Linus Pushes Back, Google Goes Agentic and Open AI Gets Bigger

Hey everyone, happy Thursday. If you don't know me, I'm Alan Shimel. I'm the CEO here at Techstrong.

I haven't been on in a while because I've been on the road all the time. But it's good to be back on the gang and to see some nice, familiar, friendly faces. Let me introduce you to our friendly face gang today.

Who could be friendlier face than this man, Fred Wilmot? Fred, good to see you, man. Happy to have you here.

Great to see you. Speaking of happy faces, she always has a smile on. Okay.

She's in Austin, not Houston. It's Gina Rosenthal. Good morning.

Gina, welcome, and thanks for coming. Joining us from, he's still up in Amsterdam smelling the tulips, I guess, or whatever it is they do in Amsterdam, but- Smelling a lot of different things up here, yeah ... Yeah, I would imagine.

Our man about town, John Swartz. John, how are you? And back from the streets of Minneapolis, in his high castle, our chief content officer, Mike Vizard.

Hey, Mike. Good to be back and good to be home. It's good to have the gang assembled here and be part of it.

Before we start, just a big shout-out and kudos to our friend Steven Foskett, who actually hosted the last couple of days. I was afraid I was going to be Wally Pipp. But- ...

good for Steven, and thank you for sitting in and doing a great job, as always. Mike, what do we got for today? Shout out to John because he's been doing double duty.

John too. John's on vacation while he's doing this. I know.

Yeah. Well, I can do this before I go to the pub, so- That's all right ... it's all good.

But- See ... yeah, no, we missed you. We missed you, Mike and Alan.

We missed you a lot, and this was mentioned on every show. So we're glad you're back. I think you got the thing a*s backward.

You should come to this after you go to the pub, and then we'll see what happens. I would be interested in next time. I was thinking of putting it here where the water was, actually.

That would be nice. Yeah. Should've done it.

All right. Why can't we get good European beer here in the US, by the way? We can.

You've just got to go to know where to look for it. I don't know. All right.

I've never found it. We just can't- Mike, I can see where this is going. Let's try to stay on focus here.

Let's start where this been. All right. So Linus Torval was in Minneapolis for the Open Source Summit.

And he usually shows up once a year or so to give his thoughts, and he's been vocal about what's going on with vulnerabilities in Linux, and John and the gang talked about that yesterday. But he said something out loud that I think most of us are thinking, and he has a love-hate relationship with AI, and he's generally positive about it, but there are things that are bugging him, and there are little things that are issues, whether it's the fact that everybody's reporting the same vulnerability over and again, and other folks who are saying that AI is going to have this massive negative impact. He was generally positive on the outlook of AI, but there are these things that everybody seems to be finding annoying.

And also we saw, I think it was late last week, when there was a commencement address, and they started booing the speaker about when he came up because he was associated with Google and AI, and it's kind of like, hey- Well, he mentioned AI too. Yeah. Every time he mentioned AI.

It was like a drinking game. Every time Eric Schmidt, who used to be the CEO of Google, mentioned AI, he was booed. And it didn't dawn on him, and so he stuck to the script, and he was amused by it, but it played out really badly.

Yeah. Yeah. No, I wrote an article on this, though.

Do you blame the people for booing? Look, these are young kids who are graduating school that can't find jobs. It's one of the toughest job markets for recent college graduates in memory.

And they're being told at least one of the reasons is that a lot of these companies are no longer interested in hiring junior people, junior developers, junior marketers, junior whatever. And so that contributes to them not having a job. And at the same time, you got these tech bro oligarchy billionaires that are not likable people.

You know what I mean? Did you, by the way, what Marc Andreessen just said? Yes.

He said they preferred AI agents to humans because they don't get drunk, they don't miss work because they're sick, or they don't have personal problems. Blah blah blah blah. Of course.

But what, you don't think they all feel any way? Let's face it, you look at the current state of the oligarchs versus, let's say, the robber barons of the early 1900s. These people are not nice, likable people.

