AI Needs Discipline as Ethics, Profitability and Engineering Collide

Hey everyone, happy Monday. Welcome to another week of Techstrong gang. We've got a great gang for you today.

Let me introduce you to our gang members. We've got our friend, Chris Velazque. Chris, good to see you.

The one and only Kate Scarcella. Kate, it's great to have you on. I haven't seen him in a while, my friend Mitch Ashley.

Mitch, looks like you're back home finally. Good to have you. Actually not on the road.

What a concept. Yeah, I hear you. And of course, Mike Bazar.

Mike, I see you're home as well. We're all home. It's amazing.

It must be the day after Memorial Day. That's what it is. We are.

Right. Everybody's home from the long weekend. So I thought we'd start off with a joke today.

Okay. The pope, a rabbi, and Iman walk into an AI data center. The pope writes a 43,000-word letter.

Well, it's not a joke. It's our first story. Mike, are we mixing?

We had a wall between church and state. A lot of people wanted to take that down. Do we need a wall between church and AI?

What's the story? Well, yeah. The separation of church and state just means that the state can't tell the church what to do, but that doesn't mean the church can't tell the state what to do.

So there you go. How does it dominate? So the pope has written a thing called an encyclical, which is a fancy phrase for a letter, some 40,000 words long, about the state of the human condition in AI and calling for the disarmament of AI.

And that term is chosen deliberately because it means it's a call for more regulation, it's a call for not militarizing AI, it's a call for keeping people at the top of the AI food chain and not the other way around, which now not everybody out there is listening to the pope, but Kate, you've got that last name with a vowel at the end of it, so I'm going to assume you have some familiarity with the subject. I'm going to start with you. At the very least, it seems like we're having a discussion now that maybe we haven't been having enough of.

What do you think? Yeah. Well, first, everyone has the right to speak from this moral framework, religious, secular, political, technical, whatever.

What I find interesting, at least in American politics, is that we have been openly using religion forever here, but since the 1980s, and of course it's been a lot longer, to position this cultural war argument and national identity, which for me has been since the 1980s. So it's hypocritical that suddenly we are shocked that religion has entered AI in the debate from this other direction. So I think it's not that the question should not be that should the Pope be allowed to comment on AI?

Yeah, of course. Everybody has the right to say whatever they think. And I don't care where you sit and what position in the hierarchical structure, wherever you are.

But the question is, I think, why are we comfortable when religion is supporting power, but uncomfortable when it challenges power? And I think that's the question I think that I'm hearing today. We embrace religion when it's supporting our view, but all of a sudden when it's not, it's like, "Whoa, hang on there, dude.

Let's not do that. " So he brought up some valid points, and at the end of the day, I think it's okay to have an opinion regardless of where you sit. Yeah.

" And of course, the Vice President of the United States claims to be a practicing Catholic. But will more government officials listen to this issue, and can it become more of a political issue simply because the Pope's talking about it? Well, it's already a political issue.

But Mike, I wrote an article on this one as well, and the important thing to remember here, though the Pope came out with this 43,000-word, in essence, letter position paper, other major religions have spoken out on it too. Right? Some rabbis and some of the Jewish sects have said, Hindu, Buddhist, Muslim, many of the largest religions in the world have kind of staked out positions, and they're not dissimilar to, I think, what Pope Leo came out with.

I think here's the important thing. You didn't see them come out with this about cloud computing. You didn't see them come out with this about Kubernetes.

You didn't even see them make a big stink as much about the internet, frankly. So clearly, there's something different about AI. Clearly, something different about AI.

And is it that, I think at the end of the day, are we dealing with a boogeyman? Some perceived boogeyman that has all these religious leaders, no pun intended, spooked. OrIs there really something here?

When there's enough smoke, there's usually a fire. And what they're saying is, look, this has the potential to really have such an impact on humanity and on the human condition that we need to do something. We need to not use it for war mongering.

