Loading summary
Transcript65 lines
- [00:00]
A
Foreign. Once again to the 3 Smart Guys podcast, sponsored by Directions on Microsoft, your independent source of information on all things Microsoft enterprise software, cost management and licensing. Hey, it's been a while. Vacations, travel, conferences. Good to be back here with my friends. I'm Barry Briggs, joined here by Peter o'. Kelly. Wave, Peter and George Gilbert. George Wave. Today I thought we'd focus on the two big conferences recently, Google I O and Microsoft Build, which took place in the last few weeks. I think to a certain extent we're all suffering from a little bit of buffer overflow. Peter suggested we call this episode Too Big to Flail and we'll talk about whether that's really, really the case or not. And we'll probably also get to the Snowflake Summit, which also took place last week. George, you attended, I believe. But before we begin, it's been two months. I've been doing some interesting things. What have you guys been up to in our hiatus? George?
- [01:18]
B
Well, nothing I think is as interesting as you. I mean one of my big highlights was the Snowflake conference and, and then I've been building up my research platform around services software and just the big takeaway was that this system of intelligence or digital twin that I've been harping on for a couple years, I understand now you can't really build it without this new intelligent client and that's why I think we're all going to be talking about the same thing, which is what why Snowflake had to have this cowork client and co co client for builders and why OpenAI and Anthropic are going to need intelligent backends because all the intelligence they capture from and interacting with users is going to need to turn into something that Microsoft showed like with Work IQ and Fabric IQ and then Foundry iq and you know that, that Google also showed with their knowledge catalog. But everyone's converging on the same space.
- [02:27]
A
We'll talk a little bit more about that in just a second. But Peter, what's been going on with you in the last couple of months?
- [02:33]
C
Last couple months? Well, I'm most immediately just back from vacation, so I'm relying on my friends Gemini and Claude and NotebookLM to help me sort through the buffer overflow of all of the latest announcements. One thing that I've been doing in my spare time is exploring personal knowledge management tools, kind of a crossover with that so called second brain kinds of approaches. And I think that's related to what you were just talking about as well, George, in the sense that that could be really central to the next client.
- [03:02]
A
But you know, go ahead. No, I was going to say it feels to me like Notebook LM is kind of the great unsung hero of all these AI tools that it's absolutely invaluable.
- [03:15]
B
I actually listen to as always, the great interviews are on the verge where Nilay Patel, he just does these interviews where he distills down these really critical issues into simple questions. And then his gifts, he draws them out and he talked to Sundar Chai from Google about how all these products ultimately have to fit together. And whereas NotebookLM started out as this labs project, it is going to be like the folder kind of intelligent client within Gemini and other products. It's not just this island. And then Peter, as you discovered that even Microsoft's working on something similar.
- [04:03]
C
Yeah, and it's moving fast. Like NotebookLM just also added automatically synchronized Google Drive file integration. So that was a big wish list item. But yeah, anyway, that's pretty much what I've been up to. And Barry, you've had a pretty interesting spring in the early summer.
- [04:21]
A
I spent three and a half weeks in Korea and China and as a tourist and I do have a few observations coming back from, from China in particular. That one was that one of the surprises was the infrastructure was stunning. It was, they, the roads were in absolutely pristine condition. We went, you know, hours and hours outside of all the big cities and at no point did I ever see a pothole everywhere. You saw electric cars and they told me that something like over half the car sold last year in China were electric. And I think that has to do with infrastructure, meaning that they're charging stations everywhere, which is, which is something that's been long been promised here in the states and has not been realized yet. Another aspect of the infrastructure thing were the trains which were amazing. And George, you know, you live right in California, you know, as far as I know, still hadn't been any real physical progress yet on high speed trains. And correct me if I'm wrong, but you know, we were on multiple trains between big cities, all of which went 200 miles an hour whizzing by quiet as can be probably more comfortable than air than airlines for sure. So it was really a pleasant way to travel. Finally, China's a very high tech country, but obviously has been discussed ad nauseam in the media. The cultural values are very different, you know, so it is, it is indeed a surveillance society. And if you walk down the Bund in Shanghai there are cameras. I think I counted like every 75ft to 100ft, something like that, literally everywhere. And so that takes a little bit of getting used to. But on the other hand, there is literally no street crime. You know, it just, it's just, it's just non existent because of, because of all the surveillance. So you know, it's, it's, it's an interesting trade off because I know certainly here in Seattle we're having that debate about putting cameras in as a, you know, you know, particularly with the World cup coming, you know, privacy versus say versus security and safety and so forth. And it's, it's still going on. I don't know how it's going to turn out. And of course there are robots, There are robots everywhere. And you know, I saw a couple of robots boxing at one of the shopping malls in Shanghai. So, you know this, there's a lot of technology and it was, it was fun. See, I certainly had envy for some of the electric cars and electric SUVs which were very luxurious and probably cost dollar for dollar about half what they cost here in the state. So very, very interesting. Obviously there's lots of, lots of downsides to Chinese society too, but, and this was a very curated trip. It was, it was nonetheless fascinating to see.
