Ep.219 – The $14 Billion Tell & How the AI Giants Just Validated the Advisor

Josh Lupresto (00:00)
Here’s a number that I want you to sit with for a second. 95%. MIT ran a study on 300 real enterprise AI projects. 95% of them produced little or no measurable impact on the bottom line. So it’s not the demo. The demo looked great. The deployment was the issue. That’s the part where the model really

Has to exist live inside of a customer’s environment, in a real company, next to legacy systems, dirty data, compliance rules, and all the humans that didn’t ask for this. the model worked, but the deployments didn’t. So why does that matter to you? And why am I doing this episode solo? It’s it’s a little bit hot off the press because the two biggest AI labs on the planet just looked at that 95% number, and instead of building a better model.

They did something that should make every single advisor, agent, MSP VAR sit straight up in your chair. They got interested in your business. And they raised billions of dollars in private equity money to do it. So let’s get into it.

Welcome back to Next Level BizTech. I’m your host, Josh Lupresto SVP of Sales Engineering at Telarus And as you see, I’m flying solo today because this news cycle moved a little bit faster than some of my guests’ calendars did. So honestly, the story is just too good to wait. you know, today we’re not talking about orbital compute, we’re not talking about SpaceX IPOs, the looming anthropic and open AI IPOs. Today it’s just about AI.

And models. I think there’s some great stuff that we want to chat on. So let’s set the table here on the models first, because you can’t understand the rest of this without it. So we are sitting in what I think is no exaggeration, the most concentrated stretch of model releases in history in this industry. Because in the span of basically what, a a a month, you’ve got open AI shipping out GPT 5.5 and then 5.6 coming out.

Google rolls out Gemini 3.5 Pro, Anthropic drops Claude, Opus 4.8. And not even talk about what happened with Fable and you know these regulatory captures or whatever you want to think about why this got blocked and what the government’s thing about Fable getting pulled was. But that doesn’t even also count in before you get into these open weight models, right? You got north of 500 models your customers can theoretically build on. So there’s a lot happening in this compressed news cycle.

So here’s the thing that I need you to internalize because this is the whole game. The model is no longer the hard part. Say it again. The model is no longer the hard part. So for a decade, think about this. You know, you had these smartest engineers on the planet. They wanted to work on the model. That’s where the status was, that’s where the money was. That’s where the frontier was. We talk about these frontier models.

So that’s something in the last 12 to 18 months. The frontier was here and then it moved. the benchmarks, you know, they all keep climbing, the capabilities, it’s all there. These models can pass as graduate level exams. The problem is for the vast majority of businesses, the use cases, it it looked like they solved them. So if the model is solved, where did all the difficulty go? Well, I mean, it it didn’t disappear.

It just it just moved a little bit. The cheese moved. So it where did it move? it it moved into the integration layer. So it went from, you know, which of these 40 models do I actually route load route this all these workloads to? It moved into all the rest of it. It moved into the evals, it moved into the test harness that catches the hallucinations before they ship. It it moved into rag, it moved into the data is a mess, Mr. Customer. I’m sorry.

Into well, you know, let’s chain five tools together. And if any one of them breaks, then the whole thing fails. But I mean, it also moved into compliance, it moved into security, and then it moved into wait a minute, wait a minute. Can this thing actually touch our customer records? Like, have we really thought through this? So, so where’s the kicker? the kicker is for anyone running in production. So it think about this. If you’ve got three frontier models with all three different vendors, they’re landing in, you know, new stuff landing in the same 30 days. Single vendor lock-in has never been more expensive than it is right now. you know, Bezos gave this example. It was it was about innovations are about a door. What door can you go through that closes and you cannot open to go back through, right? That’s a very thoughtful innovation. You’ve got to think through it versus other doors. we can test, we can try. So this idea of single vendor lock in has never been more expensive, I think, than it is right now. So what’s the move? in my mind, the move is multiprovider. the move is a router, the move is pinning your versions so OpenAI doesn’t silently upgrade you under the hood and break your customers’ workflow on a Tuesday. We all remember that anybody that’s kind of supported IT and tech, you know, the Windows, what is it? Windows Tuesday updates. You’d come back in and all those things, patch Tuesday.