We need to make sure people feel fulfilled and have jobs and something to do. So, whether or not the government is going to heed the advice of the theology folks, of the theologians, I don't know. But clearly there's something here.

" I don't know. Chris, what do you think? Well, you look at what gives any organization godlike powers is really authorship, and that's what we're talking about here, right?

Information systems don't really change over time. But they are politics. That's where they come from.

That's where politics comes from, right? We had for hundreds of thousands of years information systems that were oral, relational, you literally knew who you were talking to. Trust and so forth was a different model.

But for about 12,000 years, we've been writing it down. So who's writing it down? And on this particular story, I understand completely.

One of the big publishers of authority in modern human history over the last couple of thousand years feels it necessary to say this because authorship itself is at stake. You can change all sorts of things in human society by offering with authority or with scale or with both, and that's what AI brings us. I personally hate that religion is getting involved and vice versa.

I hate that because it stops us from actually thinking practically, practical about what we're facing, and creates this other element that I don't think is real. So... Well, there's religion and sociology, and they get mixed.

Religion arose, I believe, as a mechanism for social function and structure, right? And it has issues in the current day and so forth, has a lot of value to it as well. So a lot of what we see as religion is really just a cultural structure, right?

And as you know, I'm deeply involved in Yemen, which is an interesting case. There's plenty of them, but there you have a religious and a cultural structure that are continuous over many, many generations, many centuries. And which parts of a religion are just social protocols and which parts are theology?

I think in the modern conversation in the Western world, we blurred that. But the social structures, and again, the Holy See, the Vatican, the Catholic population is a huge demographic, it's a huge information processing system. And they're just like any other one way or the other, theology aside.

Yeah, I think the mixing of sociology and religion is a good point, Chris. Frankly, how could religious leaders not weigh in on this? You drew good analogies too.

Why didn't we talk about this or react to it about the cloud or the internet or past innovations? We have several factors going on here. One is in our culture today, how many movies over the past 34 decades have been about machines or AI or something taking over humanity?

Usually not to positive ends, in most cases. So it's built into our culture about this dystopian future that AI is one of the things that will definitely lead to it. Certainly is if you believe science fiction leads to reality for a lot of things, which it tends to.

I think the other is it's about authorship, but it's also the point of view, is it constructive? Is it condemning AI, and we should ban it all, take it off the planet, wipe off all the disks and every piece of silicon that does AI? No, that's not a practical thing.

Is it about safety and controls and what should we be doing? And yes, it's through a religious lens. Yeah.

How could it not be, given who the author is? But it does largely revolve around weaponizing AI, which a lot of technology definitely is used in weapons and weapons systems. But in this case, we have the tech bros claiming they're going to control everything on Earth, and nobody will have jobs anymore.

And we kind of self-destructively create an environment where there's a lot of hostility around AI. I think that's also part of what the Pope is reacting to. I don't think that the controls and guardrails are all that good from what we've seen so far, so maybe that is where the conversation needs to at least be had, and everybody can agree that maybe they should be better than they are.

But, we've talked about this on previous shows. I just can't tell if the guardrails belong on the data side or if they belong on the AI agent, or is the AI agent just exposing all the weaknesses in the controls anyway, so... I agree with that, Mike, that AI is just exposing all the controls.

Mm-hmm. So, yeah. Or lack of.

Or lack of, yeah. But go back to the Anthropic co-founder who was there, who by the way, not Catholic Right. He was there to lend, dare I say, moral support.

But he clearly, Mike, you said it half-jest, but he clearly was saying, "Hey, glad you're doing this. " And that really is, I think, a fundamental piece of this. Religion is weighing in.

And again, I don't want to make this just about Catholics. I think most of the major religions in the world are weighing in. Maybe not as clearly or as in-depth as Pope Leo did here, but I think clearly what you're saying is, don't count on the tech bros policing themselves.

Well- Right? The people behind the technology. And don't count on our government- Yeah, I- ...

policing either. Go ahead, Chris. To be clear, I have preferences.

