Amazon’s OpenAI Bet, Microsoft’s AI Dev Plans, and Code Quality Warnings | TSG Ep. 990

Hey, you remember the girl who danced with everyone at the Prom, don't you? Was her name open ai? You're watching Textron Gang.

Hey everyone, it's Friday. It's the Friday before Christmas week. So it's a really good Friday.

A lot of us are hopefully already feeling that holiday cheer or just waiting to just push, you know, get get to the holiday break as we we this crazy year of 2025 starts drawing to a close. Thanks for joining us today on Textron Gang. Let me introduce you to our all star panel of gang members.

We've got, uh, Gina Rosenthal looking great. I don't know if that's a new camera or what, Gina, but that you are looking like an eight K today, my friend. Um, joining Gina, we've got, uh, Alistair Cook who's joining us from, uh, the Land of Oz, I assume getting ready to spend the holidays at home, as well as John Swartz and Mike Ard and myself.

Welcome gang members. Mike. You know, we've only got a few stories left for the year, but it's gonna go out with a bang.

Open AI is, uh, dating again. What do we got? They keep On coming.

They sure do. Keep on coming. I'm having a hard time sorting this one out, but, you know, on the face of it, it looks like a hell of a discount for open AI to use AWS processors, but what do I know, John, you're closer to this thing.

What's going on? And can you make any sense of this one? Uh, Jesus, you know, every day I try to make sense of this stuff.

You know, there's a photo circulating on, on, on x. I don't know if you saw it, and it's funny that Alan would would mention the Girl of Dances with everyone, but there is a photo of seven executives standing around each other. Sam Altman is in that group and Well, Is that the time cover?

Is that the cover of time? Yeah. Oh no, it's another one.

No, this is another one. Oh, okay. This is, yeah, this is another one where our standing one and, and you've got Su Sundar, you've got Tim Cook, you've got Zuckerberg, you've got all the usual suspects, um, Nadella and, and, and the caption is, it's your turn, Jeff Bezos, to invest in open ai.

And that's actually, it's going viral. But in any event, according to the information, which does have really great information, uh, there's a, a story that Open AI is in preliminary discussions to raise at least $10 billion from Amazon that's gonna be part of a larger fundraising round that if the information has subsequently said could reach a hundred billion dollars, which would give ultimately open AI evaluation of $750 billion. Now going back to the Amazon deal, which allegedly is, is being talked through, this would value, um, chat GPT at about $500 billion and include open AI's adoption of tra.

Um, it could be, in a sense, would, it's, if, if this comes to pass, Amazon would include, they would be exploring commercial opportunities and potentially selling a corporate version of chat GPT, uh, on their service. Uh, it's, it's interesting because in a sense, Amazon's competing with Nvidia in terms of a AI chip markets, and a lot of people are actually, we, I think Dave Nicholson yesterday talked about more competition for Nvidia on this front. So not only Amazon, we've got Meta and others.

And so Mike, to go back to your original question, it's, I, it's, it's, it's head spinning. I don't, I don't, I don't quite know where this is going. I think they're all going in on open AI until they turn against them and compete against them like Microsoft did.

Um, it's, it, it's, it's, it's, it's just losing me to speechless actually. Well, I think you have spot on in that the Microsoft announcement was a precursor for all these other organizations to now invest in open ai. And one way of looking at this, Alan, is to say, Hey, you know, we're just passing the hat around for open ai, tossing a couple of billions, and away we go.

That is one way of looking at it. But, but, you know, there's a method, there's a method to Sam Altman's Madness, right? And I've said this before, I think I said it on yesterday's show.

Sam Altman is the pied piper of ai. He is the closest thing to Steve Jobs we've seen in the Valley since Steve, you know, unfortunately died way too young. Um, but here's the thing about all of these open AI deals, they're not that simple.

It's not just passing the hat and throw in a couple shingle, you know, sheckles from your pocket. It don't work like that. It, there's always a, you give me this, I'll give you that.

The money really doesn't change hands. We just move it from one spreadsheet back to your spreadsheet, from my ledger to your ledger. It's a give and take.

So for this $10 billion investment, and, and John, I don't know the particulars, I'm sure you do, but for that $10 billion investment, OpenAI is pledging to do, you know, $20 billion in hosting with AWS on train infrastructure or something. So for the 10 billion that, uh, Amazon's investing, they're going to get 20 billion back in in revenue. OpenAI is a company, you know, you mentioned the $750 billion valuation, God bless 'em, but they're on the hook for a trillion and a half dollars guys.