- [07:26]
B
I make one comment on it, which is they are better than anyone at building airports, bridges, roads, you know, trains, the top down stuff that they could specify in a five year plan, you know, the entrepreneurial what's going to bubble up next. We're, you know, still pretty good at that. And they're also very good at saying, well, we're going to be really good at electric cars and solar panels and batteries because that's important. And even if, you know, they wind up with a huge surplus of the stuff and then export the rest of, you know, dump it on the rest of the world. Yeah, they're very good at that.
- [08:06]
A
Yeah, I saw, I saw in a small, small city of only a few million that we were walking down the sidewalk and there was this shell, just concrete walls of, of a shopping mall. And you know, we, we didn't even know what it was. We asked our tour guide what's that going to be? And, and he said it was going to be a shopping mall. And we asked when it was going to be open and he said this summer. And you know, we were kind of taken aback because it was, there was just nothing there other than the, the concrete, other than the foundation and the, and some of the walls. So yeah, things happen fast. Anyway, let's move on to our topic here. George, you made this comment just before we started. I thought it was really interesting that coming from all the different conferences and there's still one more to come, the Apple Developer Conference in the next few weeks. But you sort of, you made the comment that it felt like everyone's fighting the same battle. What did you mean by that?
- [09:14]
B
Something I didn't fully appreciate until the last week or two, which is that we're witnessing the birth of a new intelligent client, kind of like the PC or with Windows or the browser or the mobile smartphone. But to support this client, the intelligence it needs lives on a new back end and the two need to be co designed. You can think of Clayton Christensen's Integrated innovation or Jensen's Extreme co design, but the two inform each other. I didn't understand or appreciate fully to what extent. That's because the intelligence on the back end, it's not a bunch of knowledge engineers going off and modeling an enterprise or your personal workflows, but it's partly bottom up from the client interacting with business users and builders and capturing their expertise and their workflows and then trying to harmonize it somewhat autonomously on the back end. And when they can't harmonize it autonomously, having builders disambiguate it or presenting choices to users, in other words, the two grow up together and there's this network effect, like making one smarter makes the other smarter. And so you can't, I don't think you can effectively compete on either the front end or the new back end without you can't do them independently. You have to do them both.
- [10:59]
C
Do you think?
- [11:00]
A
So how does the Spark RTX fit into that picture? You know, I, I saw it and I, I, my initial response to that as a, as an AI enabled client was do I really need this? And I think what you're saying is the answer is yes.
- [11:19]
B
I think the answer is that like there's this think of LLMs as adaptive compute infrastructure and there's going to be some adaptive compute to go along with your deterministic compute, which is like your cpu and it's going to live on the client, it's going to live on the back end and it's not all going to be one big frontier model in the sky because there's this Pareto frontier where sometimes you're going to need something that's faster and cheaper or sometimes you're going to need something that's fine tuned and specialized to a particular task or domain. And so yes, this new PC has silicon so that you can have edge intelligence. Just the way Apple has designed their mobile devices and their PC to have a neural accelerator so you can have edge intelligence. They haven't delivered it yet, but now partnering with Google, they're going to have on device intelligence.
- [12:28]
A
So unmetered intelligence, Peter, seemed to be the buzzword of the day at Build. I have to say that when I heard that, my first reaction was I get it, about the distribution of the tasks and that some things are better suited to running on the client, some things are better suited to running on the serv, you know, the Pareto frontier and all that, you know. But on the other hand, part of me was thinking that, you know, that Microsoft may be unwittingly cannibalizing themselves by putting more of the inferencing, you know, for free on the client side. You know, as you think about particularly Spark rtx, if that's the first wave of a new generation of PCs that have built in, you know, Blackwells or whatever you know, doing, doing inferencing, you know, to what extent does that, you know, change the economics of how we think about, well, computing generally, Microsoft specifically and, you know, an end user computing.
