Unveiling Google Gemini Three: The Future of AI | Utilizing AI Podcast Ep. 5

Google just announced Gemini three progressing from understanding to thinking to action with their frontier AI model. This includes Nana Banana image generation, uh, anti-gravity coding assist. Uh, what should we make of Gemini Three, apart from their use of crazy code names.

We also discuss in browser AI and coding assistance, as well as news from Microsoft Ignite, including Fabric IQ work, IQ and Foundry IQ on this episode of utilizing ai, welcome to utilizing ai, the weekly podcast focused on practical applications of artificial intelligence from the Futurum group. Each episode brings together diverse perspectives to explore news and new use cases for the ways in which AI is transforming enterprise IT and the industries it serves. I'm your host Steven FoST, president of the Tech Field Day Business Unit here at the Futurum Group.

Before we dive into the discussion, let's meet who's on the panel today. Hi everyone. I'm Brad Shiman, and Steven, thanks for having me.

I am an analyst with the Futurum Group looking at data intelligence, analytics, and infrastructure, and I'm an unapologetic, uh, believer in all things ai. So bear that in mind as we go. And I'm Nick Patience on the AI platform's practice lead here at futurum.

I've been looking at AI for 25, almost 26 years now. Um, I kind of, I, I agree with Brad. I am a, um, an un unapologetic supporter of AI as well.

And I'm Steven Foskett. I guess I'm the, uh, designated skeptic here. Uh, now I, I'm unapologetically enjoying and using it.

In fact, um, I did some really cool stuff with it this week, but, uh, I am a little skeptical of the business model and, uh, and financials of the whole thing. Uh, but we'll get to that. Uh, not on this episode though, because we've had actually some new announcements this week.

Um, key among those, of course, is Gemini three from Google, and I was thrilled to see that, um, you know, a little bit of inside baseball here. Google actually made this announcement, uh, in a university setting. And, um, the first, uh, official discussion of it was with a bunch of young people who are just getting into the industry, and I love that vibe.

Um, I also really enjoy Gemini. I've been using Gemini, um, as my sort of go-to engine for a while now. Um, just sort of to, to, to get stuff done during the day, uh, along with, of course, apple Intelligence and Gemini three, um, at least so far looks like it works.

It's like Gemini, but better. Um, I guess, uh, uh, Nick, let's, let's throw this to you first. Um, what's your initial impression of Gemini three?

Yeah, I agree. I guess it is, it is Gemini, but better you'd expect it to be better after all. Um, that is, that is the kind of, that is the kind of point, the way they've, they pitched it to me.

Um, the, uh, recently was sort of the Gemini one family was about, um, understanding Gemini two family was about, uh, thinking, and the Gemini three family is about action, so taking action. So this is where Gentech and, um, comes in. Um, but there's obviously still, you know, reasoning model and, and, and all that kind of stuff.

And I think the, some of the stuff that was, you know, pretty interesting was the, um, you know, Nana Banana being part of it. Um, and there's been some really interesting images going around, actually interesting images right now from an enterprise AI point of view, not the usual, um, stuff that we see, um, of, you know, astronauts riding bicycles, um, and, and things like that. But actually, uh, more, you know, it's ability to, um, you know, create images that are useful in presentations, not necessarily presentations themselves, although it's probably getting, it's getting there.

Um, and also, you know, this kind of advanced planning capabilities and things like that. I've been digging into it a little bit. Um, yeah, we're a kind of Google shop here at Futurum, so, um, so we've been using it a lot.

You know, I'm, we all have, I know. Um, but I think it's, you know, it's, it's really interesting. And there's also, um, there's other things around it.

Maybe Brad wanna talk about the, um, uh, anti-gravity and that, that Cain as part of it as well. Yeah, it's, it's a fascinating, uh, release, um, for a lot of reasons, like Nick just mentioned. Um, and it's funny, isn't it, it seems that now we have, uh, a cadence every year in early December, we get together and talk about all of the new groundbreaking, you know, from, what do we call those, um, frontier scale models that come out.

