Google Remy AI Agent, NVIDIA Networking, SAP AI Push

Hey everybody. Welcome to the latest edition of the Techstrong Gang, and we got an awesome lineup of all kinds of fun things to talk about, but let me introduce our guests. Let's start with Fred Wilmot.

Fred, good to see you. How are things? Good to see you, Mike.

Great, sunny Seattle. Awesome. Don't hear that very often, but maybe this global warming thing will change that, so we'll see.

Yeah. Alastair Cook, long time. Haven't seen you in a while.

Where are you at the moment? Well, I'm here at home in New Zealand, so I'm not sure if it's going to be a sunny day today. Look at this, it's- ...

well, it's probably nighttime where you are, I think. It's not morning. But yeah, here we are, Techstrong Gang.

We're global, man. We're everywhere. All right.

Gina, always good to see you. How are things? Good morning.

It's good to see you too. It's actually kind of chilly here today, and that's why I have a long-sleeve shirt on. All right.

And John is, if you watched the show yesterday, still in Las Vegas in a room- Yeah ... that suddenly I was able to pick out and identify the hotel just by the decor. So.

Yes. Yeah, it's going to be the first day of 100 degrees in Las Vegas is tomorrow. It's like 95 today.

Ooh. Welcome, John. You're welcome.

Yes. So, I am engaged in a bit of nostalgia just because of the way my mind works, but if you're of a certain age, you remember a show called "Happy Days," and there was a fella named Chachi on that show, was played by Scott Baio. And later on, he went on to do a show called "Charles In Charge," which was basically he was the personal assistant for this family, and that brings us full circle to what's going on here.

Google apparently has ambitions to deliver a personal assistant to folks. There's reports about this, and we're all going to get one, and it will be on our phones, and will be with us at all times. And maybe my personal assistant will negotiate with your personal assistant.

We'll see how that plays out. But John, what's going on? Yeah, so you mentioned the reports.

I think it was Business Insider first reported it. They had access to some internal documents, and as you said, Mike, it is kind of like Chachi, or "Charles In Charge," I'm sorry. They used to call it Chachi In Charge, too.

This thing is called Remy. And I guess the idea, it's kind of like OpenClaw. The idea is to own a personal agent layer.

This is what Google wants to do. They want to own a personal agent layer through the native integration with Gmail, Calendar, Drive, and it's something that standalone providers can't replicate. And this focuses on the ability of an AI to navigate websites and apps to achieve certain specific goals.

It's interesting, so the name Remy comes from the Latin word remigis, which means oarsman, which means doing heavy lifting for users across whether they're work, at home, or at school. So it's built to operate with minimal supervision. I think I'm just going to read a couple of its capabilities.

Summarize emails, draft replies, organize calendars, navigate Google's ecosystem, including, as I said, Gmail, Docs, and Search, and analyze user behavior patterns to anticipate needs and execute multi-step workflows. So it is an interesting concept. It's in the dogfooding process right now or that aspect, but I really look forward to seeing what this can do, and I'm also kind of somewhat spooked by it as well.

There you go. Gina, you ready to have your life run for you by Google? What do you think?

Heck no. I've got all that shut down on my devices. So I feel good to go.

But I think it's interesting because it's not just the Google Workspaces that are involved, it's also Google Home, which I do use because my thermostat is a Google thermostat, and some of the other acquisitions that they've made in the IoT kind of area. So it's something definitely for me to pay attention to. And the one big question I had is there's also a story going around how Google is silently installing a 4 gig AI model on personal devices.

Ooh. So is that in anticipation of launching this? And it makes sense to me kind of architecturally that it would be, but it's also Google...

It sounds like it's a great deal because, oh gosh, I can connect all my things, and I have automatic, and blah blah blah, this sounds great. Aside from the privacy issues is the issue that you're now going to have to have a device that's beefy enough to, number one, have enough space for all your pictures and videos to install this AI model, and then to be able to run it. So they're pushing that downstream to the consumer.