I like a highly distributed, independent system structure, but that's a bias, and maybe I'm wrong about this. Maybe it has to go otherwise. The most interesting thing I'm involved right now is the Standard Council of Canada has a working group working on the technical specification for systems of AI systems.

Right? Now, I have my thoughts on this, and evidence, and provenance, and whatnot, but getting a window into where we are, where the peers at the table, the big AIs, the smalls, and nation-states and so forth, and thinking about this. Again, state my bias, but I think it's clear at the end of the day that centralization just does not scale.

It's not just undesirable, and it is. And again, when you have an otherwise centralized information structure like the Catholic Church saying, "Hey, we can't count on it," right? The same sort of thing that the gentleman from Anthropic is saying, right?

This has to go down, I think, down to local AI, and Kate, the kind of stuff that you and I'm always rambling about. Yeah. So I don't think it's just desirable.

I don't think it's just a civic good. I think it's thermodynamics and the ability for systems to actually function that's going to continue to drive these sorts of issues down locally. So no, you can't count on a government or a corporation to do it, right?

Even if it was their job, I don't think it's possible. I think part of the challenge, though, is where does that line, what Kate was talking about, and where their religion is kicking in an opinion versus what are our laws going to look like? Because too often the laws wind up being overly influenced by a particular religion, and then we get into a bigger mess, right?

That's not the problem we face in this country these days. Mm-hmm. Too often today, our laws are too much shaped by who puts the biggest dollars in someone's re-election campaign or PAC, right?

I think that's a far bigger problem in this country than whether how many people listen to the Pope anymore. That's a good point, Alan. And I think sometimes the religious argument will be the second argument to get everybody to come along with whoever the biggest funder may be.

So- Mm-hmm. That's it exactly. Yeah.

And then they go shopping for the particular religious opinion they're looking for- Exactly ... to back up whatever their money play is, right? Or the religious argument becomes the distraction from what's really happening.

Right. That often happens in our- Yeah, public life ... communications and society of making a big deal out of something to distract while you're doing exactly the opposite.

Mm-hmm. Yeah. We've seen that.

Oh, how cynical. I'm sorry. I've been on the road too long.

I've seen many things, Alan. I have to tell you, some of the stuff I've seen is... And I remain to be the grumpy Gen Xer, so- Well, yeah.

But let's not kid ourselves. Some of that money winds up in various pockets, right? Oh, yeah.

There's that too. Hey, you know what? I'd like to talk on this more, but we're about out of time on this topic, Mike.

Go ahead. But speaking of AI and money- Yeah. What about AI and money?

How did that happen? So Alan has a post up that's definitely worth a read, and it's pointing to the fact that we have shifted a gear, and AI is suddenly no longer about, well, let's make it as accessible as we possibly can for everybody. We've got to actually pay for this thing.

And we're starting to see some shifts in policies and shifts in positioning, and some of it is starting with the tenor and the tone of the conversation with Wall Street, where there's more emphasis on how efficient some of these companies say they are, or about to be, or will be as they get ready to launch their various IPOs. Alan, you wrote this article, but I'm going to let Chris have the first word on this thing. It seems like, is this conversation shifting to being all about the Benjamins all of a sudden, Chris?

It kind of always is at the end of the day. In economics, that is calories. That's what makes these organisms, these systems work, right?

And I find myself having to defend, because I am a bit of a hippie at the end of the day. However, I'm a capitalist hippie. I think trade is a wonderful thing.

It's an evolutionary pressure, and you can go, look, saber-toothed animals and so forth. Biology gives us a lot of examples of what can really work for a little while, until it doesn't. And we find that resilience works over and over again.

So yeah. We can sit here and debate, and we have. Who, which wins this?

Is it the model vendors and so on and so forth? I have my thoughts, but the fact that we're talking about this right now, after the last 12 months and before the next 12 months, tells me it'll probably work out. I think a year from now we'll be saying, I think, that it was the folks at the bottleneck.