And, uh, so I'm just waiting for the next deep seek to, to roll out, uh, before the end of the year. I'm sure it'll happen. Um, but, uh, for Gemini, I, I find it interesting for, for a couple of reasons, um, first to touch on the nano Banana facet, uh, best name ever, by the way.

I'm sorry, can we just take a moment here and say, yes. Did you have nano banana and anti-gravity on your Bingo card this week, Right in the center? Yeah, I mean, what's with that?

Okay. Yeah, go, go ahead, Brad. Yeah, Yeah.

And, and it's like you mentioned Steven with rolling it out in, you know, with younger people, uh, and those in research in particular, because Google has really shown in this year that amongst the US based model makers, that it is, uh, right there with anthropic in terms of, you know, bleeding edge, leading edge, uh, innovation. So, uh, I get that a lot. And Nano Banana is a good reflection of that, because, like Nick mentioned, it's not just about funny images.

It's, it's actually like if you're working in the enterprise and you're working to build software or maintain or manage software, think about what it means to have an image generation model that can actually output, you know, correct syntax, correct grammar, correct English or whatever language, uh, uh, labeling for a very complex, um, diagram of, uh, something let's say all the way down to wiring for, or nano chip design even, let's say, I don't think that's outta the question, but think about what that means. If you can, instead of having to pivot over to a CAD cam design, you can actually just have generative AI create a design that's consumable by software itself, other AI agents or just software. That's, that's amazing.

Think about it, like, at a really high level, just for building software, if you can, um, actually create an interface that that is, you know, not, not just a sketch, but an actual, you know, workable interface that can be broken apart by the model and built as software that, that's, that's amazing to me. So I think there's a lot of practicality with Nano banana. Um, the second thing that I wanna touch on really quick is, um, that, you know, we all need to bear in mind that it's not cheap or easy or or fast to develop these frontier scale models.

And that Gemini three is not actually a new model, like Nick mentioned. This is all about tool use and refinements to an existing foundation. 5, the end of 2024.

So it's, it's, you know, not an entirely new model. It's just doing a lot more with what it already had. 5, uh, was that it would run home to mama, uh, every chance it got in terms of, you know, coming up with, uh, a, um, API calls for its own models that were from 2024.

And so you're coding in the middle of 2025 and on, you're like, why are you doing this? And it's because the corpus was trained on that. Well, you know, they've actually, you know, made the model much better at not just using like MP MCP servers to do this, but actually better at, you know, being able to, to sort of see the context of what you're working in so that it uses correct API calls.

So I, I appreciate that about it a lot. And if I could just say one caveat, uh, also that I, I found with working with Gemini three is that, um, it's not very chatty. I, I, I missed that.

5 because it would, it would actually, it did, it did, you know, do some glazing now and again, like, oh gosh, how did you come up with that? Great question. But, you know, you could look beyond that and instead really enjoy the fact that it would share with you some of its thinking and reasoning tokens about how it got arrives at, you know, oh, this is what I meant to do.

This is what I need to do. This was a dead end. I tried this and it worked.

You don't get that with three, it's just all down to business. And the business is pretty good. So I, I guess I can take that as a, you know, an acceptable, uh, downgrade.

That's pretty, that's pretty interesting. They, it's kind of like, um, it's kinda like seeing a band where the lead singer doesn't talk between the songs. It's just, just the songs, you know, that's all you're gonna get.

That's right. Um, but I, I was born without the screaming, uh, to, to clap. Exactly.

Exactly. I think, I think it's interesting where we sit at the moment this week, uh, two days ago it was, uh, chat GT's third birthday, um, and how quickly we, how far we've come and how, also thinking back, um, earlier this year, I can't really put an exact point on it, but somewhere in Q1 where, um, you know, after Deepsea came out in January and, and other things happened and, and open AI c carried on its March how Google was apparently doomed. Um, and the ad search business was, you know, gonna completely fall apart.