Well, that was my question. I too saw that report about the- Mm. LLMs being downloaded surreptitiously.

It was a researcher who found this, so I don't know if we've confirmed that just yet. But Fred, in your mind, is that what's going to be required? We're going to need LLMs on every device out there in order to make these personal assistants kind of run?

And I'm going to assume that Google's not the only one building these things no matter what we do. Yeah, I think it's obviously a race to the data you have. Right?

And you've got Google's Remy, and you've got OpenClaw, and you've got Anthropic's Claude, you've got Meta's Hatch. Right? All with different sort of ways of approaching this problem.

The worry for me is a standard operating procedure for Google, Microsoft will follow suit, which is, I didn't ask you to do any of this, and I'm not interested in you doing it. Right? And so the privacy part here far outweighs my concern.

The level of access to details that I choose or choose not to share versus the value. From a value perspective, I'm not sure I'm going to see a whole bunch of value delta here from these other three ways of operating. And the difference between what we do with all the technology, but it also has a notion of trust and an open call on every derivation of somebody's building their own agents.

You lose, and I don't mean to overload this term, you lose your own agency in this kind of a conversation, and that's my big concern around that for sustainability and safety. I don't need you loading models that I don't have access to or can't take my data out of. So if you want me as a user in this particular case, then I have heavy opt-out, heavy transparency, and I need to be able to vacate my data in any model not used for training and all the above.

And I don't see that anywhere in this approach today. " Most people are going to just shrug. Most people aren't going to think about how much of their data they're exposing and whether this is a good thing to be doing with that data, whether they're getting to choose what model, what company is using their data.

We know this. People give away their data just in order to be able to look at cat photos, and this is no different. It will continue to be people, where the majority of people will just hand their data over, and it falls to us within the industry to actually hold some light to the governance and the controls.

And as Fred says, the ability to opt out and opt fully out of it. It'll be on us to make the noise around this and to hold the vendors to account. I feel like I'm going to get shamed into it, to be honest, because I think other people will start using these things, and then they'll start- Mm ...

hitting me up for the fact that I don't have one on my phone or whatever, so they can't book a meeting with me and have these conversations about things that I'm supposed to show up for that I didn't maybe overtly agree to, but because my AI and their AI came up to some sort of agreement, the next thing you know, I'm off to Las Vegas to have a meeting with John. What do you think, John? Is that the way the world is going to work?

Yes, it's like the fear of missing out, you're going to be shamed into using this like everything else. See, the one thing I'm thinking about is the governance issue. So I'm wondering how enterprises are going to respond to this.

So their employees are using these personal agents that touch their corporate email and calendars. I wonder if that becomes an issue. I wonder about the ethical and user experience challenges of AI that executes these workflows autonomously with minimal supervision compared to somebody who waits for a prompt.

I think about all these things that are going to escalate and where it's going to lead us. It's true. " So...

I think there's two things here, Mike. One, as you mentioned, I think that a lot of people that don't have the enterprise class problem space could adopt this and probably be okay with it. But the truth is, again, this is a land grab.

It's a race, and you can do any of these things. So you're not going to miss out. There's no FOMO here.

You can do any of this with any of these other agents, no problem. They'll have the same access, broker the same way, only you control them at your behest. And I think the big deal is today, this year, 5% of the workload gets done using agents and things like this.

Next year, it's supposed to be 40%. So it's really a battle about which agents are going to do what work. And when you take away my options, I love it a lot less.

And so for enterprises, enterprises are still going to have the same requirements, the same governance, AI governance stipulations, and privacy that they have to be adherent to for all the regulatory purposes. But the average user on their phone, I think Alistair made this point really well, is that they look, what makes it easy for me? How do I use it, and how do I just make forward progress here?

I don't need to figure out open... I'll just turn this on. It's fine.

Or more likely, I just won't turn it off because I won't pay attention well enough to know what the implications are. I think people- Oh, sorry. Go ahead, Gina.