That would you know break the machines and we’d have to patch them and fix them. So I I want you to hear a little bit about what I just just kind of talked about. This isn’t a product problem anymore. This is an architecture problem. This is a strategy problem. This is a who do I trust to help me navigate these 40 AI models and these vendors to help me, the customer, not get fired problem. So what is this? This is a services problem. So let’s talk a bit more about what actually happened in May, because that I think is is probably the part here that reframes everything a little bit better. So within a matter of days of each other, you got OpenAI and Anthropic, the two frontrunners, both stood up dedicated businesses whose entire reason for existing is to put engineers inside of customer organizations to make the AI actually work.

It’s not about selling the model. Everybody’s got tokens. They’re all gonna sell tokens. It’s deploy it, it’s babysit it, it’s integrate it, it’s make sure that it really survives contact with the reality, right? The of the reality of what these customer environments look like. so let’s think about OpenAI for a second. OpenAI calls theirs, calls their deployment company deploy co for short. How creative, right?

So so so look at that structure, because the structure kind of tells you everything. You got to follow the money a little bit. So they launched a $10 billion pre-money valuation. They raised $4 billion into it. and OpenAI kept majority control. But ask yourself, who are the investors? it was n for OpenAI, it was 19 global partners. And that’s the part that maybe, you know, put my put my coffee, my Red Bull down here for just a minute. it’s three of the biggest

management consultancies on Earth. Mackenzie, Bain, and Cap Gemini. Those three are investors in it. So think about what does that mean? Mackenzie buying equity in OpenAI’s consulting arm. So the consultants are paying to get in the room. And that is the room that you are already in. So let’s go to Anthropic then for a second. Anthropic ran the same play with just a little bit of a different jersey. It was a one and a half billion dollar

Joint venture with Blackstone and Goldman to just the same thing, embed their engineers in their financial services customers. So by that way, you think about it, Goldman’s kind of backing both sides a little bit. So the smart money is it’s not about picking a model, it’s about betting on the deployment layer, regardless of who wins the model race. I think we’ve all seen from the news cycles that it’s just hard to keep up. There’s gonna be a lot of models. So time for the magic word.

The magic word behind all of this is FDE, forward deployed engineers. So you know, it’s not a it’s not a new idea per se, right? You’ve seen that in the CX environments, you’ve seen that in other environments. you know, Palantir kind of invented this years ago. The the the forward deployment engineer is not somebody that sits back at HQ just lobbing some software over the wall. They go sit at a customer, they learn the customer’s workflow.

their data, their actual mess, and then they kind of bend the technology to fit it. They’re, I guess you could think of them as half engineer, half consultant, half therapist. we’re like full time therapists sometimes. So it think about that from a jobs perspective. I think it’s one of the hottest jobs out there in AI right now. These these ransom numbers on this, the jobs

On the the postings for FDEs are up like 800% year over year. Comp is insane. It’s you know, mid-level, it’s 380,000. Principals at these frontier labs, though, you know, somewhere between 600 and even up to seven figures. You believe that? Seven figures for the principal engineers. So the labs are throwing money at this, like it’s you know, the the model team they’ve all racing to build in 2023.

So if you step back for a second and you think about why? Why would OpenAI, a company whose entire, you know, ethos is, you know, we build the God tier models, why would they turn around and raise four billion dollars of private equity to build a consulting business and staff it with people who fly out and just sit at the customer site?

Because they figured out the same thing that I think the MIT study figured out, which is the model is not the product. The outcome is the product. We’ve been talking about that for a while on this podcast. It’s the business outcomes, business outcomes, business outcomes. It’s less about, you know, the techie, the bits, the bytes. It’s and it’s so hard for me and and and and others as engineers to not want to go into those bits and bytes, but your customers trying to solve a business problem.