Um, and here we are, um, as, as Brad said in, in December and, um, of the hyperscalers, I think it's fair to say, um, it's looking like the one that's, you know, currently in the lead in terms of AI from the chips to, to everything to the apps at the up, up, you know, the top of the stack and ev and all, of course, everything in between, which is where the, uh, raw meat and potatoes is. Um, but, um, I think it's, uh, you know, it's in a, it is in a really, um, really interesting position. Yeah, you can't count Microsoft out, um, on that, but man, Google is doing everything right, um, in the AI stack it seems.

Um, they are, yeah, they have really solid models. Um, you know, we, we should probably also talk about anti-gravity versus cursor. 0 just came out, um, with a lot of the same features and, and anti-gravity is right there.

But, um, you know, it, it is really interesting that, uh, Google is able to, uh, to develop, um, really solid, useful tools, uh, get them out there to customers. Um, you know, I, I wonder about the financials of, of it, but I suspect given what Nick just mentioned with Google's own, uh, silicon that they're using here, that they, that those may not be as all that bad either. Um, and that we may be seeing the emergence of a really solid, um, offering here.

One of the things, Brad, that I wanna react to is, um, one of my greatest frustrations with dealing with LLMs is the sort of, um, compounding chaos that you get when you're trying to work with an LLM because that, I don't just throw something at it and say, give me a result. I like to try to iterate over that and say, okay, that's good, but now this, and what I found, especially with the OpenAI models, is that it tends to spiral. It tends to compound, and you get just absolutely bananas, sorry, Google results, especially, um, when it came to chat GPTs image generation, it would really just go off a cliff after four or five images, and you'd end up with something that just was completely wrong.

And I would have to basically say, forget this. Let's go back to the start and try that all again. Um, so far I haven't experienced that much as much in Gemini three.

And I'm curious if you all, first off, have you experienced this or am I just in inept and second off? No, you were, is this something that's getting better? Yes, and you're totally right.

You're totally totally right and that, that, that's spiraling of, of images to the fact that you completely reset and say, this is all completely useless, let's start again. Or often what you end up doing in those situations is going to another, you know, going to clawed or going somewhere else and just trying something and then, and then getting frustrated and then building the slide yourself from scratch. Um, but I think it's, you know, I think these, yeah, these things are, you know, they are, I mean, I guess the thing is we're always told they're only gonna get better, but they do get better, but they get better when the next version comes out.

And so, you know, you, because you're not your, because I always say your stuff is not being used to retrain that model. Um, and, you know, obviously, as Brad talked about earlier, the cutoff date proves that it's not being retrained. So yeah, you do kind of have to wait for the new version.

5, um, earlier today, and it's the first one I found that actually, I mean, I've been talking from an analyst building slides a lot, so this is a very kind of narrow point of view. It's the first one that actually built a slide, a proper one, you know, a PowerPoint slide. I mean, what we use Google Sheets, but in PowerPoint format that it was actually good and all the words were correct on it, and yeah.

And all those kind of things. So I think, um, you know, in terms of personal productivity for me, that that's, yeah, that's huge. But yeah, I think, we'll, you know, the images, the image stuff is, um, it's always the first thing to be publicized, isn't it?

Because everybody wants to, you know, put out all, put out all the silly stuff. Uh, but I think it's, it's gonna be really, really useful in all sorts of, you know, the creative industries, but also, you know, everybody else who's not a creative professional, but is a knowledge worker of, of any kind. Yeah, very true.

My, the first thing I always use these models for is to, to work on designs, airbrush designs for my 1970s panel band, uh, the ones that, the, the little half moon windows in the back. Um, but, uh, How many panel vans do you got now, Brad? Um, yeah, so, so, uh, it, it's the same for all use, all use cases for these, whether, whatever, if it's a multimodal, uh, or reasoning or, or just a, you know, uh, use drains, um, I'm trying to think of the word now, sorry, uh, a ba a basic chat response model.