- [13:34]
C
Yeah, I think this is going to be really challenging and as you point out, there could be some negative consequences for a lot of presumed business models. On top of this, back to Jevons Paradox, or however we pronounce that, where it's basically, hey, just keep turning it up because no matter how much you increase supply, the demand is going to go along with you. But I do think there are lots of innovative things that can be done with local intelligence and especially when we think about privacy and security. And again, this is, I'm sure, going to be central to the Apple Worldwide Developer Conference here in the next couple days. But does this change things or is this additive? You know, is this going to be. Now it's better because you have the ability to leverage both of these things. I do think that there's a fair amount of just believe that this will come together. I think the distribution of this is not going to be easy, especially when we're crossing over into what we've discussed a few times as neuro symbolic, where it's not just LLMs, but it's also tying into both your personal data and enterprise systems. How do you make sure that you're going to do a coherent job of that without having inadvertent oversharing?
- [14:47]
B
For instance, I would add on the Jevons Paradox, I just build on that thought, which is as we have adaptive compute not just agents, but the thing that responds based on context becomes cheaper and more pervasive. If you have it on the client, you might be in a position to generate more of your user interface or the user experience. The other thing is the trade off when you fine tune models is that it makes it harder for the cloud vendor to be multi tenant. So you might have the fine tune model on the client and the general intelligence on the server or, or if you have something fine tuned on the server, it might be for serving a department or a process.
- [15:41]
C
So a question for you, George. Just thinking about your comments about the synergy of client and server or server side working together. Do you think these are going to be tightly coupled? Do you think we're going to see like a new proprietary shift where things are going to be better together from a single vendor, or do you think that this is going to be something that's going to remain like Microsoft at Build wants to be the most open platform. Is that maybe situational?
- [16:09]
B
Yeah, but it depends how they're defining platform. I think their platform includes their 365 Copilot and Spark. And then user interaction with that enriches work IQ and fabric iq. You're capturing more and making that intelligence repeatable and shareable and harmonized. So I do think better together and then the message is, oh, build on this platform. Whereas like OpenAI and Anthropic, they're going to have very high volume intelligent clients, but they don't really have anywhere other than your personal harness to capture the intelligence they learn from you. The skills, for instance, other than let's say a skills catalog or a plug in catalog. But there's no place to harmonize it and make the organizational intelligence smarter.
- [17:11]
A
So do you see a kind of an emerging architecture of, you know, one of my friends talked about, you know, a new generation of client server, you know, whether it's, you know, or mesh, you know, of sort of cooperating, you know, agents that are all over, you know, sort of widely distributed across. And there's a kind of a separate but related question which is, you know, something like 90% of the workloads in AI are around inferencing now. And that's a pretty straight line kind of straight line kind of task. And do you think there's an opportunity to build new kinds of data centers that are strictly for inferencing only, that are cheaper, that can use, you know, maybe less sophisticated hardware, that sort of thing? I realize I asked two separate questions there, but take a shot at either one.
- [18:06]
B
Well, actually, it's funny that you mentioned that because our infrastructure guy, who's a complete genius about the entire infrastructure stack, has pointed out that, you know, everyone's, there's, there's many pretenders to the Nvidia throne now for accelerators, but this extreme co design is when agents, you know, it's not just about inference because the inference or adaptive compute is calling on tools. And if you want low latency in the call and response, it's not just, oh, send it over this interconnect to my CPU in another rack, you know, the two are designed together, they're on the same memory bus, you know, and that maybe, you know, there's. That the CPU and the accelerator are now turning into one unit because the adaptive compute, the, the agent and the tools it calls on are, are chattering back and forth at low latency. So I think that also sort of
- [19:14]
A
says that the, you know, the tools themselves have to be co located with the, with the inferencing engine. I think that was part of your point. Yeah, Peter, you know, from what you saw, what was, what was the biggest headline of all the conferences?