Models are a dime a dozen these days, and they're lovely things. However- Again, maybe I'm biased right now because I'm focusing on this systems of systems thing, but it's how it all goes together. That's where the evolutionary pressure's going to come.

So, yeah. But yeah, we're going to find out. I think the era of just trillions of dollars into things because they sound good, may be waning a bit.

To your point about sounding good, Alan, is this just noise for the investor community so that they'll throw more money? No, I think this is one of the normal things about the AI cycle. It's maybe like everything else in AI, it's at a scale greater than we've seen before, but it's perfectly normal, right?

Mitch, we've done the startup thing. Chris, you've been involved. At first, that board says, "Take the money, get as many customers as you can.

Market share, market share, market share. " First, they want just market share. Give it away for free.

Then they want top line. Have people pay for it, don't worry about the cost. And then just when you get that good comes the next quarterly board meeting, and they say, "Okay, well, now we've got to worry about the bottom line.

Now we've got to worry about the bottom line. We've got to do this profitably. " This is startup 101 right there, my friends.

Except now it's not just millions or tens of millions or even hundreds of millions of dollars, it's billions and billions of dollars, and you've got companies that are looking to IPO at a trillion-dollar valuation, and they're not profitable. A trillion-dollar valuation, they're not profitable. What are they waiting for, a hockey stick to kick in?

So this is somewhat normal, it's just on a whole different scale. I think one thing that's happening here, the value chain dynamic is in most cases, the people creating the models are also creating the tools that use the tokens to consume the models. Right?

So there's so much more emphasis now on efficiency and use of tokens, but it's not by the model makers, it's not by the tool makers, it's by the users, it's by consumers. There aren't- These are the people who have to write the checks. Yeah.

They go, "How did I spend $20,000 this weekend on my Hermes project that just ate up everything, every token on the planet? " Right? So it's a bit of a, "Well, hold on, wait a minute.

'" Wait, but it gets better. So going to Alan's point about history, the history of these companies is eventually they don't make that bottom line thing work, and then they look to get acquired by somebody, right? And basically that's their exit strategy.

But in this case, Alan, I'll ask you, who has a few spare trillion dollars to buy some of these companies when they hit that inevitable roadblock? You'd be surprised. You would be surprised.

Look, though some companies are overextended, perhaps Oracle, right, in terms of the money they put into AI and some of the regular techs, there's an awful lot of Wall Street money just sitting here waiting to scoop up something here and do it well. I would say yes, the companies who don't figure out how to do it profitably do disappear as separate entities. " And that was the whole point, is I was replacing the person to do it more cheaply here.

It's the same argument where, "What do you mean the cloud costs more money than keeping it in my own data center? " Hmm. There is that.

So you have two pressures. You have the Wall Street people and look, I've lived this life with the VCs and the Wall Street people. They make it very clear, "You're going to do this profitably.

Anything that's an offering that's not profitable, cut it. " Well, suddenly your next funding round is dependent upon being profitable with the last funding, right? Exactly.

So it's not just the pressure from the end users because the end users, they're the drug addicts in this picture, right? They may complain about the cost of heroin, but they ain't getting off the heroin. Yeah.

" I don't know, but I think it gets worse. I think we wind up with this, oh, it's just too big to fail syndrome going on again. Suddenly, these people- Well, that depends on, right, if you have that in the White House, and that's the ultimate drug dealer or the loan shark.

"I'll give you the money. " Right. So will the United States taxpayers soon own 10% of OpenAI or Anthropic or whatever it is?

I don't know. You heard it here first. Let me play the hippie capitalism.

It's a global world, right? Again, I may be American, but my country or other countries will make choices in this era, and there'll be winners and losers. Right?

And you look at where the value is to be found. Mitch, you mentioned value chain, right? And supply chain has occupied a lot of my horizon for the last number of years, and it comes back to a lot of things.

In the last segment, we were talking about sociology, cultural structures, right? It turns out when you try and apply those to AI, there's no person there. They don't know all the unwritten rules, right?