Um, it doesn't matter. It's the same because, you know, we're getting better at how we handle context windows, how big they can be, and we're getting better at the attention mechanisms we have that tell the model how to work with the content it sees in those. And so, like you're describing, Steven, you know, it, the, the challenge often when you're working with these, and this is especially the case for age agentic development, is that the further you get into a session, the the less likely you are to, to end up where you thought you were gonna go.

And it's, it's, it's like, you know, I, I can think of it, it's, it's, you know, a human, it's the same as a human being in terms of fatigue sets in confusion, sets in conflicting ideas set in waiting sets in. Like, well, what's the most important thing in this huge context window? Oh my goodness, how do you know that?

And, and so, you know, and I, I say this, Nick's probably tick of hearing me talk about it, uh, 'cause we do internally all the time. And that is that it's not about prompt engineering. It's, it's all about context engineering.

You, you really have to be very careful about how you put data in front of a model in the context of what you're asking it to do. Because, you know, the more we work with these, the more I have found that, um, you know, the, the, we, we ask more of them, and therefore they do less until the next iteration, next Christmas comes around. So you kind of have to, you know, be very c cautious about how you use them.

For me, like working ag genically for software development, um, uh, my best friend is the forward slash compress command to take everything I've been talk everything that's gone into that conversation and compress it down. Two things that does first is it saves me a ton of money, and the second is it keeps the model from going off the rails, or less likely to go off the rails. The other is, is simply just cut it off, just clear the cash, clear the memory, and start over because there are limits.

The more we ask of these, the less they're able to do, honestly, you know, until we figure out how to do it better. Yeah. The frustrating thing for me is yeah, when you're in there and you know, you literally will say, okay, keep this part the same and fix this other thing.

And this, I've experienced this with coding assistance, I've experienced this with writing, and I've experienced this with image generation like crazy with image generation. You'll say like, okay, I like that part. Keep that the same.

Now let's modify this other part. And somehow it will change that part too. And with coding assistance, that can be especially pernicious, because you can find that the function that was working and, and that you got done with before is suddenly off the rails again.

Um, ha Has anyone tried a coding assistant with Gemini three Anti-Gravity? I guess I have. Um, have you I've spent, yeah, I've spent, I've spent some time with it.

Um, and, you know, I, I, I was, well, I wasn't actually that anxious to use it because I'm not a fan of Electron and I, I really load the VS code, um, because it's, it's really just JavaScript with plugins all day long. You know, I like, I like something like Zed, that's, that's sort of purpose built to, to, to be, you know, a code, an IDE and, um, anyway, my, you know, preferences aside, I at getting to an outcome, you know, that outcome, if I say a very simple prompt, you know, make, make me a Pomodoro timer, you know, it, it will just set up my environment, get everything in order, like an actual, like good working order, not just half, you know, witted sort of, you know, A POC that you would not wanna share with anyone because they, you'd be too embarrassed to show them, but actually good coding practices to get there. Um, but I, I just don't, you know, I, it's, it's not something that I think is fully baked yet, and it's got a long ways to go.

And I think it also opens up a lot of, of potential issues for us because of, its, its use of what we, you know, have come to know as, as browser use or computer use. Where models are, are given the ability to, to work with your system itself. I mean, with these coding agents, we already give them access to RM forward slash you know, Tilda Tilda rm, you know, to delete everything.

Um, but, uh, you know, it's, it's like that with, you know, just handing the reins over to the horse and saying, take me home or drag me through trees. Let's go for the former. Okay.

You Got a lot of metaphors in there. First off, um, I think you're gonna have to use nano tomato, uh, to design the UI for your pomodoro ti timer. Um, also a baked, uh, fully baked pasta is delicious, especially with Pomodoro.