They're egomaniacs who are consumed with amassing wealth. And this generation, these Gen Zs or whatever you want to call them, they don't like that. It gets better, though, because this week, too, there was a CEO who was talking about their financials and said that AI was going to help them, I guess, not need as much low-value labor, is what he called it.

And then, a day later, he winds up walking back those comments because basically everybody started pounding him and the company, and there's a lot of folks saying, "Hey, if you're going to have that attitude, guess what? " So it just- Let me try to bring this back to Linus, though, Mike. Yeah.

Well, the thing with Linus is, though, I think we're all in the same boat here, right? Despite what everybody feels about AI, for better or worse, everybody seems to be using AI, even when they don't particularly like it. But I don't see anybody not using it per se because they have this issue Well, to me, and again, it goes to that article I wrote, it comes to this article that Joe Jackson wrote, AI is the technology everybody loves to hate.

But, as you say, doesn't stop them from using it. I think it's very hard to argue with the benefits it provides and can provide, and where we could go with it. But to a certain extent, we hold AI up to a much higher degree of perfection than we would, let's say, a person or someone else, right?

We saw it with the AI coding. The average person writes code, and for every X hundred lines of code, we get X amount of bugs on average. And when AI first started, it had 10X amount of bugs, but as it got better, at some point it crossed below the X amount of bugs per hundred lines of code.

So it was actually coding better than people. But that wasn't good enough because we want AI to be perfect, and that's the only way we're going to be happy. And in Linus's case, it's a case of, damn, this would be a great job if only those damn customers would go away.

Right? I would love to make Linux more secure. They're clogging my feedback loops with bug reports, though, and I got too much to do to really focus on it.

So hang on. I get what you're saying, but I don't agree with you. You don't have to agree with me.

I don't. That's what makes it interesting. Well, let me tell you what I think.

I think some of what Linus said was, yeah, there's a ton more bug reports now, but nobody's doing the extra piece. And this is a junior-level thing. All of a sudden, everyone's able to be a product marketer.

I've seen tons of pro-- That's what I do. Tons of posts about that. Linus is going crazy because there's so many bug reports because anybody can do it, but the right way to submit a bug report is, "Here's a bug report.

" Yeah, but the fact of the matter is the bug bounty programs have been around a long time, and they've been paying money. They generally don't pay for the remediation. And Fred, you would know more about this than I.

If you just report the bug, sometimes they do attach a fix for it, but more importantly is the proof of the bug. Fred, when you're dealing with bug bounties, is that your... Yeah.

So I think the Linus part of this problem is a broader part of the equation around, hey, we've got a whole bunch of capability now, and we're producing a bunch of duplicated bugs that I would say are low hanging fruit. And so some of those things are hard to go through the dedupe process for somebody trying to, a single person or a Linus, to go through and evaluate. Obviously, there's an impedance mismatch, right?

So we would also say that some of those super obvious things, you could probably use some agentic flows here to simplify the work stream and not have that same sort of problem. So it's more of a old world, new world part of this problem. And the crux of that, why that's important is when we start talking about things like the Linux kernel, that is the ring zero on trust of all the things.

So it becomes very impactful and important. But to the point on the bug bounty part of this, yeah, the submitting PRs for bugs that need to get fixed, that's a byproduct of an automation world problem. Yeah.

Bug bounties have been around a long time, and one of the topics of debate my good friend, Katie Moussouris, talks about regularly is how to marshal what those look like, and also the potential variants in the future. So everybody made a big deal about Mystic, great. Or Mythos, sorry.

Mythos. And then everybody is now also making a big deal about which models have what level of capabilities. But that isn't really the problem.

The problem is the triage, the problem is the remedy. The problem is being able to provide the value that helps illustrate that difference. So without getting into the love-hate part of that, I think it's just more of a, if we were to look at the data-driven part of this problem, we would simply say, "Hey, look, now I've got 1,000 bug reports where there's maybe 100 bugs.