Yeah, Right? The ma it's just Yes, exactly. Don't go into one and half trillion.

They're still in. It goes to by a hundred percent. 4 trillion.

Exactly. So what are we, and, and you just, you add up all, uh, yeah, it's just like, there's like a tail game. This is, So one of the things I was trying to think about is, okay, so like, where are the places that could be a single point of failure for organizations that are adopting ai that are trying to use the different things?

So in, in one sense, you've got, okay, if OpenAI, um, does whatever the money shenanigans are with Amazon for the, um, what is it called? Traum chips Cranium. Yeah.

That gives you a choice between, supposedly between Traum and um, NVIDIA's chips. But it also Complete No, no, no, no. It's a different, it's different.

The tra or inference chips. Okay. And the inference chip, you got Amazon, Trane, you got Google's, uh, chips that they're making, and then of course, you're a Broadcom, Right?

Right. Those are probably the bigger inference chip competitors though. But though, what's his name, Jensen and, and Nvidia wanna get into that, but they clearly have the gp, the training chips, But they're also like, Google's chips are made by Broadcom, so it's a little fluid, and some people will use training through train versus the other chips that AWS has.

So it's a, it's, it's not that clean, Right? And so the, but the question is, if you look at what OpenAI does and how much money they are owing, this is my question, maybe it's a dumb question, but let's see. Um, and how much they're now owing by and trying to make up with these different partnerships.

If the partnerships fall apart or if it doesn't really work, what is their, what are they trying to prove that, what, what do they have to do to shore up their business, making their open AI system better and getting more people to adopt it and use it, and then that rush to get the more adoption, does that have any impact on what it actually does, which is influences how information is derived from data? And I think there's a little bit of a danger there. So there's more than a little bit of a danger.

There's a big danger, and it's not to open ai. And I'll explain to you in a second. First of all, what OpenAI is doing here is classic business.

They're locking down territory. They, this is akin to European explorers planting flags on every shore they landed on, right? OpenAI is in a land grab trying to grab as much of this market as they can.

The problem with this model is if open ai, Sam Altman has put open AI at the center of it, if it falters 60 plus percent of, of, of booked revenue in the future from, for Oracle, for all of Oracle is OpenAI, even Microsoft with a, what is it? Two $3 trillion valuation. A huge percentage, 30 something percent of their book data they're waiting to recognize comes from open ai, right?

There is so much open AI is on the books Core Weave, a huge percentage of core weaves revenue is, is open AI pledges. Now it's gonna be at Amazon too. You know, Amazon AWS is what, a 13 billion or 15 billion year business?

Something like that. You know, how much of that is going to be open ai uh, pledges. So this is, this is a setup for a classic fall where if it, I did this on Shimmy says a couple weeks ago I had the Jenga Tower, right, with open AI's tile right in the middle.

You pull out open AI's tile, the whole Tower falls, The whole thing collapses. Yep. 4 trillion that you astutely referenced, Alan Open AI is like hemorrhaging users to, to Gemini and, and to Claude.

Absolutely. And this is created panic within the company. But here's the one thing I, maybe this plays into it, I'm not sure, but somehow maybe this Johnny Eye relationship with OpenAI involves Amazon eventually, where you have some sort of Consumery device and Amazon is 60, but Amazon's talking about business version, not a consumer Version.

No, no commercial. No. They might use it for commercial purposes.

So, so let's wait, let's get Alistair in here for a second, since he's our new guest for the moment. Alistair, what's your take on this whole thing? Well, I've been looking as, as all of us have at the ridiculous amounts of money that are flying around in, uh, this generative AI space, these chat bots, and it, it comes back to me to where the heck is all the business value that this money is supposed to generate.

There's the gap for me is there's a whole chunk of money, and as you say, this is a, this is a land grab for compute capacity across the world by open AI and under them by core weave as well. But all of this compute capacity that's gotta be built for all of this spend, how the heck are we gonna actually get business value out of it? I don't, you know, still hemorrhaging cash, As you say.

Well, that's, that's what the Krishna Armand, Kris, yeah, Armand Krishna of IBM said mm-hmm. You know, 'cause we've talked, so forget just open AI for what they, what the, the powers that be. And it's the seven or eight people in that picture you talked about John, or in the time man of the year, it's the architects of ai.

They have them sitting on a beam like the workers, you know, there's the eight, I don't know if you guys have seen this one. It's a great shot. Anyway, they're talking about $8 trillion in data center build out over the next three to five years, $8 trillion.