Um, and, um, and we actually went down to Amish country this weekend here in Ohio, and the horse absolutely can take you home when you're drunk, um, in the middle of the night. That is a, that is a dangerous practice, sir. But I'd let you know.

Um, but Nick, let's, yeah, so let's talk about coding, uh, and browser use. I think browser use is one of those things that people are a little shy about. Yeah.

I mean, do you mean shy about admitting they're using it or shy about the, uh, or wanting to actually get started with it? Yeah, Like, like, do I really wanna turn my browser over to Gemini? Yeah.

And I, and I guess they're all, they're all pushing you, perplexity is pushing you, and they're all pushing you open ais we're all pushing you to do that. Yeah. There's obviously these kind of, um, core security risks of the, you know, the kind of, um, the, the kind of, what is it in, in indirect prompt injection, um, and, and things like that.

I think it's, it's, it's interesting, you know, because we're kind of back to the nineties again, aren't we? With, with browser walls? I'm, yeah, I think we're all old enough to remember, um, you know, Netscape and IE.

And all the others, um, that, that, that were around. I think it's, you know, I think there's, yeah, there's, there's gonna be, I think it's gonna be pushback on the corporate level anyway. Isnt corporate it on those kind of things.

I mean, there's, there's, there's only so much shadow ai, um, you, you can get away with, um, in a large corporation. I mean, you know, smaller companies than, than, yeah, I think, you know, people will, I mean, obviously the, the, the switching costs then becomes so high, don't they? Because if you are investing in, you know, if that becomes your browser, and obviously that, you know, becomes incredibly sticky, that's obviously why the vendors want to do it.

Um, so you just don't, like, like we'd been discussing earlier, you don't just go, well, I'm not gonna use that image generator anymore. I'm gonna jump over here and use this one. Um, but I think at the moment, we're still very, very early in how effective this stuff is, which is we've just talked about for the last 15 minutes, um, that people will not probably want to commit, um, to, you know, I'm only gonna go down this road road because, um, or as it's pre it's impressive as, as say, you know, Gemini three years, there'll be another state of the art model along in a minute.

And so, yeah, you want to, you want to try it, surely. Um, and so I think it's, uh, you know, I think people are gonna be, wanna be more, um, flexible in for, for a while yet. Yeah.

I, I feel like we're entering a new phase, um, with development in particular that, uh, it's not, you know, with a agent tech development in particular, you remember how we were talking about vibe coding that Capar came up with last year, and, uh, it's vibe coding is now just a, you know, AI coding or just even coding. We're, we're, we're, we're, we're moving very rapidly toward new, new ways of, of building software and the enterprise. And, and as we do that, like Nick is saying, you know, you're, you won't got so much tolerance for risk in the enterprise, you know, and you, you don't go beyond that.

And if Shadow AI is threatening it, it gets locked down pretty quick. So I, I, these tools are, are, you know, evolving rapidly to, to really match those requirements. And I've seen it with some interesting recent, um, you know, a agent tech tooling like we had, and I just cannot believe this name.

It's, I, I'm not gonna say better than Nano Banana, but on par with Nano Banana, and that is Bob from IBM. Bob is an IDE vtech development. And Bob, Bob, Bob likes literate programming, if you've ever heard of that.

And also spec driven developments. And those two things together, um, those two ideas are, are really informing the, the maturation of a coding with ag agentic tools right now and more. I can't believe that Microsoft let the trademark on Bob Laps, so I didn't pick it up That, because it was such a success.

Absolutely. Yeah. Right.

Yeah, but so tell us more about that, Brad. Yeah, Yeah. So it's an old idea as, as you know, um, we've been talking about it, it's old ideas come around, old problems come back around, and I think we'll talk about those in a minute with Microsoft.