" Okay. We need a better system to be able to fix the problems that we have. It's a known problem in the industry.

It's great that the entire industry is burning their tokens to help remedy these things. Bug reports for open source projects do not get you money, and so how phenomenal that the community wants to illustrate this for you, and Linus, get a workflow to solve this problem for you. And I think that's the consensus around how you think about this problem for folks that are dealing with the bugs.

I would love it if software that I wrote in an open source project garnered 1,000 different bugs and bug reports that I had to triage, and then I would hate it because I had to triage 1,000 bug reports, not because AI's doing any of that work. I think there's two things that are important here, though, Fred, to recognize. If it's not white hats finding those bugs, you can bet there's black hats finding those bugs.

For sure. And so for every bug that gets reported, there might be another bug or three or five that the bad guys have found and they're using to do bad things with. And so-It's a good thing that we have this many people looking, running these models to find these bugs.

However, again, this is a common issue we're seeing in AI. It's a scale and governance issue, right? Sure.

The scale of the amount of reports we're seeing is overwhelming our human-based or our human-design systems to deal with the scale, and so we need a better system for that. I agree. Here's one really interesting thing that Linus said that I think everyone should really sit for a minute and think about.

The notion that because AI found a bug, that that bug should be considered public. Now, why is that interesting as a statement? Because that says that all AI things are equal, which is inherently not true.

No. All users of AI are then therefore inherently equal, which is also inherently not true. And so if a random person found the same bug as an experienced bug hunter, threat researcher, then I would say there's something to hold water there.

But that isn't the world we live in, and everybody doesn't find the same bugs. You have a low-hanging fruit that people can certainly, and this happens all the time. As soon as you start going deeper into both your use of tool set, your experience and the outcomes, you find more interesting, better things, and it requires more effort.

So I think this is an interesting statement to make, and I would say maybe a little tone-deaf. Whether or not that's actually the reality we live in. " And none of that is true.

In fact, like you said, Shimmy, the governance part of this and the scale function of it can only operate with these rules. Otherwise, we're just like everybody else. We're just like the animals, as they say.

Yep. Now look, Fred, you remember a time before responsible disclosure was a settled thing, right? You used to have security researchers who would just announce bugs in products without even giving the people a chance to fix them first.

Yeah. And it was a chaos. I think all of those issues have been pounded into the ground for the last five days on this show.

But the thing that I would come back to, it's like, I think it's okay to hate AI even though you're going to use it, and I think we're going to have this as the new normal. Or does AI get better to the point where all these little issues that we're talking about go away and- Or are you hating AI because you're holding it to an unrealistic bar? Or are you hating AI because big tech went out there and said they laid off all these people because of AI?

That's- And yet- Yes ... and yet if you go look at their numbers, basically, their hiring levels are right back where they were pre-COVID. So, how much of this was just them hiring too many people, and AI is a convenient excuse?

Yeah, I think that's it. Right. You lay the blame at the doorstep of AI for being a bloated, overstepped companies that cashed in during COVID.

And it's convenient, but also moving forward, it also has a deep impact amongst users. Well, the other thing is that we talk about hating AI, but we don't talk, like AI is a tool. So AI is a tool, and like Fred really explained well, AI usage blows up all of our current workflows and work processes across industries.

So until we get a handle on, okay, we use this tool, and then something blows up over here, so we've got to change the workflows to make the workflow match what the tool is able to produce- Right ... it's all going to stay the same. So we have to fix the workflows for program accordingly.

You've got to work this. The rat has to work its way through the snake, so to speak, right? At each- Mm.

This theory of constraints. As soon as you move one bottleneck, you get the next bottleneck. We're at a bottleneck right here, though, and we're over time.

But Mike, is Google the answer for any and all AI? Are they in the catbird seat in the AI race? Yeah.

It was interesting. They had their big developer conference this week, and they were touting the fact that they have, I think, somewhere around 375 customers that are consuming over a trillion tokens. And what they've done is inserted AI into just about everything that they provide, and ultimately, that's driving up consumption of their models.