And so in this case, the outcome is the product when it’s you know, especially as we’re talking about AI, it’s so outcome driven. And the gap, I think, between the model and the outcome, it’s filled with humans. no surprise. The private equity money is the tell that you know, PE PE doesn’t chase the the technology, right? PE chases durable cash flow, stickiness, and and just longevity.

So if you think about you think about the structure of these deals, this bakes in a massive guaranteed return. You’ve got the smartest capital allocators in the world looking at this landscape and deciding, okay, the bet is not the model. It’s the deployment. And that’s that’s not a side note. That kind of is the story. All right. So we’ve laid the groundwork a little bit.

And I i i i i if that gives you a little bit of color, right? The last couple of months have been a little wild. And so what’s great about this? The great about the the great part about this in my mind is that this is such validation for the channel. It’s crazy. So let’s let’s connect you all the way back, right? If you’re listening to this and you sell technology for a living, what I described should get you very excited and feel like a tailwind, not I

Don’t look at this as a threat. And let me kind of explain why, right? It’s it’s it’s an abundance mindset. If you think about though, you know, for for years there’s been this nagging fear of like, it’s commoditization, you know, the technology is gonna get so good they’re just gonna push a button. and the customer doesn’t need a guide. They just swipe a credit card. We know we heard it with what? We heard it with cloud, we heard it with SaaS.

And now we’re hearing it with AI. The model’s so smart. It’s just gonna do the designing all by itself. who yeah who who needs these, you know, engineers and and all of these things. So look, here’s the reality. The reality is the news just proved it in dollars. Again, you it’s the 13, 14 billion dollar tail. The technology got better, and I think the need for a guide went up and not down.

Up is the critical part here. think about for a second, think about the access layer, right? this is where a lot of us started out, so where we cut our teeth. Connectivity got faster, it got cheaper. and and did it disappear? No, it did not, right? How how long and are we still doing all of this connectivity? Did you know, and and and the advisor.

Constantly plays a more and more and more important part of that. you know, with all those changes over twenty years, what did the advisor do? What did you do? You kept selling into that stack, but you also moved up the stack. You started selling what? SD WAN and SASE and then, you know, kind of this whole security posture, and then just it keeps going and going and going. just cloud, it just never ends. I think at some point then.

The value migrated to orchestrating and and the trust layer. The trust layer has never gone away. Is is all the kind of ups and downs that the channel has had over time, suppliers coming and going and building and investing. What’s stayed constant? The fact that it stayed constant is relationships matter and people buy from people they like and people they trust. So AI is just, you know, this evolution of AI, I think, is just

The exact same thing. It’s just been compressed into 18 months instead of, you know, 20 years, right? Think about this content that we’ve been talking about. It’s just, it’s moved so quick. So what’s the model layer, right? The model layer is just this kind of raw access layer to the intelligence. It’s it’s commoditizing in front of our eyes. 40 vendors, 500 models, you know, tokens.

Finally dropping in a price that’s reasonable. and the value is migrating up the stack to the same two places it always migrates to. It’s integration and it’s trust. So what’s the great part about this that I think these frontier and and and you know these deployed labs cannot do? The part that I believe and have always believed is is that we’ve believed is structurally yours. So

OpenAI and Anthropic here are pointing their FDEs at the Fortune 500, JP Morgan, all of those guys, the giant financial institutions, they are are are big enough accounts to justify million-dollar engineers flying out and sitting on site for six months. That’s crazy. The the maybe the math works at the very top of the market, that top 1%.

But as as as I’ve learned, you know, scaling a business is hard and there’s a lot that comes with that. that math does not work for the mid market. It does not work for the remaining ninety-nine percent. It doesn’t work for the regional manufacturer, the multi-site healthcare. you know, that’s the all of these things, the the mid market, the SMB, that is the bread and butter of this channel, right? It’s ninety-nine percent of the market that’s out there. So I don’t

I don’t think the labs even understand those accounts and never would consider putting an embedded engineer in those accounts, but the the the the theory is, the thesis is those customers all still need our help. And they can’t do it with these these these labs that we’re talking about. The economics just don’t support it.