A we don't have enough power generation to generate to, to, to power those $8 trillion. B, if you are gonna have, if you're gonna invest $8 trillion, what's a reasonable return? Right?

And you're not gonna make it up in one day, even though every five years you've gotta redo these data centers with the next generation of equipment and infrastructure. But so what's a reasonable rate of return? How much money can it make?

How much money can it make to, to justify that sort of cap CapEx? And, you know, it was one thing when Google and Amazon and Meta were using the, the hoards of money they've been sitting on since COVID to invest in this. com bubble, you know, and when, when, when stuff goes south, someone's going to get left holding, you know, with a short straw here.

But That's what you're starting to see evidence of it already with Oracle. Yeah. I Mean, well, that's what the government's doing, right?

It's they're gonna use the government money to do it. That's what the US Department of Energy is in charge of the whole, um, whatever the executive order was to build out the AI data centers in the us it's gonna be Taxpayers. Are you saying, are you saying the taxpayers are gonna get stuck with this at the end?

Is that where we're Going? I think that's a possibility. Well, we're already paying, we're gonna pay higher utility costs for it.

But, but guys, this isn't just a US problem. The EU is bought into this. Our friends in Saudi Arabia are, you know, head down into it.

This is that $8 trillion is not a US number. It's a world, I I should mention that the eight trillion's a worldwide number. I think the US number is closer to five.

So Alan, for the people who have forgotten, how does that Pi Piper story end again? He, he leads all the children outta the village, right? It doesn't end well, not, it doesn't end well and never, never To be seen, never to be seen again.

Yep. Yep. Mm-hmm.

It's, it's a scary thing. But that being said, you know what, you gotta give Sam Altman credit the same way you had to admire Steve Jobs, moxie and showmanship. So do you gotta admire Sam Altman too.

He's, he's, he's placed them in the middle of everything, pivotal in the middle of everything, a company that doesn't even have $20 billion in revenue, hemorrhaging money, like, like a, like a, a, a cut, you know, artery. And, and, and he, but they're in the middle of everything and, and doing these deals. 10 billion here, 200 billion here.

It's, it's, I mean, one day they'll do business school case studies on this And then, but to John, and they Could either triumph or a tragedy story. I'm sorry, Mike. I mean, look, maybe, maybe we're all, we don't see it that he does.

And, and maybe ai, when you start putting it in robotics and the GenX and everything else, and it puts us all outta work and we just sit on serbo chairs and grow fat, right? And Sam Altman looks crazy as a fox Maybe, or to John's point, you know, everybody moves over to LLMs from Google and Microsoft and smaller LLMs and open AI winds up being, you know, the biggest buble in history. That's a, that's a possibility actually.

Yeah, you Would, there Is a better, You know, when there is a PT Barnum element to it, right? So, I mean, Steve Jobs was a PT Barnum, but he, I mean, he did delivered in a sense, this one, we is yet to be known. And, and if you wanna be in the middle of everything, you also take a lot of risks by being in the middle of everything, you're, you're something Falls, falls aside or a kilter.

There's a lot of possibilities for things to go badly. I'm not saying it will, but it just raises alarms for me at least. Absolutely.

Well, it's Friday. Let's not end on a be note here. We got a lot of good stuff coming up on text and gang.

Um, let's take a quick break, Mike. We've got more. We've got Microsoft AI agent machinations.

I love that you're watching text and gang. You've Earned it. The spotlight, the responsibility, the weight of teams, companies, and entire industries fall on your shoulders.

Lives depend on your decisions, your home life included that work. You are protected physically and digitally. Nothing gets through your team without a fight.

But in a globally connected world, everyone sees you, including those who mean to cause you and your organization harm. And now home your sanctuary attackers see an opportunity. Your digital front door is wide open.

And what compromises your home can breach your board work. Because the devil's greatest trick isn't targeting your workplace firewall. It's convincing you that your personal life isn't at risk.

Black clerk, digital executive protection, defending the new attack surface your personal life. Hey folks, we're back and we're talking about the latest AI announcements from Microsoft involving VS code. And if you have been following Microsoft over the years, sometimes that M stands for mud as in clear s and Microsoft is basically talking about how they're going to embed this kinda agent control plane into BS code and their app dev strategy.

But at the same time, they're also, um, deprecating certain other AI related tools that they have created over the years. And it seems like they're trying to make a play to be kind of the central control mechanism for all AI agents related to anything to do with software development. Gina, I know you read this story.