And, before anybody else gets off the ground and starts kind of establishing some sort of alternative universe, Google's just saying, "Hey, I already have the points of distribution. I already have the customers. " And this, I think, is going to be par for the course, and I think Microsoft's probably got the same kind of idea going here.

So ultimately, I don't know, John, is there room for all these so-called other startup companies, including whether it's Anthropic or OpenAI, to actually go deliver some sort of application when fundamentally this is becoming a feature? What Google did, we're not surprised by what they announced, but I guess it's the speed of what they're doing this And they're basically positioning Gemini as the foundational operating layer for enterprise and consumer services. 5 million developers utilizing those models.

So in a sense, they've introduced this vertically integrated agent control plane that spans everything from customer silicon to Gemini Spark. 5 Flash, which is developer as reasoning in action. To mention the buzz is Google Search is undergoing an agentic overhaul, so now you're going to be able to actively synthesize data, monitor topics, and dynamically generate custom dashboards.

It's overwhelming, and it always kind of brings me back to something that I think Alan always mentions. Like in every major market, there are three or four major players who carve out to the dominant niches of them, if there are three or four, but basically about three. And I think that's what Google is doing, obviously.

They're leveraging what they have, and they're doing it fast, and they're doing it smart. And I suspect Microsoft will do the same thing, which really interests me in how it all plays out among those who are scattered below their levels. Yeah.

Alan, is it all over but the shouting? Well, John, I'm glad you brought that up, and Mike, I think there is an argument to be made that the top three at the top of the food chain at that frontier model level is locked in. It is Anthropic, OpenAI, and Google.

And quite frankly, Anthropic and OpenAI will be jockeying with each other like Bill Gates and Steve Jobs. For that, they're the new kids on this block, but they are the locked-in top three. The fact of the matter is, I think when you look at their relative positions, Google is in the pole position in the AI derby, right?

When you look at they got Gemini, they had DeepMind, they've got Google Search, they got Google Workspace. They've got Android, they've got so much, Google Cloud. They have the most tools to do it all themselves.

But here's the thing, and I hope it goes down this way. You're going to have these three behemoths at the top of the food chain, and they're going to have these general frontier models, the monster models. But a lot of people have foretold that going forward, the monster model may not be the right tool for every AI job, right?

Because you can't run that monster model locally. You can't run it on your phone-- Well, right now you can anyway, run it on your phone. You need to be connected to the cloud.

You need bandwidth. Can I use a small, angular, nimble model for the particular job I'm doing? I'll give you a for instance.

I've been experimenting this week with a virtual EA. I gave her a name, a background, everything. She's great.

Half the people she's writing to think she's real. They're having conversations with her. She does a great job of managing my calendar, scheduling the interviews, all of this stuff, responding, add a calendar stuff.

And I was talking about it with one of our folks here at Techstrong. " They said, "Well, can I make my Perplexity agent do that? " And I suppose you could using one of these frontier general models, but it's just a hell of a lot easier using a model, and by the way, I think ultimately this one's based on Meta, maybe Llama or something.

But it's very rigged out for a specific purpose. Is that the future? Google's already talking about that.

We had a show talking about Remy, and that's their personal assistant for everybody, and that's- Yes, they do. I saw it. I actually compared it to what I'm using this week, and I got to tell you the truth, what I'm using blows it away.

Yeah. Because the Google one locks you into Google. Yeah.

I think they're right to capture the market they have, but I think that will be the market they get. The rest of these things that are broader in scope, when I look at an agent harness that I use, Google's not in it. And the agent harness includes, for coding, for example.

Which platform I use for this and what instructions I give them, settings, rules. Which models, sure, the hooks for agents, and then also which agents I've delegated to do what tasks and so on. And there is an agent control plane here.

