 2027. And at that time, that’s when the current vSphere eight environments become really end of support. So what that means is anyone that’s using vSphere eight and isn’t onto VCF, they’re not able to be compliant with things like SOC or PCI or HIPAA. And

It’s not just an in-place upgrade like what vSphere has been in the past. It’s actually a re-implementation, a re-architecture of requires more hardware than what they had in the past for the management plane. So there’s a lot of different moving pieces. And when you marry that with the amount of time it takes to do the migration, to do the planning, to buy the hardware current lead times. I mean, if people aren’t actively planning that project right now, it could be the thing when I said that it could be a their blocker of AI, it could consume all the time.

in their infrastructure teams and their IT teams next year if they haven’t already done proper planning.

Shawn Cordner (01:03:29.122)
Are you finding peop most people are on top of this or a little behind?

Bryan Smith (01:03:32.94)
I’d say most are very behind. You know, that it’s something that’s been talked about for a while. And you know, we’ve been having the conversation for multiple different year and multiple years, but just in the last quarter, we really saw this hockey stick adoption transition where people they may have tried different technologies, seen what else was out there, and we’re still seeing about sixty-five percent of people stay on a VMware platform, about thirty-five percent look at other different technologies like a Nutanix.

And but now they’re having to make moves, you know, and they’re trying to look at hardware and hardware costs have gone up, like I said, four hundred, five hundred percent in the last, you know, twelve to eighteen months. So they’re getting sticker shocked that they didn’t expect. So that’s where a service provider also that pre purchased gear in advance is if people are acting quickly, they can actually still get access to hardware that is significantly less than if they were trying to deploy and buy their own right now.

Shawn Cordner (01:04:32.45)
So we’ve got thirteen months ish, give or take. So what would the timeline look like if they, you know, if someone was just starting from scratch right now? How long should they take to look at their options? How long, you know, when should they make a decision? How long is it gonna take to implement?

Bryan Smith (01:04:53.998)
Yeah, so if you if you work backwards, if somebody was in the do-it-yourself mode, that their migration is probably gonna be two to three months for their environment. they’re gonna their lead time for hardware is roughly six months for a lot of hardware right now. And there a lot of existing hardware that people have isn’t doesn’t support VCF.

Yeah, so they don’t realize they’re in an unhealthy state because they’re making assumptions. We already have hardware, so we’re good. And so they’ve not done that type of check to ensure that they have that availability. So you know you now you’re already at nine months, then you have to get it scoped out, you have to do the engineering, and you’re probably gonna look at more than one vendor and s and there’s gonna be a couple months of time that goes into that as well. So you’re roughly eleven months, which is September.

As in a week and a half.

Shawn Cordner (01:05:52.61)
Yeah. w we can condense that time frame a little bit if they work with us, but sounds like people need to get started yesterday and best time to plant a tree was ten years ago and the second best time is today. So what’s your message to people to get up and take action today?

Bryan Smith (01:06:10.518)
Yeah, the the biggest is don’t wait at all because one, you know, working with people that may have access to hardware immediately. Yeah, you can buy down your decision. There’s a very defined migration path that most of our migrations are done within from ninety days from when somebody says go. So that you can really get deployed very quickly and it just de risks the entire thing because like for example, we did

about four hundred and fifty migrations in the last twelve months, twelve to sixteen months. in versus you think of how many times a company does their own data migration. So you know move quickly and don’t wait till the very end because there’s also gonna be a very there’s gonna be a lack of talent that is available at the very end to help.

Shawn Cordner (01:07:00.054)
That makes sense. Would you say that first order of operations is to make sure that the equipment that you have can support VCF?

Bryan Smith (01:07:08.6)
Yeah, absolutely. Doing like a VMware health check so that you can ensure that it supports VCF and then how long because there’s kind of two different phases. There’s one processor line that they’re just there’s no support at all, and there’s another one that will only have a year, year and a half of support on it. So people would have to be planning an upgrade pretty quickly. so they would end up doing two upgrades. They’d have to do one where they’re doing a migration to the

a current VCF and then they’re gonna have to do a second migration onto a new hardware stack. Yeah, if they’re trying to do it all themselves.