What, I mean, what's your take on what's happening here and, and how did you read this thing? Well, I read it and, um, it looked like it was good stuff and bad stuff. So it looked like there was really good stuff.

People were happy that, um, in inte code, well, well, not the Intelli code, that the transcript seven is, it was all updated and it's great. And it's newer model, new newer language. Everybody's happy with that.

The co-pilot stuff is awesome, is obvious. They just want more co-pilot, um, subscriptions. They don't want really, um, they don't wanna give anything away for free like they were doing with the Intelli code.

But the, the rest of it is just like, it's such a total product market. My product marketer went crazy because reading this, because the, the person that wrote this was just like, the messaging sucks. This is the worst messaging ever.

We don't know what Agent HQ is supposed to be. We don't know what all of these terms mean. We don't know what it's going to do.

So that's why I had a bunch of questions about it. And, and, and for a platform for agents, there have to be tons of questions about it. I mean, I'm trying to build an agent right now, and I'm totally within that 85% of what I'm trying, the agent came out is wrong, and you have to go back and, and rewrite it.

And, um, there, there's so many security implications that I don't even think I've caught all of them with my dumb little agent that doesn't do anything. Um, so I'm terrified to put it out into the real world to see if It'll welcome to tomorrow. I think everyone has, you know what Gina, everyone I speak to has a very similar experience.

I tried to build an agent and it sucks, right? I, as, as a, when we were at AWS on the way in, in the cab, uh, from the airport, there was a sign, or maybe it was in the airport, Databricks, our, they said, our agents don't suck, But why don't we call 'em bots? That's the thing I don't understand.

So to have this are Bots, well, bots, bots have a bad connotation security wise, right? These would have a bad connotation security wise too, because they are bots. Yeah, right?

That's all they're doing. My, my little agent is nothing, you know, it's got one little piece of, of an open AI call, but it's basically calling a a, an open. Yeah, a, it just, it just, no, it's, it's just calling APIs.

It's not doing anything interesting. It's a freaking bot and it's full of security holes, even my little stupid agent. But I think that's, um, what the big deal if, if you're gonna put like this agent platform out, if, if the people who are going to write about it that really do understand all the Microsoft, um, um, releases and everything, and they're like, this, I have no idea what's going on.

'cause Microsoft didn't tell us what's going on, but they want you to use a platform to manage all your agents. Like how in the clear is that platform? Do you know what the platform's doing to the agents?

And there's secret agents behind it that ma that Microsoft wants you to use. And it was really confusing, Mike. Yeah.

So here's the, here's the deal. A couple of months ago at the big Microsoft Ignite conference, you know, agent HQ is a platform and a strategy, and away we go This week, agent HQ is a statement of direction. We don't know exactly what this thing is looking like.

And I got a feeling that maybe there's not a whole lot of code to go with it right now, John, this kind of feels like classic Microsoft. Oh, it just Sure does. Yeah.

That's funny. That's funny how they changed the narrative, like just a matter of of months. Yeah.

This is, this is to be expected. And you're right, it's clear as bud and it's, um, the, the whole thing, and I'm glad you mentioned this, Gina, the whole thing about AI agents and what, what they are bots, right? And it's just a marketing phrase.

Maybe it's skewed, maybe they think it's, it's more embraceable, but it's, it's, this is part of this larger issue that I have with AI agents and I, I belabor this, belabor this argument, but it's just, it's just too much overwhelming CIOs, CISOs, you just put, your heads must be spinning. Like, wait a second, I thought you said this was gonna be the case. Now you're saying this, it's like, how do you, how this is why the adoption of AI agents in some cases is slower than anticipated because there's just so much noise and so much spaghetti being hit against the wall.

So guys gather round, let me try to make sense of this for you. Here's the deal. Here's the deal.

First of all, in terms of AI agents as bots, today's AI agents, six months from now, a year from now, we're gonna look at them and say how quaint, how quaint they are. Because most of the things that we're calling agents today are bots or even API calls, right? They're making API call kind of things.

And that's not really what the future of AI agents are. The future of AI agents will be autonomous tools that are persistent and aren't one trick ponies. Right?

Now what we're seeing are what I'm calling one trick ponies, right? I want my AI agent to take my whole workflow and work that workflow from A to Z, not A to B, but that's the promise of AI agents, right? That they will go A to Z in terms of Microsoft, it is classic Microsoft.