So, getting value out of that in business turns out is a little harder than it looks. And in the value chain, we've been getting away with a lot of things. The opportunity is to collapse that and not have the inefficiencies.

Whether it is not accounting for environmental costs and competing that way in the past, those sort of opportunities are there. Those are real opportunities. The efficiencies in existing systems where we've been coasting along, we're fine, is hugely inefficient to begin with.

Those are real opportunities. All these other things you're talking about, yeah, and there'll be bad bets made. Nations could suffer dramatically.

Certainly, investors will win and lose. But I think in the end, those efficiencies are real, and that's what we'll see being long-lasting. Really?

Let's go full circle then. Let's come back to the Pope. If we're going to wind up owning a piece of these companies anyway, and we're going to wind up having to bail them out, and if you think that that's how it might go, let's just cut to the chase and let's have the government set up frontier models on their own, and they'll fund that, and then we'll build the companies and ecosystems around an open source frontier model owned by all of us citizens, thereby checking the box for the Pope, because we'll have some control over that.

Yeah, but Chris is the hippie. Wait. The core technology itself is a commodity anyway, so maybe we can just move along here.

If Chris is the hippie, you're the communist. Two out of five. You finally called it.

Okay, I'm paying full salute. Go. That's just not the model today.

Right? It's about having these private companies, and the government takes a piece of the action, like that "Star Trek" episode where they base the whole society on a book about the Chicago mob, right? " Yeah.

Until I need another bailout, then I got a bigger piece of the action, until I don't need to be bailed out. But let me just say something. The point of the article I wrote, or one of the things that drove me to write that article, is actually Anthropic- Yeah ...

came out, and you know what? They are profitable. On about $10 billion, the story goes, and I don't know if this is corroborated or not, but about on $10 billion in revenue this quarter.

This quarter, $10 billion in revenue. They're turning a little over $500 million in profit. Yeah.

It ain't bad. They're turning real profit. And not only are they turning profit, but they dramatically, not as much as we'd like to see, but they dramatically decrease the cost of their infrastructure to revenue earned.

Right? So, these companies will get it. Yeah.

When you write compute cost reported drop from 71 cents to 56 cents per revenue dollar, and that's remarkable. And I think to Chris's point, talking about efficiency, that's what you see happening. So we're not, of course, eliminating it.

At the end of the day, aren't we talking about technical debt? And we're not eliminating, they're converting this operational leverage before it becomes unmanageable. If they continue along that trend- Yeah ...

that's a business. Fine. That's a real business.

Well, there's an argument to be made. Pick your vendor. I can't work in their environment just because of the cost.

I can do the same thing with a different vendor at a lower cost. It eventually comes down to not just a capabilities discussion, but what does it take to run my business? And if I can do it lower, I'll move it to the cloud.

If I can do it lower, I'll change providers of models, because at the end of the day, they are commodities. They are all in a cyclical chase- Yep ... of feature items.

So what happens, to Mitch's point, it's a commodity. So pricing does not stay flat in a commodity world. It goes down.

So the amount of revenue that you generate is not the only thing, right? Ultimately, you're going to be generating less profit per revenue cycle because you're going to have stiffer competition. So this doesn't look guaranteed to me at any level.

No, but it's about market dominance at the end of the day. I think that's what Mitch is talking about. Mm-hmm.

So. More in the future. Absolutely.

Right. So- Like the captions at first. Well, excuse me ...

in order to proceed, you have to be a monopoly, and we have to create that level of pressure and scale? No, I think he's saying that there's a strategic willingness to absorb this temporary inefficiency for market dominance at the end of the day. Yeah.

But to Alan's point, one of the things that the VCs that have been around a while will tell you is it's all about first mover. It's about getting the customers first and keeping them. Yeah.

Right? While you're continuing to consume or trying to gather more and more customers, and building that momentum is something hard to stop if you really get the flywheel running. And I think you can argue Anthropic has done that with Cloud Code.