- [19:33]
C
Well, so I think across the three of them, and George has mentioned this as well, I think one of the implicit biggest headlines is they're all converging toward a similar architecture and they're coming with different strengths and weaknesses into it. But now, you know, we're entering this with anthropic and OpenAI filing further IPOs. We are in uncharted territory in terms of the level of resources that will be thrown into this. So it's interesting back to sort of a paradox of abundance. You think of the totality of what was presented at Google I o our build 2026 and we're all getting saturated. It's like, oh yeah, there's the obligatory quantum stuff and then here's everything fits together across this and we're quietly burying our DE and just moving forward. Everybody's got the same picture. So I think it wasn't a huge surprise, but it's just observing that I think we're kind of at the end of the beginning of AI and enterprise computing. I think it's getting real and now the expectations are shifting and people are getting very impatient. Like, what do you mean? I can't explore this till summer. I've seen enough demos. I really want to use this myself and explore it.
- [20:50]
B
So yeah, and I would just build on it. And to say, I didn't really realize until this week, last week, that everyone's converging on the same thing where we're seeing the rise of a. It's not like a personal agent, it's a new client that's intelligent and it needs a new back end to support it. And that's why I think one of the things you take away is Anthropic and OpenAI are going to have to build some sort of intelligent backend to capture and support the intelligent, capture the intelligence coming from their users and support the intelligent client.
- [21:33]
A
Yeah, it's funny, I actually walked away with a different perspective which was I think you're right about the architecture converging but my perspective was that the business models now are solidifying around their, around Microsoft's and Google's core value, which is Microsoft is at its core as a developer company. And you know, Build and you know Build was explicitly targeting developers clearly but nonetheless, you know, everything they presented was really was strictly focused on the developer and that is where Microsoft's heart is, having worked there for some number of years. Whereas you know, Google, I thought one of the most interesting announcements I thought at Google was this unified shopping cart. And you know that's again that's kind of Google's core mission is to present, you know, call it advertising, call it commerce, you know, tightly coupled. But you know, bringing together Gemini and YouTube and Google search all together as us as a single, you know, a single common shopping cart I thought was now that, that's what, that's what Google does. That's the sort of thing that you would expect to see out of Google. And so I thought, you know, they're converging on their own core, you know, corporate identity in some sense.
- [22:52]
B
I couldn't let me have one one on that because I couldn't. You, you articulated much better what I'm, I, I talk about a common architecture. You're talking about how their business models make them express the products and the customer set differently. And the same thing with Apple, I think we'll see similar architecture but it's now the intelligent client is going to be on the phone and then they have to build this intelligent backend. Yeah, I think you distilled it which is the different personalities manifest in different products.
- [23:33]
C
So a couple devil's advocate points for me. So one, one of the things they took away and again I didn't have time to go through all of the sessions at Build but one of the things I took away is yes, it's a developer oriented Microsoft at Build is a developer oriented company and the most important things they said were Unix like it's going to be Linux, it's going to be no compromises for working in terminals. It's going to be a new version of Android for which they do not need to Microsoft or Project Celero that's built on top of that. So I think most of what they were talking about there was very different from what they've talked about in the past. But if we say fundamentally from a business model perspective, I think Microsoft really has crossed over into the fungible infrastructure provider. And if you look at the increasingly blue dot covered map of the world where they've got their data centers, that's what they are placing their wind bed on. And they also are going to keep these annuity strings from their traditional products alive as long as they can. And just one other quick comment on the Google side. I agree that the shopping cart is a really compelling demo but I think this is full spectrum war for the enterprise. I don't think that Google is doing this as a part time thing. I think that they realize especially with workspace and AI integrated into workspace and then being able to go TPU up all the way to the client environments and then having Apple bringing Gemini into the picture as well. I think that right now Google is seeing a better competitive opportunity relative to Microsoft than they've seen in 20 years.
- [25:14]
B
They've had the capacity because of the TPUs to agent enable more of their product line. But I do want to make one point about the split personality at Google because they know the enterprise is a big opportunity and but they're so fixated on consumer that you can see it in the split personality of Gemini 3X which is this Omni model. It's really good at multimodal input and output but it's not yet world class at coding. So they have a world class harness with Windsurf but it's not as good as OpenAI or anthropic encoding and that is constraining the sophistication of the enterprise tools built within that harness because they're training it for Omni sort of input output and that's taking up some of what they could have devoted for coding.