I'm sorry. I think people will start noticing if they've got a full phone, which is everybody I know that is not technical. Full phone because they've got videos and games and everything else on it.

So all of a sudden, their game is too slow, or they can't watch videos like they wanted to because the infrastructure required is now pushed down to the users. I think that's when people will be starting to notice, oh, what's going on? And they might think about, for two seconds, well, what do I delete?

Or do I want to delete anything to have this running? So I think that's what will happen. I can see it working more like this, I think.

Think I'm going to have a personal assistant on some device that will be running alongside my, quote-unquote, business assistant that I got from my company, and the two of them will be in their own little sandbox and kind of talk to each other and share things back and forth within appropriate sets of permissions and policies that are going to be probably overly broad and overly permissive anyway. But maybe these two things will be isolated and instead of me having maybe two separate smartphones, one for my other real life and one for work life, we're going to just have these assistants that are carved up for special functions and tasks. Alistair, what do you think?

Yeah. One of the funky things to this to me is the idea that the agent, the AI, is running on my phone because my phone's not that powerful of a device to run all of my AI. As Gina says, it's probably full of those cat photos and videos, but also particularly if you're not running a flagship phone, it's not as powerful as you might think, and spare capability to run the AI, run these agents for us.

If you've taken a look at what's going on with some of the coverage of actual agents doing things, they burn through a lot of tokens. They need a lot of compute power to complete the tasks because agents take iterative loops to complete tasks. So I see the agents running somewhere other than on the phone.

I think the AI on the phone may be good for my user interface, but not for actually running the agent. So yeah, theConnectivity between the agent me that is working and the agent me that is part of my family and the agent me that is part of a club that I'm on the committee for should probably all be able to communicate and have different sets of rules and governance around them as well. Do you suppose instead of the phone, we've heard about the Jani Ive experiment with OpenAI.

Google's a hardware company. Perhaps there's a device in the future where your AI agent runs independent of your phone. Is that crazy or is that something that...

This is something I'm wondering where it's going to head eventually. I think it may start out that way, but eventually shows up on something that feels like a phone. And are we surprised that there might be a downstream conspiracy effect here to get us to buy new hardware and upgrade our phones?

That's what I'm thinking. Part of the purpose. But how's that going to work right now when there's no memory anywhere to do anything?

So how is that going to work? We're talking a longer, maybe a couple of years. I don't know.

Maybe something crude. To me, it's an opportunity to tap revenue. And if I'm OpenAI, I'm thinking of every conceivable idea because I'm under so much pressure to magnify the revenue stream.

And- I think, and maybe Fred, correct me if I'm wrong, but this could be done in a highly distributed fashion, right? So the device is not all that smart, has a little bit of memory, and it's talking to something that is at the network edge down the block from my house, and then that too is talking to something that is in a pop somewhere, and then finally into a cloud server. But does this all become one giant distributed workload?

100%. I think the understanding of distributed computing is pretty well established, and so how best to opportunize what exists, right? The hundreds of millions of phones versus the difficulties in AI data centers is a very easy way to look at a threshold and a balance to figure that out.

Again, the question being, where is my data, air quotes, being work loaded for real? And then how do we understand how to manage and/or mitigate that? And I think that's an interesting conundrum.

So if you own a phone, right? And Google owns Android, right? So if you own a phone, then this becomes a very obvious way to obviate that problem.

And I think what'll be interesting to see is how does Apple respond and whether or not other manufacturers are going to look in like kind. And in the next couple of years, you're already starting to see this now, the delta on electrons versus photons is going to be another significant challenger into how and why, because the economics of this is not going to scale over the next handful of years anyway. It's interesting that you mentioned Apple because they now have a CEO, well, as of September 1st, they're going to have a CEO as a hardware engineer.

So maybe that kind of signals to us maybe potentially what they might do in this whole kind of scheme of things. Alistair, are there security issues to go with this? Because we've already seen prompt injections, and the next thing you know, I've got this personal assistant telling me to go somewhere, and then I somehow or other wind up in a back alley getting mugged by three guys.