But, um, you know, back in the seventies, a guy named Knuth with a K came up with this idea of literate programming where you could mix, you know, natural, um, instructions for the program with actual code and sort of treat that as a, a paradigm for developing software that was useful for both humans and machines. And what we kind of came up with from that was what everyone that's into data science knows so fondly as Jupyter Notebooks, not, not directly, but indirectly, it sort of led to that idea I'm saying, and that is now evolving and, and maturing in tools like Bob to, to, you know, sort of prioritize the thing that all developers hate the most, which is documentation. It's explaining their code.

And as we have discovered over the last few years, what, what are these tools really good at these age agentic, you know, l LMS really good at, they are really good at parsing long strings of data, making sense of it, and documenting it for us. So ideas like literary programming, kind of make your age agentic, um, experience, like I was talking about with, with Gemini CLI and the non chatty version of, of Gemini. Um, you know, it makes it into a sort of, um, how do, how do I put this sort of like self-documenting software development in which you can have this auditable sort of timeframe time slice, slice, slice of how something was created that can be audited later and used like a black box later, uh, to make software better, more secure, you know, more performant, et cetera.

So those are, those are great ideas. And the spec thing, the spec driven development is simply know what you're doing before you ask an agent, agent o tool to write your Pomodoro timer for you. And it's just start with good documentation, be clear, you know, make, make this a project, make this less of vibe coding and more of just coding.

Maybe we should talk about, um, the other hyperscaler with the big news Microsoft. Um, yeah, because, uh, it was, uh, recently Microsoft Ignite was held recently in San Francisco. Um, I was there with a couple of other, uh, future analysts.

I thought it was, it was interesting. Um, keynotes are not a, not a massive fan of keynotes, um, because, you know, they're, they're very, very well scripted and everything else. Um, but I thought this one was interesting.

It was interesting partly because who was not there, sat Satie and Adela was not there. And obviously this is, goes to part of his, his, uh, the, you know, the announcements I made a few months ago where he's gonna focus more on AI and, and also more on internal issues. And he left it all over to Judd out off the CEO of the commercial business.

Um, and it was, it was, it was interesting how, yeah, Microsoft does have some models, but Microsoft has obviously been leaning on open AI for its frontier models. It's not a frontier model, um, provider, really. Um, it's an enabler and Azure, you know, through Azure.

Um, but so, so many of the announcements, only Keynote for instance, went on for about two and a quarter hours up until about the last 10 minutes. It was all about software, which you probably think, well, yeah, yeah, Nick, Microsoft's a software company. Um, but so many things we go to, um, you know, even some elements of companies like Salesforce who've never talked about infrastructure us, I talk about infrastructure.

So it's quite revealing. I think they understand where their sweet spot obviously is. Microsoft, this is, um, it's obviously in developers but also enterprise users, um, of all kinds.

And I think that's, that was, um, that was, that was evident in the, the topics and, and the, uh, things that they, they announced. It wasn't, it wasn't till right at the end. They, they started talking about, um, infrastructure.

The other thing that was, um, unusual about it slightly was the lack of customers. Um, until sort of later on there was a couple of fireside chats, um, when, when Mercedes, uh, and a and another, and another customer and all the rest of it, they were using a fake company. They said, we have, they were, this is a made up company called Ava, and they were using their own engineers to kind of, you know, play the role of somebody at this company where they could be in, in advertising or a marketing or that co of that company.

And they're doing demos like that, which if I'm being, you know, nice and fair to Microsoft, they're obviously doing it partly 'cause it's very, very early in terms of customer adoption. Um, and obviously some of the things they're announcing obviously weren't out until, you know, weren't announced until that very day, which has also happened to be the same day, um, that Google announced Gini three. So I think it was, um, it was, it was just, it was an interesting reflection of, of where Microsoft is versus where AWS is, which is reinvent this week.

Uh, versus, versus where, where, uh, Google is. Yeah, I, I was lurking from afar, um, and, uh, from, from my vantage point, I I, I found it also interesting about, about that in how, you know, use cases. And I think it, as you mentioned Nick, it really speaks to, um, the fact that Microsoft is looking a bit further forward than, than, um, sometimes we, we get from them.