Microsoft is never the first one to a party. Ever make a party? Who's the first person there?

There's always some people who come early, not me. I'm married 35 years. I've never been early to a party in 30, 40 years since I'm with my wife.

But that's another story. Microsoft's never early to a party, but they never, like, they, they hate to miss a party. And so all they're doing right now is enunciating the same broad message that Salesforce, uh, ServiceNow, Amazon, and the rest of these people are, are enunciating, which is, I wanna be the control plane for your agents.

I wanna be the garage where you park your agents. I wanna be the Kubernetes the orchestrator for your agents. And that's what this is, this is Microsoft saying as it relates to software development, we wanna be the control plane for your agents.

We wanna be the cobe, the orchestrator for your agents, you and everyone else, Microsoft. But what Microsoft is betting on is though they're never early to a party, they're always the best dressed, you know, and they come with the nicest gifts and they bear, you know, and they, and just by their sheer tenaciousness and and market presence, they're gonna make themselves one of the top three. And that's what they'll do.

Are you seeing that I will not be getting a pony from Microsoft this Christmas then? Is that what you're saying? No, there's, there's no pony in that room.

But, um, but that's, that's you talk about. And they don't know exactly what it's gonna look like, but they'll be there. I be You wanna hit something really significant in this Ellen, about how quaint these agents will look in the future.

Currently they look terrifying. They don't look quaint. They get things so wrong.

They are so riddled with, with risks and, and dangers in them. One of the things I pulled out of this announcement was the YOLO switch. You only live once.

Turn on the agents, let them loose. 'cause you are going to get destroyed by this, right? We are not in quaint agents.

We are in the terrifying early place where it was, uh, early radiation, right? Earlier in nuclear. Your kids are playing with uranium right now.

So This is not the way this Ordered the gremlins. It's not one happens. Yeah, well, but That's exactly what I thought, Mike.

But you're gonna, you know, but hopefully we'll get through this. You know, my grandmother used to tell us about raising kids. It's just a phase.

They'll get through it. This is, Now can I mention some, can I mention something really quick about, I mean, you mentioned tenacious, that's such a great word to describe the way Microsoft tries to convey ideas. I remember as a reporter, like when I was in, uh, mainstream publications, they would have you go up to, to Redmond and they would do these back to back meetings.

I'm sure Mike went through this, these back to back to back meetings all day. And it was a form of brainwashing. And by the end of the day, until He said yes, Yeah, until you actually were mimicking what they said, you were mouthing what they said.

You go to Apple, that you have a briefing and you have five different viewpoints from five different people at Microsoft, you'd have 20 people saying exactly the same thing and they would just drill it into you. Eventually they would make their point. So I think the reason they call agents agents instead of bots, and I think this is actually an important thing, right?

They want us to think of them as more human running off of being our coworker friend that's going digital Coworkers. Yes. The thing is, technically we're seeing what we saw with really bad malicious spots 20 years ago, right?

It's just a thing that can do a thing and maybe it can pull other things. I mean, I was writing agents to build, to put, um, Linux on servers back in the early two thousands. That's all that jumpstart kickstart was.

It was something that was written with some code that said, go and do the pre-work. Go and install this little bit after it's installed. Go and do the post work.

Now do a test. That's all that these agents are doing. They are bots.

And if we talk about them, like they are pieces of software to do specific pieces of things, and we want them to start interacting with each other to be autonomous pieces of software to do it instead of pretending that it's our coworker and we should have an agent managing the agents in a whole department of agents. This is just stupidity. Let's say, let's talk about having the right platform and having the bots.

I I I'm done with agents. I'm not calling with that anymore. So, so there's a method to their madness, right?

And so the software company's been banging a drum for years about getting paid based on the quote unquote business value of their software versus just selling you a license. And this aging provides a mechanism to kind of start billing you for the value of the perceived software versus just saying, here's a bot that's an extension of your existing license, But isn't an agent just the extension of your license. I would agree with you, bud.

I'm just telling you. Right? This is why I keep being a big company working.

These are the arguments I bring up in meetings. But one other thing though that I don't, I want to just, I'll leave us with this, is at the same time they're doing this, they're killing awful lot of the free stuff they were giving out. And again, that's the same old story, right?

The drug dealers in New York City used to do this first one's free, right? First one's free. They gave it to you for free to get hooked, and then now they pay.

So clearly Microsoft has decided that, hey, the time for giving out the free candy is over. Now people gotta pay for the cigarettes or whatever this is, right? And, and I I, if Microsoft is doing it, Google's gonna do it too.