Wow. Well, the thing is, you don't even need to be in the back alley to be mugged by three guys because you're going to link this thing to your credit card and all of the things that are of value to you in order for it to do its work. In the same way that you have access to your credit card and your bank balance, your agent is going to have access to these things in order to pay for those flights for you to get to Vegas to spend some time with John.

In exactly the same way, it's going to have access to your entire life. For it to be useful, it has to. And yes, that's going to be terrifying.

Having looked at some of the fun that's been had with current generation AI tools, yeah, prompt injection, and basically the fact that agents have no idea the consequences of their actions as yet. Fundamentally, that's our problem here. There is no understanding of consequences.

It's like handing things over to a 17-year-old who cannot even think about how tomorrow is going to work. There will be big security issues with this. Circling back to where this is going to run as well, you've got to think about where your data is.

How do you run this AI close to where your data is? And the two places my Google data is is on my phone and in Google's data centers. So one of those two places is the logical place to run this AI.

Mm-hmm. Yeah. I kind of like Google having some connection to enterprise thinking, but it's not been as strong a connection to enterprise thinking as maybe some of the other enterprise vendors that are in the space.

Nobody's mentioned Azure yet, and I'm sure Copilot will have an agent that will do this for me too if I'm on Microsoft 365. The part in this that I'm not so happy about, and it just comes back to the same divide I have in my house, and probably everybody else does, too. I got half the family on Apple and half the family on Google, and of course, if we want to have a conversation with each other, we got to use something like WhatsApp.

So if we have all these personal assistants going on, maybe this should be a third-party application rather than something provided by the people who are giving me the hardware, because otherwise, these things don't interoperate as well as they should, and I wind up getting this kind of forced conversation. Or does everybody in my family have to agree to kind of be on the same hardware device so we can have one Charles in charge of the whole family? I don't know.

John, what do you think? That's an opportunity, Mike. There's this, we create this whole ecosystem.

We get the hardware device, we get the third-party app. The possibilities are endless. So, yes, because I have the same issue that you do in our family.

I'm the only person who doesn't use an Apple device for my laptop. We all use iPhones, but yes, it's inevitably going to move down this way. It's this huge opportunity.

Mm-hmm. Wait, didn't Apple sign a deal with Gemini? Is it really a deal now?

Yeah. Is it- You're right ... just everybody- They did.

Okay. Maybe that's what they were thinking about when they signed the deal, because they know what they're going to be doing a few years ahead of the rest of us, but all right. Gina, you started out with a hard no on this thing.

Has any of this changed your mind? I know. I'm a hard no.

I'm worried about the few little things that I do already, and this makes it harder for me to understand how to control that. So, no, I will not participate in it here. For my business, I do use 365 already, and so that stays in the business.

One of the things I worry about too is anything you use this for, anything that then has a record of what you've been doing becomes discoverable, and there's been lawsuits- Mm ... about that already. So there was a lady that was not allowed to take her pregnancy leave, which was wild to me.

I didn't get the whole story, but she wasn't allowed to take leave because they subpoenaed all of her digital therapy that she took via the company's therapy, and all that she had talked about with this digital therapist was discoverable. And so that led them to deny her getting any kind of help or getting that leave. So, that worries me a lot, too.

So if we give up all of our life and make all of our life digital, at what point, if it's digital, it's probably discoverable. So if the laws aren't keeping up with where we're at with AI, and we've got to protect ourselves a little bit. So yeah, hard no.

I'm going to end this here because, well, it can just go on forever, but I would point out one thing. There sure is a fine line between being my assistant and just generally being a nag that I don't want to hear from anymore. All right.

Let's shift a gear here. NVIDIA this week was talking about a manufacturing agreement with Corning, who makes all that fiberglass out there and is going to help build some interconnects for AI, for networking, which we sorely need. And there is a question, though, about how all these networks are going to get built going forward, when we need to connect all these things, because, as I was talking about with Fred, eventually all this stuff becomes highly distributed.