And you could see that really reflected in some of their announcements. Well, it's always, it's always give and take. The, the give part is they announced this, um, uh, preview of what they call, um, fabric iq, which just sounds like, oh, intelligent quotient, that must be really smart.

And and indeed it is. It's trying to bring, you know, some sort of semantic meaning to your, to your data state, which we could, we could talk about as much as you guys want. But sec the, the balance to that was, you know, they, they came out with their own version of a, um, you know, uh, Postgres SQL database, uh, called Horizon db, which everybody in their, you know, cousin has a version of Postgres.

That's why it's the world's most popular relational database. And the fact that Microsoft announced this as a part of fabric, which is, which they, you know, hinges on their idea of one lake, which is, you know, a, a data lake house, you know, it's like Databricks and somewhat snowflake. It, it's to say that, you know, we, we want a very flexible data platform that separates storage and compute as much as we can, uh, and leans heavily toward what we all love right now about, you know, bringing data to ai and that's unstructured data, but you kind of need to pay attention to operational data.

So, so I feel like at once they're, they're trying to leap way ahead with Fabric iq and they're also trying to, you know, be very grounded in, in what they're doing for existing customers to help bring them forward on this journey. 'cause if you just made fabric, you know, just all about, you know, the, the data lakehouse and don't really try to, you know, bridge that, you know, longstanding gap between, you know, analytics, uh, and operational data estates, you know, you're, you're, you're gonna just always end up with in-house customers and you won't have any real ones. Yeah.

Um, and then Can I just touch on the other two IQs? They, they announced one called Work iq, one called Foundry iq. The work IQ one was much more of the kind of the M 365 user, um, and copilot.

And this is, this is so pulling everything from the graph, the knowledge graph that Microsoft's been building for years, um, plus memory and context. And this is aiming at agents, you know, and if I'm working it, it'll know what documents have opened, who I've been emailing, who I've been chatting to in teams and all that kind of stuff. And it brings it all together.

Uh, the Foundry, um, is the, is a slightly different from the other two. The other two, fabric and work IQ are kind of bounded by the kind of data, um, they are looking at for, for very good reason. Foundry is the one that can kind of go anywhere else, including the web.

Um, and it's based on Azure AI search. Um, so it's kind of in some ways is the evolution of enterprise search, the great kind of, um, uh, undelivered promise of, of, of it over the last 30 or 40 years. Um, but I think, you know, Foundry is a very interesting approach and they also announced, um, agent Factory and Agent 365, the, the latter being their kind of ag agentic control plane, which is, uh, um, gonna be, I think, you know, very, very important for any, these kind of agentic or orchestration is gonna be massive over the next couple of years.

So that was the other two things I, well, or three things I just wanted to point out. Yeah. Well, thank you for that.

And, um, you know, I honestly, I I feel like we could have probably talked a lot more about Ignite here. Um, I think for me, the big news was that they didn't use Contoso as their sample company. Um, very disappointed to hear that.

Um, so Microsoft actually has dozens of sample companies they've used in the past. Um, I, I guess, um, you know, Google told us that Kentoso switched to Google Workspace a few years back. Um, How'd it go?

They, they did make that, uh, press release. So, uh, those of you who don't know what we're talking about here, uh, I don't know, Google it or something, bing it. Um, yeah, so the, the, and of course, one more thing we, we do have to mention, uh, this episode goes live during, uh, AWS reinvent.

So in fact, uh, Brad and Nick, I think both of you are at re reinvent as well, right? So, um, we will be talking about reinvent news, um, probably for a little while because we've, you know, we've given Google quite a lot of love here for Gemini. Uh, we've given Microsoft quite a lot of love for what they announced at Ignite.

Uh, but of course, um, I guess who expects that, uh, AWS is gonna sit on their hands? I mean, um, what, what do you anticipate at this point, uh, to be the big takeaways from reinvent? Yeah.