It is kind of the de facto, yeah, Cursor's good, but in terms of market share, it's not even close. So not to say somebody can't come out with something better and people start moving over a different direction. They haven't won the game yet, but they certainly have got some advantages.

Yeah, I think Alan's spinning in his grave as we sit here. You know what, there's a strategy play here too, right? Claude or Anthropic have achieved these results pursuing an enterprise strategy very clearly focused on developers, software development.

OpenAI, on the other hand, has pursued, well, I would posit that they pursued an everything strategy. They're just throwing stuff on the wall everywhere. But at one level, 20 bucks a month, it's a consumer play.

Yeah. And consumer may wind up being a bigger market someday. But when you look at technology best practices, they always want to go after that enterprise market.

Yeah. Which, that was one of the things I thought brilliant and when you talk about going after the enterprise market, that differentiated Steve Jobs and Gates. Steve Jobs actually went after the person.

Yeah. Apple pursued a consumer, never considered themselves an enterprise play. Right.

And worthy. Again, it's just- Bill and Gates gave the AI chat. Right.

Just to be clear on that point, it was Bill who wrote the check to bail out Steve, right? Yep. All right.

You little smarty pants. The communist speaks up. Sorry.

I never thought of you as a communist, Mike. All right. We're out of time on this one.

Let's move along, keep moving. Mike, what do we got next here? We've got some interesting stuff.

Mitchell, you were at Google I/O. Yeah. Well, some things are happening here across the spectrum of things.

So we'll start with Alan's article about systems engineering and the benefits are being rediscovered and reappreciated. And Google's talking about that thing at their Google I/O thing, but they're not the only ones talking about it. ai that has a development platform that is designed primarily for AI agents rather than humans, and it's a soup to nuts end-to-end SDLC.

But Mitch, I'm going to start with you on this. Is something fundamentally changing in the way we think about software development, and how does AI agents change that or drive that? Well, yes and no.

Yes, it's changing, but what we're changing is leveraging what we already know, and that is software is not about writing code. Software is about engineering systems, building systems. Right?

And yes, there's code that gets either written or generated, but there's a much larger process that goes on, and you can see that, and Alan really made that point well in his article about systems engineering. That's what it takes to build something that you can put into production that's of any reasonable size or scale. But what's happening in the market is, it's not just about engineering the system.

Yes, it's about creating agents to build agents and multi-agent orchestration platforms, all of those capabilities. But we're starting to see the AI stack and different approaches to filling out that AI stack come to fruition. And a lot of it is not only where does my engineering or my coding or my agent to build software, but it also has to do sort of what about the underlying software substrate, the operating system, and open platforms that it operates on, and whether it's portable or not.

People are grabbing elements of the governance layer. That, I think is the big battleground, still is today, of who actually, to our earlier point about who is securing all of this, where is that defined, and who controls that? And am I willing to hand over the keys to pick your partner, pick your vendor?

My agent control plane, however it's distributed across that stack is AWS or IBM or Google, Microsoft. It's a big commitment. It's sort of like saying, well, I'm not just picking a brand, I'm picking where I'm going to keep my money.

I'm picking where I'm going to keep my controls, where that's really going to happen. And that's what we're seeing shake out today. I'm coming out with an AI stack paper that has six archetypes of how the vendors are sort of stacking up across that AI stack and their strategies, and a lot of it comes back to structural.

Google is very much trying to be the full stack, right? SaaS vendors like a Salesforce, of course, think about it from an application and a data layer and building out sort of more horizontal that way. But we can see things shaping up now where vendors are going.

Okay. Now, I'm a little optimistic about this whole thing, but maybe from the wrong angle. I feel like we ignored all these fundamental principles for years now because human developers would kind of work their way around it.

AI agents are kind of exposing that issue, and we're now revisiting all these systems engineering conversations, and maybe the outcome will ultimately be better because the AI agents want to behave logically and want to have the stack versus, and if they don't have it, they'll make one up, right? And I think that that's a perfect example, and good for you for being optimistic. Yes, I think now is the time to be optimistic.