But Alistair, correct me if I'm wrong, but the networking infrastructure that we have in place today really isn't up to the task for any of this. So is this the beginning of maybe a reinvention of networking as we know it? Or what's going on here?

Well, I think one of the things to have in mind is there are a couple of different networks going on here. One of the networks is AI specific, the networking that runs inside the AI data center and connects all of those GPUs together to transfer data. But there's also the connection from that data center to the rest of the world, and been hearing a little bit of a rumor that there's a shortage of the fiber that connects between the buildings, let alone within buildings.

So, cross-country fiber, a little harder to get hold of than it was a little while ago, and some of that is simple manufacturing, and that's why we're seeing a support deal, a commitment between NVIDIA and Corning to make sure there is enough fiber available to actually both fit out these data centers where connectivity is moving towards fiber, but also connect between these new data centers. A lot of new data centers are being built or planned to be built. Some places, more data centers than have ever existed in the world before are being planned to be built just to host this AI boom, and so connecting those data centers to the wider internet is going to be an issue if there's no fiber available.

And connectivity within those data centers that's transitioning from being copper cables plugged into the back of the GPUs to being fiber cables plugged in. Those two transitions are both driving this shortage of optical fiber, and Corning, as the main manufacturer for this optical fiber, is needing to expand its capacity. Interesting to see that this commitment is around US-based manufacturing in particular.

That's probably a reflection of the tariff environment that's been going on, but also the desirability of getting more employment on US shores. They're talking about 3,000 jobs being created here. Mm.

The complexity and the connectivity inside the data center between the GPUs, historically, this has been NVIDIA's own proprietary NVLink technology that's the most well-known, but of course, that's only accessible to people who have NVIDIA hardware. The UI Link consortium has been building a competitor to this, and that's planned for release sometime. Sometime we'll be able to use something other than NVIDIA's technology to join together our farms of GPUs over good optical interconnects, but we're not quite there yet.

Definitely lots going on in this often overlooked part of how you get access to AI. John, I'd love your opinion on this because you've been following NVIDIA, because to Alistair's point, these guys are building a highly vertically integrated stack that goes all the way down in the network, and are we really setting ourselves up here for basically living in an NVIDIA world because everything from the interconnect up to the GPUs and the software stack that sits on top of it is all going to come from one place? Yes, exactly.

So NVIDIA keeps investing across the stack, and that makes me think about does that create a de facto platform tax for everyone else, or is this just smart scaling? And is this partnership model with Corning, is this like a new phase of vertical integration for NVIDIA? Those are the two things that come to mind based on what they just did here.

I don't know. I don't know. Fred, what are you thinking about?

I think out in Seattle, I'm sure that people talk a lot about NVIDIA, and Amazon, and Microsoft, and all these folks. " I think the challenge here is that somebody's got to do it, in the same way that we thought about the expansion for telecom. And so there's a reason why all the internet hotels are named, May East and May West, and so on and so forth because Ma Bell and the things that happened in the early cable laying days.

Now, somebody's got to lay some more fiber in order for this to happen. And I think it's appropriate for whomever is going to be able to have the use case and also the resource to do it, should do it. Will there be a case down the road where we think through deregulation as the function or allow other people access for these things?

Yeah, absolutely. The question I have is whether or not this is not a commercial requirement or a federal requirement. Because some of the conduit behaviors that happen for accessibility, privacy, and the regulation behind that also have to be part of the strictures of we have state governance around this and federal governance.

But in order to avoid the philosophy of the robber barons and the railroads and the early days of AT&T, and the copper tax that we all paid for long-distance phone calls and the ridiculous like that, does this become the same thing? I think those are optimization problem questions. Right now, capacity is the big deal, and it doesn't exist.

And so, I'm just happy that somebody's going to invest in laying the groundwork for the rest of these pieces here. Somebody's got enough optics to see that this is a problem right now, and it will be much bigger in the future. I don't know.