More, more of what Nick just mentioned, agent orchestration all day, every day, um, breakfast, lunch, and dinner, and with, with, uh, agent Core, I think is what they're branding it from, from AWS and, uh, it's such a critical aspect right now. Um, because as one of the reasons why Microsoft came out with Fabric iq, and it's the same thing we're getting from Salesforce, sorry, with, um, snowflake was Snowflake intelligence, and, and from Salesforce with, with their agent force, um, um, is all about how you bring semantic meaning to the data that, to bring the right context to these models. But it's not just about that.

It's about how do you actually bring together these autonomous or semi-autonomous elements, um, in, in a way that are, you know, non-deterministic but still deterministic enough to, to get the outcome you want in, in a sort of reasonable manner without exposing you to a ton of risk and without costing an arm and a leg. And that's all about orchestration. So that's what's what we expect to see a lot of from them.

And that also, I do expect to see more infrastructure. There will be more, there'll be chip announcements, I assume, um, from AWS, um, so there will be, we're dragging us back down the, um, the stack down, down to that. Um, there, I'm sure there'll be, there'll be, there'll be a lot.

I mean, it's, it's a big, as we were kind of alluding to earlier, you know, if, if you can train your own models on your own silicon, you know, you, there's a big cost advantage and I think, uh, you know, both AWS and, and Google is, uh, are seeing that to a greater or a lesser extent. If they don't have to, um, keep shelling out for, for Nvidia GPUs, then, uh, yeah, that's a big advantage for them. So expect expect more of that.

And yeah, we'll be revisiting, um, reinvent for, for quite a while after, uh, is is finished, I'm sure. Yeah, Nick's right about the hardware thing. We can't forget it, can't we?

Because, uh, apparently Oracle was right along with, with Exodata and, and vertical integration being kind of a useful thing. And we've seen it proven a time and again, with, with Nvidia, um, and that is that, you know, vertical integration leads to optimizations that that lets you do a lot more than if you're just white boxing it all day with just scale out, you know, for trying to solve your problems. And with AI in particular, you know, there are a lot of bottlenecks that show up in places like networking and in places like storage.

And, and so that's why we see all of the, you know, networking and hardware, you know, storage server, et cetera, vendors all leaning really hard right now on creating this sort of integration stack of, of optimization. Yep, absolutely. And, and, you know, for that matter, um, you know, we just, uh, last week heard about Nokia announcing that they're gonna be the, the, uh, providing networking, uh, connectivity for AI models.

Uh, they're investing $5 billion to have made in America networking hardware to support these things. I mean, there's, there's all sorts of, um, investment happening down at the infrastructure layer. And, um, as Nick points out, I definitely, that, that's my focus, so I'll be keeping an eye on that.

Um, we do have to wrap this week though. Um, so thank you both for joining us. Uh, before we run though, since again, it is, uh, reinvent week for our, our listeners when they're listening to this, where can they find your coverage and reactions to reinvent?

Uh, Brad? com, et cetera. Alright, Nick?

Yeah, same. I'll be, uh, LinkedIn apnic patients. I, I, I don't use XA lot, but I'm at Nick patients on that.

com. We'll be, we'll be writing a notes, um, uh, for, uh, for, for our clients and everybody else on what we see at Reinvent. Yeah, absolutely.

And, um, yeah, I'm gonna be watching it as well. You'll find me at s Foskett on most social media networks. Um, I'm, I'm using Blue Sky a lot lately.

Uh, maybe you head over there. Uh, thank you everyone for listening to this episode of the Utilizing AI podcast. If you enjoyed the discussion, please do subscribe.

You'll find us on YouTube as well as in your favorite podcast application. Uh, and also we'd love to hear from you if you wanna give us a rating or review. Uh, this podcast is brought to you by analysts and experts from the Futurum Group, where Insights meet ai.

ai, uh, the utilizing AI YouTube channel or the text TV app. Thanks for listening, and we will catch you next week.