And there's a working group building an agent control standard that is meant to be ubiquitous as opposed to a monoculture. And when I look at what the feasibility is for the rest of the technologies, Google works if you use Google, sure. Their search share goes down with every model development that happens.

Their ability to force leverage the folks that did not say that they want their data used to train AI models continues to grow... small businesses and otherwise. So is there a convenience tax?

100%. Is that the end market? No way.

And the benefit is the ecosystem here from an innovation perspective is growing at a faster rate than they can capitalize on the data that they already have. And my hope is that there is definitely going to be some arbitrage about who is going to be in the future. Is it really still just Microsoft and Google in this context?

I don't believe it will be. I think you've got to look at the math problem. If Google and Microsoft can satisfy 80% of the needs, then that winds up being 80% of the market.

There's not enough market left over for all these other folks to be sustainable in terms of what their costs are for building these models. So I think at some point- Absolutely ... do I believe that there'll always probably be better models from OpenAI and Anthropic?

Yeah. Do I believe that you as an advanced user will use those? Absolutely.

Do I think 80% of the rest of the population wants to do relatively simple things and doesn't need those models? Probably so. You know what, Mike?

I remember when AltaVista said that about search. Mm-hmm. And AOL.

It didn't work out so well. Yeah. And AOL.

Or AOL. So I think that's a good point, though, because I'm not using agents specifically because I'm trying to use all of the new improvements from Microsoft, because that's where all my data is, and most of my clients use Microsoft, so their data is all in Microsoft. And it's very frustrating as an advanced user not to use and build my own thing because I run up on the limitations, and with Microsoft because they have a lot of legacy data platforms.

It doesn't all work like it does in the videos that come out of the different hype machine. And it's very, very frustrating. But I want to know what people are going to go through so I can help them get through it, so why I continue to do it.

So I think that there's going to be a problem of you're locking people in and there's going to come along and be something that makes it just as easy for knowledge workers to do their jobs with documents and spreadsheets and all the rest of it, not so much with agents and building those platforms. That that's going to be the jump. What's going to make it easier, what's going to make it secure, what's going to keep the data secure, that's going to be real important.

Mm-hmm. The one thing I also want to bring up about Google is they have a lot of humans in the loop helping them, and my question is, that doesn't get mentioned in these really great stacks that they're showing us, of course, because they don't want us to know about the humans. There is still a lot of human work that goes in to the things that are returned from Google and the results that are returned to the common person, because there's definitely probably some legal liability if they get things wrong in certain places.

So I think that part's interesting, too. How long can we have humans doing this work at such a reduced rate all over the globe for these big models, and how sustainable is that? Is there a way around that?

To your point about that, somebody said it only takes one lawsuit to wipe out six months worth of AI productivity gains, so. If there are AI productivity gains. But let me- Look, what are we- Who knows ...

really talking about here, right? This is an age-old question. Do I want one throat to choke?

Right? One do it all solution, or do I take best of breed in the different things that I need? And Google and Microsoft are going to continue to offer you the one throat to choke, right?

And you look at AWS, right? With Bedrock. They give you the ability to plug in multiple agents- Mm.

Hmm ... into the back end and use whatever you want per job. Salesforce and Einstein had a similar kind of thing, right?

That you can plug in multiple agents and models back there. The Perplexity computer agentic solution that I play with also has Gemini, OpenAI, and Claude in there, as well as others. So is the future that homogenous or pasteurized that we just go Google or Microsoft or whatever the next one is?

Or is it more will people reject the walled garden? Speaking of walled gardens, what about Apple in this? Mm-hmm.

Same thing. Yep. Anyway, we're about out of time for this segment, Mike.

We've got to move on. Are we going to keep talking about AI? You bet we are.

What do you got next? So there was Jim Zemlin, who runs the Linux Foundation, was in Minneapolis this week and gave a keynote, and it was roughly the equivalent of a John Paul Jones speech about how the open source community has yet to fight. And the two things he was pointing at, which in some ways are diametrically opposite, but on the one hand, he was saying, look, it's pretty clear that people realize how valuable data is, but now they're putting it behind moats so that it's not accessible, and the AI models themselves are finding it a hard time to get access to additional data, at least without licensing it or paying for it.