Metal will do it. They're all gonna do it. We might be seeing the end of the, did we get, did we set the hook deep enough to start making the money here?

Because we all can't have open AI balance sheet. Something to think about. All right, let's take a break.

We're gonna come back to Sea Block. I think Alistair's gonna help us with this one. I can't wait.

You're watching Texture and Gang. 2026 marks a turning point. Artificial intelligence is no longer just a tool.

It's shaping industries, accelerating innovation and redefining how humans build, create, and solve problems from engineering and medicine to finance infrastructure and everyday life. AI has become one of the most influential forces on the planet. That's why for 2026, we recognize AI as tech Strong's person of the year not for what it replaces, but for what it enables a future built together humans and machines predict.

2026, join us. Hey folks, we're back and continuing a little bit on a theme, but, uh, the folks at Code Rabbit put out a report this week, an analysis of 270 pull requests that were made involving open source tools and concluded that the number of software and defects being generated by AI coding tools is much higher than what humans would be doing in a similar set of cases. And this goes well beyond vulnerabilities.

These are defects that essentially increase the technical debt to the point where, well, it's probably your application's gonna eventually break. But Alistair what's your take on this whole thing? Well, I think this, this study was, was good.

It was a, a great start to look at 270 pull requests. Uh, but you know, headlines from from a whole bunch of larger companies are saying that's how many pull requests we push through in a an hour or a day or whatever their headline is. Uh, AI coding tools we know are very popular.

We've seen statements from a lot of large companies saying that huge amounts of their current code bases written by ai. So it's kind of surprising then that we're also hearing stories of code quality decline. There's been a lot of anecdotal coverage of this, particularly talking about Windows 11.

Code quality was the one that I saw quite a bit. Uh, the discussion around maybe, uh, Microsoft needs to hold all of its feature development to fix all the things that have been broken recently in Windows 11. And I know correlation doesn't imply causality, but this does correlate highly with large amounts of AI generated code.

So I am kind of skeptical about how much of a problem we're generating here. Now, one of the perspectives is that if the coders AI generated and it's going to be AI maintained, doesn't matter if human can't read it so much. And this comes back to what we're covering in our last story, is how much do we trust the, uh, bots, the, uh, agents that are doing this coding for us?

And how much are we just gonna hand the keys over for the entirety of our software development? Uh, do these technical debt issues that affect human coders also affect AI based coders? But I think fundamentally we are seeing that these, uh, AI coding tools are not up to the quality of a good software developer.

They are still giving us more the intern based level of, of coding. Uh, I talked about this on a recent episode of the Tech Field Day podcast as well, that the knowledge of the enterprise scale building of applications doesn't seem to be nearly as trainable in these AI coding tools. As a, you'd hope that the very senior engineers who have vast experience haven't necessarily codified that experience in a way that can be used to train an ai.

And that's why we're not seeing awesome application architecture being built by many of these coding tools. Uh, I liked also one of the commentary in this article, which was around that AI coding tools are producing greater efficiency for developers. Uh, maybe not.

There was a study conducted midyear where the developers thought they were more productive using AI coding tools, but they com the study group actually completed tasks faster if they didn't use AI coding tools. So is this productivity and illusion or is it that we have to use these AI coding tools? 'cause we've spent lots of money on building ai.

Are we actually getting something useful out of the AI if it's taking longer and producing worse product? Well, that was mid-year. The world's changed since then.

But let me, you know, I recently did an interview with the CTO over at Veracode, right? Veracode's, big AppSec company and anno, you know, anonymizing the data that they gather from their hundreds, thousands of clients. They, they actually have been putting out periodic reports on the security status and vulnerability frequency of AI generated code versus humans and so forth.

Um, and you are right, uh, the AI generated code throws off a lot of security vulnera, you wanna call 'em security vulnerabilities, bugs, defects, whatever you wanna call it. However, according to Veracode, and you know, they don't, this is just what their data shows. What I said about midyear is no joke.

The quality of code that AI is generating is, is on like a, a, a, a 45 degree upward plane. And just recently, like in September or October, it actually crossed the level of human generated code in terms of vulnerabilities and security defects, right? Let us not fool ourselves.

Yes, the most senior best developers develop better code than junior developers. I don't think anyone will argue that, but when you take the totality of human generated code and baseline that to the totality of, of AI generated code, they, they're, right now they're roughly on par. But one is at a 45 degree angle continuing to improve.