So what do we do then for how do we deploy and keep deploying AI like we have been? you all have done some amazing things this past year for anybody that isn’t a Fortune one hundred, right? Who’s the deflor the the the the forward deployed engineer for the mid market? Great news. You are the advisor, the MSP, the the VAR, right? all partners are.

Bat forward deployed engineer in my mind, right? Of course, with our team to help you in that. Like you you’ve got that trusted relationship. you already know the customer’s environment. And you, I I’ve seen it time and time again, you already get the phone call every time. Who do I go to for this? Who’s a good fit here? Who who you know, this is this is no different. So there’s

There’s if you think about these labs that we’re talking about, there’s a second thing that these labs structurally can’t be. even in the enterprise, they can’t be neutral. Right. And that’s the beauty of this channel, is you all are neutral. you are trying to put a solution in there that is the best fit for the customer forever, not for right now, not for you know, these ebbs and flows. that’s that’s the beauty about it. So

Think about OpenAI’s deployment company. Guess what they’re going to recommend? OpenAI. Anthropics Forward Deploys Engineers? Claude, Claude, and more Claude, and then a little side of Claude. They’re selling their own model. Of course they are. But the customer doesn’t need a model. I think they need the right model for their workload. It’s the same thing as other technology problems. They need the right technology for the right problem. And that changes over time. But they also, I think they need it to have some fallback. they need it to route intelligently. They need to not be locked in. And they need integration across connectivity, security, all the other things that you are already talking to them about. So that kind of that that neutral cross vendor, I’m on your side of the table, the those labs can never own that.

It’s just a conflict of interest, I think, by if you just think about the definition of it. It is the single most defensible position in this entire market. And it belongs to you, the independ the the independent advisor. you know, we’ve we’ve said this before look, you’re you’re Switzerland. You’re the one who can actually say, for this, don’t use what the model salesman is pushing. and here’s why. And and you can mean it. So

You know, this this thread, this whole thread, I guess, as as the stack gets harder and it’s getting harder every single week, or or perceivably so, the customer’s ability to self-serve goes down, and their need for a trusted, neutral, fluent guide goes up, right? Up is good. these these AI giants, I think they just confirmed that on the thesis of raising four.

Billion dollars. What an insane amount of money to build that function for the top of the market. So they wrote the check and look, that proves our model. What great validation that is. So I suppose then the the only question that’s left, whether you show up as that guide or you let the relationship just, you know, default to.

Whoever does, if any of those FDEs show up, maybe they will, maybe they won’t. Okay, so that’s that’s the core, right? I think we’ve explained the story, the arc. let’s let’s talk a little bit about, let’s get a little practical, right? I I I want to talk about what to do. I don’t want to leave you with, you know, Josh’s passionate thesis and no playbook. We always love, you know, if you’ve listened to the other episodes, we love to give you questions, we love to give you what to do, what not to do.

Hot technology, but hopefully it’s it’s always been things that can relate to what your customer is going through to help you add more into that conversation, into those threads. So if you’re an advisor and you want to position this, here’s where I would probably put my energy. So one.

I would build some real AI fluency if you have not already. I’ve I’ve seen what a lot of you are doing, some pretty wild stuff. A lot of you are very fluent in this. I you don’t need to be a forward-deployed engineer at Anthropic. You don’t need to be fine-tuning a model, but you do need to be able to walk into there to that customer and say, here’s what the foundational model is good at, and maybe here’s where it falls down a little bit. Have you thought of this? Have you run these frameworks against it? do you have resiliency around that?

How are you tuning this? Do you have that expertise in-house? And back to the earlier thought about going through the door. Are you going through a door that you cannot walk back through if this fails? And are you resilient in that? So you know, if if y you think about you don’t want to be locked into that one vendor when three of them just shipped. So I I think that conversation is is worth real money.