So he was signaling that the Linux Foundation would be looking at expanding their efforts to create more open data sources that could be used by any AI model. At the same time, he was saying that the AI models themselves, at least the ones being built by the open source community, are catching up to the proprietary ones. And at the speed and the rate at which that happens, then the cost of AI will drop considerably because-It may not just be that the only game in town is OpenAI versus Anthropic versus Google and Microsoft.

It may be that there is another option here where the open source model becomes much less expensive to run, and people will do that instead, whether it's an ISV or an enterprise or whatever. So as I look at these two things, I'm not quite clear to what degree they might succeed. But we've seen this before, right, Alan?

Where there's a proprietary technology and then it leaps out, and then the open source community kind of catches up, and then before too long, they wind up driving the innovation. Unix Linux. Mm-hmm.

It was perfect. So Mike, in my mind, and when I first spoke to you about this, I guess it was last week or earlier this week, to me, this is open AI with a small o, right? Versus OpenAI, the company with the capital O.

Mm-hmm. But here's the fundamental issue for me that I think you got to address it before you convince me. This ain't about just software.

This is about hardware, infrastructure. You could have a great model, and I'm going to get 1,000 tiny elves to help me train it. But ultimately, it's got to run on someone's GPUs somewhere, right?

And if we're going to get it done in some workable timeframe, what data center is this running on? And so ultimately the question is, where's the real cost? Where is the real cost?

Yes, there is definitely a cost in amassing the data you need to train a model. I get it. And it's getting more expensive because people are now wise to, you just can't rip my IP off to train your model.

But the training of that model, the running of it, the inference, the use of it, is so tightly tied into infrastructure. And that ain't open source, and that ain't free. Well, but we've seen this play before, right?

How often have the cloud service providers taken a piece of open source software, turned it into a service, and monetized the crap out of it against a proprietary platform as a rival? So I think that that's the infrastructure's already there, and those guys are dying to find more things to put on their infrastructure. There also has to be somebody from an infrastructure standpoint to stand up to Nvidia.

There's lots of other things that can run these models. It doesn't necessarily just happen to be a GPU and a GPU for each workload that you're running. So there has to be something- Well, no, let's be clear.

The GPUs are more for training. Right. Right.

The inference chips are, there are a lot of contenders. Right. But the thing is that the standard for all of those things, especially for the networking, Nvidia is driving the communities to collect, the for-profit communities to collect.

There needs to be, where the innovation is, which is with the networking right now, I think, there needs to be an open source version of that. So it's not all locked into what the providers right now want to provide and want to build, but it's also open to what could possibly work and what could work better if everybody joined together and figured it out. We'll also see AI accelerators that are not GPUs- Yeah ...

that have been built from the ground up for training purposes because they basically don't need all the graphics overhead that the GPU has. And so they've shown up and found that a more cost-effective way to do training, and AWS will even tell you that some of their Trainium chips are being used for training purposes as well. So those options are starting to emerge.

So, a couple thoughts- Yeah I guess it was reflected in Nvidia's earnings yesterday, huh? That might be a non-leading indicator. A trailing indicator?

Yeah. You know what? Five trillion goes a long way on the way to seven, my friend.

Uh, yeah. Fred, you were going to say something? A couple of things to consider here.

I think the focus for the Linux Foundation has got to be more centered around maybe users, right? Getting a collective data set. There are a lot of open source models that already exist, already trained on existing data.

Data continues to compound. Agents can get at all that data because users have credentials. So, talking about open source data and hidden behind paywalls sounds great.

Users are already circumventing this problem today. Catch up. The second part of that is when we talk about the governance part here, and I think that's really, Shimmy, you hit the nail on the head.