Mike, you mentioned, and I'm going to take another crack at what you asked me because I'm also thinking about NVIDIA and what they've been able to do and navigate during... And again, tech is a lot like politics. It's all intertwined.

And I think of them, they don't project this kind of ominous, dark, evil empire that Google does, or Microsoft to a lesser extent, Amazon definitely, Meta. And they are slowly but surely, every year they're a trillion dollars richer in terms of market value. And they are, as you're saying, across the stack, they are becoming this huge dominant play, but the way they've played it, where they work with everyone and they seem to play nice with the administration, they've become this behemoth that no one seems to be overly concerned about or terrified of.

And I give them, I'm going to give Jensen in particular, a lot of credit for being able to pull that off given the big tech attitude we have now, mentality. I don't know. Just from having experience at my former employer dealing with NVIDIA trying to co-produce assets when our approaches to especially virtualizing GPUs were very different at the time.

They play hardball. Mm-hmm. And nobody stomps on their message.

Everybody plays their message. You go along with that, or you don't get to play. So, coming from that kind of collaboration viewpoint- Oh ...

it's very obvious. Oh, yeah. No, I mean in a general sense.

Oh, no. But- Like in the general public sense is that NVIDIA's considered this great success story, the hot company, but they're not saddled with the baggage of a Zuckerberg or Bezos or even a Tim Cook at this point, to a lesser extent. And, I find that fascinating how they've been able to pull it off.

Because of meetings behind the scene, I'm telling you. But that's why we have the consortium as well, is to keep people from being locked out. Because you think of some of the other bigger companies in the past, like they remind me of Sun, and I don't remember how Sun was, but I know that Sun was all locked in.

It reminded me of another former employer, EMC, an awful lot in the marketing, in the positioning phase of things. So, like this is the only solution. This is how we're going to get to AI.

This is the only way to do it. When in fact there's a lot of alternatives, and when alternatives popped up that actually looked pretty good, they gobbled them up, and they were made to fit into that narrative. So I can see it from that inside view a little bit more than the outside.

I get what Fred is saying. We need the tech in there at the forefront of this thing, but we have learned the hard way, or maybe not learned, but have seen multiple times over, including recent times, where IT becomes overly dependent upon some single vendor somewhere who then turns around and either demands a massive share of the total spending or threatens to increase licensing fees because, well, they can, and there's no counter. So Alistair, are we incapable of learning or what?

Well, yeah. We always repeat history. This is basic human nature, and I've seen both what Gina is describing of NVIDIA being very much controlling.

It's our way or the highway. I saw that from how they ran their GTC conference, where the vendors were told, "This is the booth size that you get. You can't give us more money to get a bigger booth.

"But also from the popular love that John describes of all these gamers who love their NVIDIA GPUs, have grown up to be the IT executives who are absolutely embracing that NVIDIA is now a big part of their business, and they love the success. I loved seeing Jensen standing up at GTC last year saying they're a trillion-dollar company, and this year saying they're a $2 trillion company. So this definitely has those two different sides.

The very public side of the industry loves NVIDIA, NVIDIA's doing great things, NVIDIA's partnering with almost everybody, except competitors. They really want to lock out competitors, and that's what the UA link's all about, is about moving away from proprietary architectures, moving away from things like having what was originally InfiniBand was the connectivity that's become NVLink. And that was always a niche piece of connectivity, but then NVIDIA acquired the one manufacturer in the space, and it all became their technology and not really available for anything else.

If you want to build yourself a brand-new supercompute cluster, you're probably not using InfiniBand anymore. You're moving towards high-performance Ethernet or UA link. So yeah, the sort of moat that is being built around NVIDIA is that everybody uses NVIDIA, NVIDIA is vertically integrated, NVIDIA is all of the things.

But maybe NVIDIA is not the right solution for all of your problems. Only NVIDIA wants you to solve all your problems with their hardware, like any vendor. All right.