And one is sino, you know, has been level for a while now. So the, the question is, what about six months from now? What about nine months from now?

But, but here's the, here's the soft white underbelly that I think gets dangerous. And it's a birds and the bees question. Yes, our best developers are better than AI can code right now, but if we don't train and let our not best developers evolve, where's the next generation of best developers coming from?

Right? We don't hatch these people in nurseries. We, they, they learn on the job.

And so that's the real issue is we can't say just let the best people develop. We've gotta, you know, we, we talk about things like human in the loop. We've gotta keep the human in the loop for a, a lot of reasons.

That being one of 'em. So that these developers do learn to use AI as a tool, but become better developers, right? Otherwise, this'll be the last generation of best developers we have.

I also think that we need to use AI to review the code created by ai. But there's a a tendency where I see where people are using the same AI that they used to write the code to review the code. And that's a fundamental mistake.

'cause that's just gonna, yeah, Best practices has to emerge where you, if you're gonna use AI to check ai, you gotta have a different AI on it. Mm-hmm. But you still, even in that scenario, Mike, you need a human in the loop, Right?

And I'm also, you know, I am not the vice president of the United States, but I will engage in a little conspiracy theory here. The AI code is, you know, extraordinarily verbose and seems to require a lot more computing capacity to run. And, you know, that seems to play into the benefits of the cloud service providers who want you to maybe consume more infrastructure than ever.

I'm not saying that's a deliberate outcome, but it sure is coincidental. I thought you were gonna tell me you're merging, merging with a fusion energy company. Go ahead.

I digress. I think it's, I think it's a bigger, it's a bigger picture too, right? Like when you look at these surveys, like, like what Alistair said, it was a really small amount of, for this story, a very small amount of pull requests that they looked at.

Um, and just in general, what we're talking about with the 45 degree co, uh, AI's getting better and we're, we're dropping off here, but just, Well, we're not dropping off, we're not getting worse Gina. But human coding is steady. So, so, but what I was getting at is, I, I think there's specific things that AI code is probably really, really good at.

I think the whole mainframe industry is one place, being able to go through trillions of lines of code to, to point out this is duplicated, this is duplicated, nobody knows what this means. And really help the newer generation, uh, COBOL developers that also fluent in different languages, excuse me, to help them understand what the code was to help them see where to attack. I think that's really good, even with the older code too.

But like, um, with, you don't have developers that have the, the younger developers having the knowledge of how things break and, and that puts a whole piece into us. I don't, I think that AI code, written code and checking the code, there's all sorts of places. It's really great for, like, if we're doing infrastructure as code, that changes so rapidly depending on the different pieces of hardware that are involved in an inventory of whatever an organization has.

But being able to write the basic blocks of this is a Dell server, this is a, this type of interconnect, this and that. It should be able to get to that and be easy to check and be easy to upgrade once you have, you know, what the differences are and save lots of time. Maybe help people learn to code.

I think that that should always be that human in the middle that you're talking about. This is something that, that an AI is not going to identify because there's not going to be a prompt to help it identify because the human will know that it's a problem until they go to run. It Could be, I mean, certainly teaching people to code, I, I think AI is a great tool for teachers, especially like code, like, you know, that kind of teaching, you know, very, uh, rote kind of stuff like that.

Um, I mean it, yeah, that's certainly a use case, but I think people are using it, you know, this vibe coding thing. I, I remember the first time I heard it here on the gang and I was like, I was thinking beach boys and, um, but who knew? So, so how do you envision this working in the future?

'cause in my mind it might play out this way using your statement that says, you know, six months from now the AI will be a lot better. So are we all gonna get alerts from the new AI saying that the old AI was stupid and an idiot and I gotta go fix all this code? How does that kind of play?

Well let me ask you, do you ever get an alert from a company that says, I'm sorry, one of our junior developers developed really crappy code, but we put one of our good guys on it and it's better now? No, you're just gonna get an update. Hey, there's a new update click.

It's better, it's faster improved. It's stronger. You'll get Esoteric press release from Microsoft.

Yeah. You know, We'll see We've got accumulative update for you. You know, they do that.

Yeah. But These updates will, will also, I mean there's an element here of not just the individual ais get getting better, but as, as you're saying, you, you don't use the same AI to do the code review that you use to do the initial writing. You also don't use the same AI to do the optimization of the larger code base that you use to write the individual features.