" And we keep going on the oh, wow, this is so cool model without ever getting any sort of structure around it. And I think AI, just as we've talked about for the last year on how they were really good at coding, they're really good at structure. And I think at the end of the day, that the structure is going to save us.

That as we create this structure and as we create frameworks, real frameworks, not this-- Sometimes I've looked at these frameworks and I've been like, "What? " So the architecture, it speaks to me so much. And what do you think, Chris?

I mean, gosh, you've been-- Yeah. Yeah, welcome to my lair. I've looked forward to this segment today so much, right?

And it said AI harness on the notes for today. And look, we were just talking about maybe the Pope should have a frontier model. And let me overstate this.

I don't care about models. Doesn't matter, right? They're all good enough to use in a system.

It's like, and again, like human systems, we're trying to replicate, right? Where do I find the perfect human, the perfect employee? Stop trying.

Don't. Good enough is quite good systemically, right? We have stumbled along through this technology thing for the last 50 years without ever doing a lot of the things that really should be done.

And yeah, the technical debt means that now we need to fix a bunch of those things, but it's the same thing. It's a systemic world, right? And again, I watch for indicators over time that I see on my daily, weekly basis, and the fact that the International Standards Organization, which is not the acronym, but the ISO, this group right now is just thinking through, right now, what a system of AI systems looks like.

Which means it hasn't been written down. And again, I have my own thoughts on what it looks like, but I see what our peers are pondering out there, and I keep going back to this show every week and saying the same things. I think it goes very local.

I think, Kate, you're exactly right. They're really good at structure. We're not that good at structure.

We've gotten away without putting the structure in for a long time. This is forcing us to put structure in. But with that structure, these things, these LOM systems are kind of amazing.

But the big frontier model running in space or at some massive corporation's headquarters, not really. Maybe. It doesn't seem to play in anywhere that I'm looking.

Mm-hmm. Well, Mitch, let me ask you this. What does the future of software development look like?

Because we created Java and all this other crap to find a way for humans to talk to machines in a way that was at a higher level of abstraction that an AI agent doesn't need. So maybe we're just going to all go back to assembly, and the AI agents are going to do this thing with all this structure underneath it, and all these tools that we created for humans are quaint but obsolete. I remember that class in college, all that other crap that we created.

That was a great- Why did we have 101? There's one thing that we like to do is abstract things, right? It's another layer, it's another programming language, it's another, it's another.

And to your point, I think the programming languages, it will become less if not no issue, right? I don't really care if it's what it's written in, if the AI could rewrite it or update it or change it, whatever, because I'm not working at the code level anymore. A lot of people are taking that stance to really leverage the most out of AI is get out of the editor as much as possible.

0 doesn't look like a code editor anymore. It doesn't look like an IDE. It's not.

It's a different platform. But to your point, I love Chris and Kate because talking about architecture and structure, that's ultimately what it's about, right? It's about assembling the pieces in a way that you can both deploy, move quickly, add the capabilities that you want.

It's not about any one feature that any of these vendors or capability that they're offering. It's about constructing what you're trying to deploy and create, whether it's onto your desktop for your own enjoyment of your own personal assistant, or it's processing massive amounts of data. At the end of the day, it has to structurally be sound to be able to take on complex tasks.

And that's where, to your point, Chris, don't care about the models. I just care that they keep advancing and can take on tougher and tougher issues. Whether it's this one that wins or that one that wins.

Pricing aside, I'll use the best tool. Matter of fact, I'll write code or write agents to pick the best model for me or save me the most money. And to your point, Mitch, honestly, my whole TED Talk was basically if humans are systems, if we were to act more from a structure standpoint, we would all be better off at the end of the day.

You're going into Scott territory there. Okay. We're video logic nerds.

My only problem with this whole thing is like, all right, so the AI agent is going to go build something in some low-level language that we probably don't understand, and then if we don't really understand how it works, then we're going to ask, I guess, another AI agent to validate whether or not that thing works, but we won't know why that AI agent validated the first thing in the first place. So Mitch, at the end of the day, are we as humans going to know how anything works? Well, it's a good question.