I will just end it here and point out that just because the handcuffs are made out of gold doesn't mean they're not handcuffs, right? All right. Ian, let's continue on with this theme, though, because if we get to our third topic, we're talking about lock-in in a lot of different ways and forms today, and the latest and greatest in that path or trend is, well, SAP has turned around and bought a company called, I think it's Dreamio.

Is that correct, John? Yes. All right.

Yes. And they're a provider of a data lake. And then at the same time, there are reports now that SAP is trying to limit who can access their APIs because, well, they're afraid that all these AI agents will just pile in on the data that is stored in their platforms, and they won't get compensated for that.

And part of that issue goes back to these conversations we've been having about SaaS apps in general and AI agents because, well, theoretically at least, if we use an AI agent to build an app or a process, it will just pull data from wherever it finds it and turn all those SaaS applications into a headless service of some type. And I think SAP doesn't necessarily want that to just happen. Gina, is any of this sounding familiar to you?

And I can't help but wonder if, as SAP tries to build a moat around its APIs, maybe that's just the first place that they're going to retreat to, and maybe the last line of retreat is they're just going to hold onto our data in the data stores and say, for lack of a better phrase, "It's a roach motel. " Oh, wow. That's a visual.

And that's scary, right? Because when you're talking about large language models, and being able to access and make sense from the lots and lots of tables that enterprise data exist in the enterprise data, you want that to be open so you can get in there, get the right information, mix it with some new information, and figure out what's really going on there. So, the roach motel is very scary, and the locking you down is very scary.

So, the Dreamio, is basically a way for them to have access to Apache Iceberg, to a different type of data lake so that they can... Since SAP concentrates on all those tables and spreadsheets and databases and everything else, they want that business data to be in a data lake where it's easier to access and easier to use AI on it. And they also acquired another company, let me get the name.

Prior Labs. Prior Labs. Yeah, thank you.

Which they work on the tabular foundation model, and that is an LLM, but not an LLM. The model is more built on this tabular data that exists in most of the world's business data. So they acquired them as well.

So now they have a way to have this data lake that will accept the tabular data. They have a tabular data model that they can apply and start using that on the world's data. So they think everyone's already locked in to SAP, so they should just keep them there, but make their work easier for them inside of SAP, so they don't have to go anywhere.

But then the problem is, people that have their data inside of SAP are using different types of models to get their data out, and they're using the APIs to do it. And I know I read an article yesterday, and I went looking for it and couldn't find it. It's not just SAP.

There was a couple of different organizations that are saying, "Okay, wait. We're not going to give you free access anymore to just come in here with your little AI tools, grab all the data out of this, and take it someplace else to process it. You bought us.

We are the data company that does all the tabular data. We're going to keep you here, and you're going to stay here, and you're going to love it. " So it does come back to is that the right way to go?

And will an old-school company like SAP, with all of their knowledge and their understanding of how enterprises use this type of dataWill they be able to convince their customers, "Hey, you really don't want to go anybody else. We've got the expertise. We've now got the right kind of model.

" For me, I just look at it and think about the data cleaning efforts, and I know how salespeople are, and I know how many ways they try to spell different names so they can hide their stuff. They're never going to change the humans, so I don't know if there is a way that AI can help with that, but the data's going to be dirty as long as there's people touching it. " At some point, do customers just not look at that and go, "Wait a minute.

" Because, well, it is the customer's data after all. Yeah. It's also a risk-reward, right?

So SAP is not alone in this, but by the time SAP decides they have a policy decision on how to take over AI as controlling the AI path for all the data they have, but also partnering with rivals so they can help control the way that AI interacts with all of those things, by the time that SAP has arrived at that place, you have officially basically thrown all of the meat to the wolves. And so the expectation here is clearly the same, that most people that have invested in SAP have already invested a ridiculous amount of money in time, implementation, and the data structures around it. And it's probably not small amounts of data, right?

If you're using HANA or whatever, now with partners, you'll have access to Databricks and Snowflake and the other sort of really monstrous warehouses. And now you have some stipulations on rules about how you get access to that, right? This is going to put a lot of pressure on the ecosystem about how to interoperate.