- [26:20]
A
Can you hear me? I'm getting a message that. Okay, good. Yeah, I can't let this go without without mentioning my favorite buzzwords which is I have to read it. Humanistic, unmetered artificial agent general intelligence was what I that was what I took away from the combination of Satya's and Mustafa's keynotes on the first day of build. And I wonder you know to what extent we're getting just buzzworded to death here and it's just reaching, you know, syntactical or semantic overload.
- [26:52]
C
Yeah. Or I mean another way to look at it from again, kind of a devil's advocate perspective is you have to put an and between each of those words and say that is Microsoft's narrowly defined value proposition because it will have trouble if it can't have all of those things. So humanistic goes back to the inflection roots with Mustafa Suleiman for instance, but. And then future tense on the artificial general general super intelligence. But yeah, I think that that would clean the bingo. The bingo, buzzword, bingo card in one fell swoop.
- [27:28]
A
Yeah, I would, I would.
- [27:31]
B
I, I took it as a value statement that's not yet reflected in products in the sense that what I think they're trying to signal with the word humanistic is that it's not meant to. It's, it's an anti Dario and Sam message which is we're not going to put every white collar worker out of, out of a job. We're going to augment you. That's because that's part of Microsoft's ethos, you know, empower every worker and organization. But what I question is whether that's reflected at all in the technology yet.
- [28:05]
A
Right, right, yeah. Anything you didn't expect that you saw in any of the conferences?
- [28:14]
C
I didn't and again I was mostly focused on the keynote, but I was surprised by what was conspicuously missing from the Microsoft keynote, which is sort of like copilot, maybe as lowercase cnow and power platform didn't seem to be part of the story. But again, I think those were all based on a lot of very detailed discussions about what's the relative positioning going to be. I don't think anything is going away. Back to the two big to flail theme. You can't just pull things from the market. But I think the emphasis there, to me it really comes down to two or three strategic bets that Microsoft is placing here. And one of them, which maybe was a surprise to some people, but Microsoft coming in and saying they will be in the top four for frontier models and yet also they will be able to leverage commoditized models. If either of those sub bets turns out to be wrong, they, they have a big problem, A really big problem.
- [29:11]
A
I guess what really surprised me was, was this Project Solara. And you know, when I first, when I saw it, you know, I thought, oh, I can't say the word here because I have one of those consumer Amazon consumer devices right next to you so I'm going to spell it but it's a L E X A for the enterprise and you know, you know and Peter, you, you mentioned this in one of our previous conversations. Let's see, is it Zoom, Kin, Connect, Hololens? You know another, another one for the, for the almost immediately obsolete of Microsoft hardware innovations. Do you think, you know, do you guys think that that's, do you agree with me or is my skepticism warranted or, or is this the real thing now?
- [29:55]
C
I, I definitely don't see it having a consumer device impact, you know but I, I could see it in the enterprise and maybe it wasn't an accident that they like what had looked like a badge, like a smart badge going back to Xerox 30, 40 years ago but ubiquitous computing stuff but a badge that says here, here is a reason that you need this in addition to the smartphone that you are never leaving behind. So yes, maybe an opportunity there but I think it might also be seen as Microsoft recognizing that if they can't get resurgence with Windows and more relevance on their own controlled mobile platforms that going to be increasingly problematic. Especially as Google and Apple partner together and collectively own most of the smart non PC devices in the world.
- [30:46]
B
My take was on the sort of the core emphasis which was work iq, fabric IQ and fine tuning the new open models that the Maya models which was if that stuff is ready for prime time then they have sort of now put a layer on top of what Snowflake and Databricks, you know and even Google have done if they're ready for prime time. Because they're saying we're giving you the next gen platform which is the neuro symbolic, you know, it's the adaptive compute with your business rules and your tacit knowledge, you know, encoded in a graph and a way to fine tune your models and then a way to operate within those policy guardrails. If that's ready for prime time then they have a very coherent story. Microsoft's always good at telling the story. Let's see if this is version 3 and ready to sell the story.
- [31:51]
A
Yeah, I have to say that, that you know I was, I was really interested in the WebIQ portion of the keynote. So I immediately went after the keynote, went to the developer site and none of it's actually available yet and you know, reminded me of something we always said when we moved back in our, back in our youth at Lotus, which was that I think it was One of our CEOs said that Microsoft has no future tense. You know, it's there. Of course it's there.
- [32:21]
B
So that's a great.