It’s hard to have it credibly yet, especially if you’re these frontier models, right? So I I I I think that’s your opening. I I I I tell you from doing this hands on myself, the more you learn, you you learn a lot from building, you learn a lot from doing. but riding shotgun in these engagements, I’ve watched a lot of you ride shotgun as as we’re talking about these, you know, routing options and the deployments and the consulting and the SOWs. But I think you also learn a lot from just the education.

Look, we we try to give you education that’s valuable on previous episodes and future episodes of this show. You’ve got these things out there, Sam’s minute snippets, you you’ve got all of the AI training events, all all of the events that Toleris puts on. There’s a lot of ways to learn. You’ve got the LMS. you’ve of course you’ve got you know, the big culmination coming up, you’ve got Partner Summit here coming up in August, but there’s a lot of places to learn. And I would just you know encourage you constantly stay immersed stay in that keep positioning yourself and just double down on these absolute basics right it’s it’s you’re this neutral advisor orchestrator lean in on the fact that just like anything else you don’t focus on selling a model that’s not a weakness I don’t I don’t think that’s kind of your entire moat I’m gonna stay vendor neutral I’ll help you pick the right intelligence layer match it to the workload

Integrate it with all the other things that I’ve helped you with connectivity, security, infrastructure. None of those labs guys can say that in one sentence. I don’t they don’t know about those things. And quite honestly, I don’t think they care about those things, which is great because there, you know, there’s a lot in this, I think, to understand. So third here, I guess own this deployment gap back to the 95%. that’s not something to be scared of AI. That’s your addressable market.

Every one of these pilots that’s failed has a has a customer that is proved they want this. And they just couldn’t quite make it stick. So just show up as that person that you know, helps with that gap on the outcome. And you look, you don’t have to have the FDE muscle alone. this is what what we’re here for, this is what the engineering bench is for, the relationships, the architects, these are the people that are here to help you from our team. ourselves and and and and the suppliers. Look, you’ve got that that group and you know.

Maybe point number four five, I guess. move now. The window of of the AI advisor is very differentiated. and I I don’t know how long, you know, does that stay 18 months? Does that stay 12 months? I I just think it’s moving so fast, it’s hard to tell. So, but what I’ve seen in all the early conversations, the advisors who are planning the flag in twenty-six are the ones that are planning it at at the right time who own the relationship when it matters. So final thoughts here, right? The the headline everybody read was, you know, AI Labs is is is launching from these, you know, private equities and these forward-deployed engineers. The story beneath the headline is I I want you to walk away with is probably the most sophisticated technology companies and the most sophisticated capital just looked at the future of this market and said, the durable value isn’t just the model. It is the deployment, it is the integration.

It’s the human who sits on the customer side of the table and actually makes the technology the deliver, makes the technology deliver. So that human at the top of the market for you know, if you’re if you work for open AI, that’s a million dollars. That human for everybody else, the other ninety-nine percent of customers for the the the the mid market, the SMB, the customer who needs that kind of, you know, neutral advice for 400 vendors is the advisor. It’s you.

The the stack is getting harder. That’s not bad news. That is the business model. The harder it gets, the more the end customers, the more the world needs people who can navigate it. and these guys just paid $14 billion to confirm it. So looking ahead here just a little bit, keep your eye on whether these FDEs, you know, is as you hear about that in some of your customer accounts, and just watch how fast these orchestration layers mature and we’re gonna keep it.

You know, keep an eye on where the next opportunity opens up. And just like we try to bring you hot topics here, we’ll keep tracking it. So look, that’s the story for today. the AI giants validating the you know, the the the fourteen billion dollar tell I think. So look if if this helps reframe how you’re thinking about AI in your practice, in your agency, share it with someone who needs to hear it, share it with the team. I’m your host, Josh Lupresto S V P of Sales Engineering, and this has been Next Level BizTech.