And so as we see some maturity, we will see those, those ais uh, getting better at doing the more strategic element. Starting with a, describe how this should work rather than describe the single feature that you want to have, or it'll be describe the path, the, the story. If you're gonna go agile gi give me the story of the user journey through this application and, and the AI will generate from.

So when you're talking about human in the loop, it may well be that we see a very big shift from that human looking at the individual pieces of code and needing to maintain those pieces of code to ai. Looking after all of that, in which case the, the those code is, is great, but we're gonna want to then optimize it, do the functionality that we've previously done with a compiler. It's gonna be to optimize and, and enhance the, the code that's actually being written by the coding AI to be better at execution time and development and production execution are two completely different spaces.

You may well not do that optimization until you're getting close to that continuous deployment phase in your development pipeline. So you absolutely, we will be seeing people using different ais for different purposes at different phases of their development as you use different humans for the same thing. I think this is where that 45 degree uplift and effectiveness is gonna be useful and eventually there'll be a critical point where the AI code is better, but we're not there yet.

We're still seeing lots of that legacy code that we're going to see an improvement in sometime in the non determinant future. Right Now, I know we have a predict conference coming up, but um, here's my prediction for the next year. We're gonna see a lot more stories that start with Company X had experienced application outage because some software updates somewhere didn't go as planned and a lot of it will be traced back to AI code.

Do you think we could retroactively bring the CrowdStrike on it? That will seem like, Is that because look, what's that, That, that, that may seem like, you know, a mere footnote in history compared to what we might see. Okay.

I mean it's not like it hasn't happened with humans coding that and that. Mm-hmm. You know, so, and that's a lesson there.

Should we hold AI to a different, uh, level Standard Standard than we do humans just because it's ai? Should it be perfect? 'cause certainly human code's not perfect.

I think we have done, to our credit a better job of reviewing code and kind of preventing disasters. And when there are disasters minimizing the impact of that, I think in the age of ai we may be trusting too much in the machine and not having those review cycles in place. And then we're, 'cause we're all in the name of productivity and speed and break things at all costs, maybe we're not paying enough attention.

I Think someone will have to be held accountable for it, right? Because if the, if there is revenue lost because the code was incorrect or even worse human life lost or that kind of thing, God forbid that it will have to go through, at least in the us the same courts to say you're at fault and you are required to pay this much, whether it's AI or whether it's a human. That's gonna have to happen.

Mm-hmm. It'll be interesting. Hey, we're about out of time.

What a great discussion today gang members. Thank you so much for coming on. Uh, as Mike mentioned, we do have our predict event coming where the, the good analysts of the RUM group are going to be leading the charge this year in our predictions for 2026.

That's on January 15th after the holidays. com right now. Click on Predict.

com and you could register for that event. We've got some great, great predictions and analyst, uh, kind of look aheads. I'll also mention that later today I'm gonna be doing my year end Shimmy says, my last shimmy says of the year where I'm gonna give kind of my look ahead, uh, for what's been an a year for the record books, that's for sure.

Well, we will not forget 2025. Um, so stay tuned for that. I also just wanna also point out, I'm assuming says I did yesterday Thursday on should we break up Big Tech?

I saw a debate recently, a video of a debate, uh, pro and con on this, and it's fascinating. So check out that Shimmy says, um, a what about the next tech field day? Are you guys off till the new year?

Yes, I, I'm, uh, doing the planning and organization for the next tech field day, which will be AI Infrastructure Field Day. Uh, that will be at the end of January 28th, the 30th. It's gonna be a nice big busier event back in Silicon Valley.

Very cool. John, I see a CES in your future. Oh, I think I'm gonna stop by for a couple days.

I actually got a a, I reached out to Nvidia just to reach out to me. They're gonna be doing a big announcement the day, two days before CES starts with Jensen. I don't know what it's gonna be, but lasted, he talked about physical ai, so maybe something along those lines.

Very cool. Gina, what do you have? I'll be at, um, Alistair's Tech Field Day event, the AI infrastructure next month.

Very cool. All right, Mike, I know what you've got. You're waiting for pictures and catchers to report And I am not leaving the Tri-state area for at least another 30 Days.

Good for you. This is great. Enjoy it.

Well, it's not warm though. All right. I hope wherever, whatever you are doing for the rest of this year and the holidays is, is good and enjoyable.

It's a time to recharge refresh, but 2026 isn't gonna slow down, so stay tuned. We here at Techron, at Techron Gang and all of Techron Fu Tech Field Day, we're gonna try to bring it to you. So check it out.

But for now, this is Alan Schmo and we're out.