We used to work on our cars. Do we work on our cars anymore? No, we don't.

We have cars that are much more complex, much more intelligent. I think the same thing is true with software. And by elevating what we work on to a different level, now we don't have to worry.

Maybe it's a hobby, but now we don't have to worry about turning down the carburetor this weekend. Yeah. I think it's important to keep in mind that if you're maintaining the code, yeah, you need to understand it.

If you're not, if you're using a code generator, you're using a different system to do that for you, going in there and mucking with the code, another technical term, you actually do yourself more work by getting in there and changing things around. And now the code generator, the agent, or whatever that you're using, is suddenly trying to dance around your changing and wiping those out, and you have to go back and reapply. " Well, it could do that, but it sort of defeats the purpose.

Well, look at the way you said that, Mike. And again, this is the anti-systemic systems we've been building. That they're not systems.

They're long linear chains. If I build a system where, let's take AI agents or any other component, if I don't know if this is perfect and it is the input into this thing, and I have to know if that's perfect, and I rely on making that decision, we're all going to die. My mitochondria right now is doing wrong things everywhere, but they're not single pathways.

But we've built single pathways into everything. Yeah. And that's why we're having problems with systems, because they're not systemic.

So we need to go a little interstitial here. Wait, Mitch, did you just say that humans in the middle, that humans are essentially in the way? Is that where you were going?

I'm saying there are times we can make things worse by trying to make it better. That's what I'm saying. Messing too much with the TV dial, right in the rabbit ears.

Just leave them. Yeah, sometimes, if the Wi-Fi is working, don't mess around with it. Just let it do what it's doing.

And if I don't like the code, I'll just ask the AI agent to do it over, right? Essentially. Absolutely, yes.

Well, and you already see contrarian where OpenAI now, you can bring Codex into your cloud code environment and use it as an adversarial programming agent to say, "Now analyze this, and make it better. Rip it apart. You would do it differently.

Why? " To a large extent, the code doesn't matter. The documentation you're generating code from, that matters.

You know what this illustrates? " And then it just became a huge mess. , Home Depot's closed, everything's a disaster.

That's this. Like, just step away. I'm starting to see how they're going to make money on these models, though.

So basically, for everything we do, we need three opinions. Three AI agents who will then have this adversarial conversation with each other until it goes round and round and round, and then, I just spent 50 grand figuring something out. There are already architectures for constructing your environment to have agents who are judges, agents who are architects.

And that's the whole point with the harnesses, guys. Mm-hmm. With a harness, I should be able to use multiple agents, not interchangeably, but for the right job.

Mm-hmm. And I shouldn't be locked into an agent. Now, of course, again, it doesn't make a difference on the end user if the underlying technology is not efficient enough to turn a profit, Mike.

So though you may spend $50,000 with three different AI companies, if it costs those AI companies $75,000 to provide you with that $50,000 of value, they're still not a good company. But on that, we've got to end it, guys. We're over time.

It's an interesting conversation. I think we're going to talk more about harnesses, though, because I do think that's a big key buzzword we're going to focus in on going forward. But for now, hey, thank you, Pope Leo, for speaking out.

It's good to see the church at least trying to be relevant. I think the Pope just needs to come up with his own AI agent, then we slip it in the harness, and we're good to go. How do you know he already doesn't?

Maybe the models are trained on the Vatican's core principles. He has a non-compete over there, so he had to put it under someone else's name. Who knows?

The agents are German. We'll be back tomorrow with more, I'm sure. So stay tuned for that.

But Chris, Kate, Mitch, Mike, thanks so much. Thanks so much for all of you watching. Again, you can watch this on demand on Textron TV, our OTT app on just about any kind of screen you watch on, or on our YouTube channel.

But until tomorrow, thanks everyone. This is Alan Shimmel. We're out.