Shawn Cordner (01:07:41.26)
And can you dig in a little more about the compliance risk?

Bryan Smith (01:07:44.386)
Yeah, so the compliance piece comes from there’s a requirement on this all compliance that has to be supported software. So once the vSphere environment becomes into support, you can no longer be compliant with any of the core things that are out there. And there’s a lot of risk to the business because th also a lot of cyber insurance requires that you are using supported software. Wow. Or you know, you could get fines of

over $100,000 a day in certain compliance groups if if you’re out of compliance and a breach happens. So you’re essentially t end up taking on all the different risk. And a number of clients also have commitments to their clients that they have a SOC their SOC compliant or their PCI compliant. And they would then have situations where they’re in breach of their client contracts because they did not maintain the compliance. And it’s not something that IT

and the compliance team are really having those conversations. So I think it’s sneak up on a lot of people.

Shawn Cordner (01:08:46.85)
So technology leaders and compliance leaders in heavily regulated industries are aware of this, but they’re not collaborating or they’re generally not aware of this, or what is the actual breakdown?

Bryan Smith (01:09:02.69)
Yeah. I think that it’s they’re aware, but they’re not aware of you know that this isn’t a normal in place upgrade. So they’ve had you know this cycle many, many times before going to V Sphere seventy, V Sphere eight, no big deal. And but it is because it’s a completely new stack that they have to re implement the time to do the implementation, even on their existing hardware, is gonna take a lot longer. They’re learning a new stack that that takes

takes a while and there there could be new security holes. And then they’re also used to in the the previous at buying hardware, availability and price points were very different. So you know when you go to ask for budget and it’s 4x what the expectation was, that also can slow the process down because now the CFO’s asking you to go back and check things because this can’t be right, which is slowing down your entire buying process as well. So it’s more

That the world is just different and this upgrade cycle is different than what it’s been in the past.

Shawn Cordner (01:10:05.57)
Probably worth flagging too that you know we’re going into budget season for most businesses. So this should be probably at the top of the list when you start to sit down and think about what your budget looks like for twenty seven.

Bryan Smith (01:10:17.386)
Absolutely. Because if you’re not having this conversation, one, it can really throw off your entire budget or could greatly consume budget that you had planned for other things that were gonna be meaningful, like AI. So and now you start to have cascading impact in kicking further projects down the road.

Shawn Cordner (01:10:36.174)
All right, any other words of wisdom for the technology leaders that are listening here, whether it’s about AI cost governance or whether it’s about VMware?

Bryan Smith (01:10:47.884)
I’d say the biggest thing is you really have to just move into a continuous learner mode. And there’s so much change happening so fast. Being that continuous learner, making that just part of who you are and your personality. And we’re also seeing a shift in the community and how much people are willing to share information with their peers, working with partners to have those conversations, yeah, it’s can be some of the highest value.

for sure. And then second, I would say you just have to start that you know people sometimes get a little bit of analysis, paralysis with this. And you know you just have to dive in and start learning and start doing and working with your peers overall. And pick a platform that isn’t gonna lock in all of your future decisions that gives you that optionality so that you can continue to make adjustments for what fits your business best. And

I would encourage you to try to have the least amount of vendors you’re interacting with that while also giving you kind of choice across technology platforms.

Shawn Cordner (01:11:57.87)
All right, well you’ve been very generous with your time. You’re a busy guy, I’m sure. So thank you. I appreciate that. Where can people find you if they wanted to catch up with you on this subject?

Bryan Smith (01:12:08.482)
that always on LinkedIn is one of the easiest places, happy there or my contact information is directly on the expedient dot com website as well. people can get a number of different reach outs and look forward to interacting and and learning from you guys as well.

Shawn Cordner (01:12:26.19)
All right, Bryan Smith, thank you for your time and expertise.

Bryan Smith (01:12:29.198)
Thanks, Shawn. Really appreciate it.