- [32:23]
A
Yeah, yeah. You know, the one thing, one thing I took away too, you know, particularly from the Celera announcement, but you know, kind of generally too, was that this whole notion of, you know, agent as something special is beginning to feel like, you know, a meaningless word. And isn't it really just an app at the end of the day? You know, and it's an app that has an LLM in it. Okay, great. But it's an app and you know, I have a, I have a word processor that has a spell checker in it. Okay. I have a deterministic app that calls a non deterministic LLM. Okay. You know, it's a. It's yet another dll. And you know, yeah, you can have autonomous agents, but, you know, website's autonomous. A website works just fine all day long without human intervention and so forth. And so, you know, I, I just wonder, you know, are we just really talking about just the next generation of applications?
- [33:18]
B
Yes, with a new architecture.
- [33:22]
C
And I think one fundamental difference is autonomous really means it will not be in all cases directed by humans. It will be taking the initiative on its own with I'm sure, highly variable results.
- [33:37]
A
Well, that's the bottom line, right, is that there is a stochastic element to this which you don't have. Amazon website, for example, you push the purchase button and you expect that a certain sequence of events is going to happen. You know, with, with energetic application, you expect that it probably will happen.
- [33:58]
B
You know, there's an interesting point. It's just coming at this from two angles. Like Andrej Karpathy says, you know, are we summoning like agents as humans or as ghosts? And some people call them like smart microservices, but they're adaptive. But there's another point, which is we say we've built deterministic applications for 60 years, but because they're islands and humans are the ones that bridge their islands, the output has been probabilistic. You don't get the same thing for every input because the human is the one that's kind of synthesizing it and that's not deterministic. So, you know, it's a new architecture and there's a human like, or ghost like entity now that's helping to bridge, you know, and, and it's driving us to eliminate more of the islands.
- [34:56]
A
I've got images of Patrick Swayze and Demi Moore, and in the back of my mind right now. So we're on the eve of the Apple Developer Conference. It's the one conference that hasn't happened yet. Any thoughts? Do you think there'll be any revolutionary announcements or evolutionary. Will Siri become real? What do you think?
- [35:22]
C
I definitely think Siri will become real. I probably should have spelled that just like he spelled the other assistant and long overdue. But yes. And I do think they have pretty broad portfolio of things to build on, to leverage with it. So I'll be curious to see the extent to which they're going all in on Gemini and if there's room for others and if OpenAI is going to sue them for what they perceived as a revenue opportunity, that's not going to be there anymore. And I wouldn't be surprised, like, they're also expecting to see new device platforms from, from Apple. So maybe that's where the Celera came from, was, hey, we need to do a preemptive striker.
- [36:06]
B
I. I think there's the chance for Apple to show something very impressive because the groundwork was laid years ago with Apple Intents, which is kind of like mcp, you know, connectors for every app, but where there's some standardization of the nouns and verbs. So that if you've got that, that's sort of like unifying the data and action space, which we've talked about for the enterprise. But they've sort of been. They've sort of done that and then so has Android. But in other words, the platform is there. They just need the programmability. So the question is if there's going to be on device intelligence, and that on device intelligence is coming from Google, is it strong enough to turn user intent into programmability that then gets converted and manipulates the applications? There's a lot of moving parts in there, but the groundwork is there. And then there's the question of how much of that might live in the cloud too, and then how much might drive the need for Meteor and Meteor devices.
- [37:18]
A
Right, right, right. So we're on the verge of new wave of hardware from Apple, potentially a new wave of hardware from Microsoft, maybe a gadget at some point from OpenAI and Johnny, I was saying, so it's lots of new stuff coming for the consumer and for the enterprise.
- [37:42]
B
This is no way a bubble right now.
- [37:44]
A
Yeah.
- [37:45]
B
So much left to be done.
- [37:47]
A
Yeah. On the other hand, our last question of the day, the market took a really big hit today. The NASDAQ's down 1100 points. Microsoft was down. I just looked 2.6%. Now, some of this is partly due to Jobs forecast and likelihood of an interest rate either staying the same or rising. But do you think there's also potentially an AI pullback in all of this? Do you think that's, that's part of this? If not, or even if you do, do you think there will be an AI pullback at some point?