Instead of talking about how to get to a place where you can help drive the value down on the cost, the whole reason of the theory of constraints is resources aren't just chips and memory and bandwidth, right? It is also the ability to get access to data. It is the ability to drive down the marketplace.

And when people are tired of paying for it, they're going to create their own models and train their own models to do the things that are necessary. Large language models don't require, you don't require large language models for every task. And you don't require GPUs to run some of these models locally.

And you don't require a number of these other things. And my concern here is in an effort to, and the Linux Foundation historically has taken a lot of bullets for being super politicized. A lot of money in there, not a lot of outcomes.

We've already had open source projects contribute to the Linux Foundation for the agent harnessesStart there. Please make some headway with this. It's been more than six months, haven't heard a peep about it.

And outcomes that are necessary to drive collective open source communities forward should harness the value instead of talking about the governance. So please use the things that you're building for in order to prove the things that you're building, so that we can make some more rapid advancements here, and not in what sounds to be an effort to remain relevant. Yeah.

Fred, did you just tell them to get off the can? Is that what that was about? Subtle.

And in fairness, that's an oversimplification. I don't mean to be glib, but the simple fact of the matter is we can't sit on the couch and wave fingers at people at this stage in the game. You've got to demonstrate by doing, and you've got to make some progress.

Otherwise, we become irrelevant. Doesn't matter whether you're a nonprofit that's been funded to do this particular set of things, you're a user at home, fighting for a job that you're not sure where is going to land, or you're an organization battling it out with titans. It's all the same problem, and you don't get to sit on the couch and read the newspaper and complain about it.

No, but what you do do is wait in the grass to pounce at the right moment. Mm-hmm. And again, I have nothing against Jim and the Linux Foundation.

They've done great stuff. But they're not going to jump into a fray, or a market, or a battle unless they can win. And I think what they're doing here is beating the drums a little bit to see what comes out of the bush, and before they pick a horse.

Yeah. Or a cow, or whatever their good animal. Or maybe it was just a cry for funding.

Could be, too. " No one likes a loser. Right?

No. Are we placing bets now? Is that what we're doing?

Speaking of which, bet on the Agentic AI for DevOps, a DevOps experience. It's coming September 24th. I'm going to start banging the drum on this one now.

We're going to have some great speakers, and the theme this year is, which Agentic AI horse are you betting on? Mm-hmm. Or multiple.

We'll have trifectas, perfectas for all of you horse betting degenerates out there. We'll have all kinds of action. Are we going to hook up to a prediction market?

How are we going to do that? That's an interesting thing. Maybe we'll make a model or something to do with that, but possibly.

Anyway, hey, I want to tie a bow on this one, though. Fred, you said it, I said it. I think Gina, you said it as well.

The issue that I think we're really confronting is not the ability for AI to file bug reports, to do code, to produce documents. The issue is one of scale and speed, and that becomes an issue of governance. How do we take systems that were not built at machine scale and use them to respond to machine scale input?

And that needs to work its way through the system, and I think we're going to be running into it over and over again. Well, I just need to find a way to do that, that we can afford, because this is getting- And the economics have to work. Mm-hmm.

But speaking of economics that have to work, we've got to pull the plug on this and do something that's sponsored following. So that's going to wrap up today's Techstrong gang. Big thank you to each of our gang members, as always, especially John, sharing his European vacation with us- Yeah ...

and the great conversation and everything else. If you like what you saw today, be sure to subscribe to the Techstrong TV YouTube channel, or you can check us out on the OTT app, which is available, Techstrong OTT app on iOS or Android or Roku, Apple TV, anywhere you got a screen, Amazon Fire. tv.

We just redid that site. It's fantastic. I know the designer.

He's a genius. I designed that site. The Techstrong gang is live every weekday at noon, Eastern Time, and you can watch it on the Techstrong network, including all those places I just told you.

But for everyone here, I'm Alan Schimmel. It's great to be back. It's great to have Mike and John and Fred and Gina with us.

We'll be back tomorrow with more, gang. Take care, everyone. Bye-bye.