And Salesforce is doing this. Snowflake is going to be doing this. They just don't have the sheer magnitude of capacity and capability yet on all of those fronts.

So it's very similar in my mind to what Google's doing. SAP, super deep. Now they're going to go broad, and they want to control the strictures of the way that you interact across this broad horizontal view.

And it is what every data lake/data storage is the race to the AI layer across them all, to rule them all, the one ring to rule them all for Alistair's benefits. Alistair, what's your take on this? Is there going to be a free my data rally down by the courthouse, and everybody's going to come with their protest signs or what?

Well, you'd hope that my data belongs to me, even if it's stored on SAP's cloud service. But maybe, yeah, there's definitely going to be some protests around this. But I think this is the beginnings of SAP saying, "Well, we're actually going to run your agents rather than using external agents to access the data that's inside SAP.

" And some of the dialogue in here is around the fact that access to your data is much more of a restriction. The readiness of your data and the understanding of your data is much more of a restriction than what language model you're using or who you're using to run that language model. And so SAP is positioning itself to be the organization that knows how you're storing your data because, well, you're storing it in their system and around their structures.

And so their agent is going to be the smartest agent to work with that set of data. I think whether the customers will accept that, that will be an interesting thing to see over time. I think the ability to simply access the data directly, rate limits, quotas, monitoring, that typically reflects the fact that there's a cost to this.

Presumably, I can increase my rate limits if I pay you more money. That will be very much an SAP way of working. But I think also shifting to, I'll pay you to run an agent that gets the insights out of this very specific data, rather than using an inefficient large language model to access the structured data, because LLMs are good for unstructured data.

They're not great for structured data. We'll have a different type of AI model that is designed for the specific structured data that understands the structure we already have. And that's definitely some of the point about the Cryo Labs acquisition is to build the AI that knows about SAP applications, and therefore the agent that runs inside SAP.

So I don't think this is complete yet. I think we're still waiting to see SAP say everything's going to be agents and access all of your data through our agents. You mentioned Salesforce before.

They absolutely are on board with we're not so concerned about selling licenses for humans. We're very much about agents are going to do the things. I think SAP will follow the same path.

John, are these vendors basically positioning themselves now to say that I am a friendlier totalitarian regime than the next guy, and that's why you should pick me over them? And is that just kind of missing the whole point? Yeah, I think there's this theme here I'm starting to see like we have a genuinely open data and agent platform for you, but what we're doing in the case of SAP is we're tightening the system of record mode.

Yeah, it is. It's a lock-in, the fear of it. We offer the all-in-one solution.

We'll make it easier for you, but then once you're locked in, you're locked in like a jail cell. There you go. Fred, can I program my way out of this, or is this just the way the world is going to be?

I think it's the way the world is today. I think we're just going to see another evolution of that rather than a revolution, and remains to be seen whether or not revolution is actually possible here. But there's an interesting thing here, just as an aside with what SAP is doing, is the whole notion of zero-copy data sharing.

So wherever your data lives, that organization is not going to get punished for you using SAP to do the things. It is really the semantic layer they're really trying to layer on here, and I like the way that Alistair described it, too. When you think about it, if I have a company that does the notion of tabular models, and I then basically create a dictionary of all of the other resources you have data in, and I maintain sort of the best way to do it, it's how search works today.

There's nothing unique about that from the standpoint, but it's the eyes and the fingers that will do the talking and the walking. So I'm anxious to see what happens. All right, folks, I'm going to leave it here because I think in all these instances, it's pretty clear that if you give up control of your data, you've given up control of everything.

So need to think long and hard about where that data's going to be and how I'm going to access it and who has control thereof. Folks, I want to thank everybody for sharing their knowledge and insights. Fred, Alistair, Gina, John, as always, something to think about out there.

And meantime, folks, thank you all for watching the latest episode. tv lineup, and we'll see you all tomorrow.