- [38:23]
C
I think there is a general expectation that categorically things will move forward, but there will be losers and some of the losers are massively capitalized and deeply entrenched and there will be disruption from that. So yeah, I do think there will be a selective reset. And again, part of this might be coming out of next week with Apple developer conference. The market might just pull back and say between Apple and Google and Microsoft and to a lesser extent with aws, there's just the dance floor is going to get pretty sparse.
- [39:05]
B
Yeah, I think the secular trend of adding intelligence to enterprise and then to consumer devices will continue. I think what we may see more is more movement along the Pareto frontier because right now everyone's always gobbling up frontier intelligence, but we could see more value engineering like on device or, or mid range models that decide to call on frontier intelligence when they need it. In other words, we can be a whole lot more efficient about this. And cost pressures may make companies move away from token maxing to measuring business outcomes and the cost of achieving those outcomes. But I don't think the build out is going to slow down, at least anytime soon. I think we just might get smarter about it.
- [40:05]
A
Yeah, I wonder if we're sort of in that phase, you know, where we were, you know, with the cloud about 15 years ago where, you know, where developers, you know, were going out to AWS or Azure or GCP and just grabbing themselves the biggest honking VMs they possibly could and you know, you know, overload, you know. And then at some point some, somebody, a CFO came along and said, you know, do you really need this 128 gig machine? You know, can't you do it for something smaller which will be a lot cheaper. And I've seen that personally in some of my own work and we haven't announced this yet, but I'm doing, I'm doing some work with, with an AI application and I've discovered that there's a lot of stuff you can do with these mini and nano models that you don't need the big, full, full scale, you know, you know, full LLMs for. So you know, maybe we're on, you know, at Some point on the cusp of this sort of bright sizing of the, of the model for the application kind of, kind of mode at this point.
- [41:09]
B
Yeah, that's. I. That part. I think that that reset I think is, is, is. Has to happen. How imminent it is, you know, you know is, is an open question, but just think about. And that was Microsoft's whole message which was if you take a GPT 5.5 and your own reasoning traces and you might distill some more and then fine tune a 5 billion parameter model for your domain, you're getting GPT 5.5 level intelligence. That was a clear message.
- [41:49]
A
Yep, yep. Cool. Any closing comments, Peter?
- [41:54]
C
Anything.
- [41:55]
A
Anything to close us out on?
- [41:58]
C
No, I, I think from, you know, if I project it as a Microsoft Enterprise customer, I would be looking at those big bets, you know, so the big bets including is Microsoft really serious about making the investment to be in the top four or, and, or can it, can it play the best of all of them as they get commoditized? And to your point, on the agentic stuff, I do think we're seeing pretty fast meme cycling here, so maybe we'll just sort of. Yeah, we're not going to talk as much about agentic. It's all about context now. But that again, like what's really going to persist here as something that's going to be valuable to mainstream customers, not just cool demos.
- [42:40]
A
George, any closing comments?
- [42:45]
B
The one thing that may keep this, you know, going going forward where we might have paused in the past is like when we went from ChatGPT, the ChatGPT release was like the browser moment and then Claude code on, on Opus 4.5 or 4.6, I forgot which. The one in the fall that was like the introduction of the smartphone and rather than that being separated by like 12 years, it was separated by three years that the improvement is going so fast that that secular trend is less likely to see like imminent consolidation. We might do value engineering. Right. Sizing, but the improvement is going so fast.
- [43:35]
A
Yeah, I agree. And you know, I think, you know, from my own closing remarks, I just say build on what you said, George, which is that that has huge implications for Microsoft in terms of focus.
- [43:45]
C
Right.
- [43:45]
A
You know, they're going to have to stay focused. You know, again, I don't mean to pick on Solara, maybe it's going to be the greatest thing ever, but it felt like a bit of an outlier, you know, some extent. You know, and there were, you know, and I think the challenge when the big companies a lot of smart people is staying focused on what delivers the most value to customers. So once again, as and as always, it's been a real pleasure chatting with you. Thank you. Thanks to our sponsor, Directions on Microsoft. Again, if you need a reliable, independent information source about anything Microsoft Enterprise, whether software, services, costs or licensing, come visit us at directions on Microsoft.com it's right up, right up there. And if you like this podcast from the three of us, please give us a thumbs up. And if you have any feedback on anything we discussed today, and if there are any topics you'd like us to focus on in a future episode, put it in the comment and to the Sessions YouTube page. Again, don't forget to subscribe. Thank you very much.