AI Revolution in the Workplace: Jobs, Risks and Trust | TSG Ep. 923

Hey, is the AI wrecking ball wreaking havoc in the workplace? You're watching Textron Gang. Hey everyone, it's Alan Shimo.

Happy Monday. Wow, that weekend, it feels like it was just Friday, but here we are on Monday and we are a, got one of an amazing lineup of Techron gang people to talk about some amazing stuff with, we were talking in the Green Room before coming on. com in 20 13, 20 14.

And here we are, 11 years going on 12 years later. Still talking DevOps now with AI and everything else, but whatever we're talking about, I'm just happy to see them and have them be part of it. Of course, Mitch was there too with me.

This is even before Mitchell. Mitchell was still at the Cable Labs company, but I was working in DevOps, so it was good stuff. And we got a great, great panel today.

They're Andy, Mitch, and Sanji are joined by all stars. Tracy Reagan, Kimberly Bates, Kimberly. It's so good to see you, Mike.

And of course, everybody's favorite Yankee fan. They won last night. Mike.

Mike, you go man. There you go. Um, so look, as usual, we've got a lot of AI to talk about today.

You know, I was, I, last week I was out in Napa at this Jfr s warmup event, and it just, it boggles my mind how quickly this has so deeply rooted into our, into everything we're doing in tech anyway, and, and kind of turned it on its head. I, I use the term AI wrecking ball. 'cause I think in some ways it is a wrecking ball, but sometimes you gotta, you know, it's like when they're building new hotels on the Vegas strip, they gotta implode the old stuff to build that new stuff.

Not a, not the way we usually like to do tech. But Mike, what do, what do we got? All right.

Well there, our folks seem to be pushing the limitations of AI, is the way I would kind of look at it. We have not one, not two, but three separate studies looking at folks who are saying that they keep starting to use some sort of project using ai, but they can't complete it 'cause it doesn't quite fit for purpose or whatever they were trying to do. The AI just wasn't up to the task.

Or it could be they were just didn't have the skills to get there because, well, it turns out you gotta be pretty savvy about how to use this stuff. Andy, I know you've been playing around with this stuff, but a lot of folks I talked to when we've seen them gone from we need you to use AI to now they're using AI and they're reporting that they're slightly disappointed. Yeah.

Look, I mean, the AI revolution's absolutely upon us. There's no doubt about it. These surveys are really interesting in some of the stuff they discover.

Um, you know, I really, uh, it's non-trivial that we're gonna lose 45% of certain jobs. Uh, uh, of that 45% of jobs are gonna go away. Types of jobs.

Interesting. I really, you know, when you think about clerical jobs and, and low skill, you know, we can talk about whether that is actually a thing or not. Uh, but one of the things that that's gonna happen here is paralegals are gonna go away.

This is really interesting to me because it signals the complexity is where we're gonna get a lot of advantage from if we get advantage. 'cause you are right Mike, there's a lot of people trying and failing, but everyone's doing it. Another thing that came out of this res research was the lack of compliance, governance control.

There's a lot of people using it. Sometimes they're using official, uh, l lms. Sometimes they're not.

I mean, I'll tell you mate, I just bought a new laptop. I've got a button here which fires up copilot, right? What if I, I happen to have a corporate license to copilot, but what if I didn't?

Well, all of a sudden, maybe I'm doing strategy work, I'm doing competitive intelligence. Maybe I'm doing product planning and I'm pumping my stuff into a public LLM and I wonder if my competitors can go and ask that LLM what I'm doing. You know, there's a lot of concern here about the jobs, and I get that there's a lot of concern about trust and using this AI and creating AI slop.

Um, and you know, we're seeing this a lot in sort of social media, but when we do this, uh, legal pleadings are being kicked out of court because they're creating cases that don't exist as precedent. And so there's a lot of concern. You know, I think we've gotta get a lot more controls around what we're doing.

Uh, a lot more understanding of who's doing what with AI ways to build up trust. And we know there are things like, you know, human in the middle and rag and stuff, but we've gotta get ways to address the findings of these surveys. Um, finally, the last thing I think is really cool in this article by John Schwartz is that it, there are five ways to address this and some of them are pretty obvious training, but others, you know, there's some good advice in here.

So I think we've got a good way out. But I'm very concerned about the slop, about the governance and compliance and about the lack of trust that we have in these, in these tools. Andy, that is fabulous.

What you had to say. Um, I've been traveling a lot the last two months. Um, personal travel.

And in that travel, I've been talking to a whole lot of non technologists. So I'm getting, I got outta my bubble. I've been outta my bubble and I'm talking, and I've been with a lot of Aussies, new Zealanders, um, Canadians, Germans, et cetera.

And those conversations, when they have found out that I've been in, I know what I know about ai, which is about this much, but yet I know that much more. Sorry, I'm doing the Italian thing. I, this wasn't even in Italy.

Um, I know so much more than the rest of these people. And that's scary. And so what they globbed onto me to say, well, what does this mean?

What does this mean? What does this mean? I'm a teacher.

I'm a I am a, a lawyer. I am a researcher on all these things. And so when you, when I come from that visual of there, I, and I read this, especially the Audacity research report.

I came away with a couple things in those conversations. One, the LMS and all that work is really good at data, memory and recall. Really, it's good.

Well, it's almost really good because it makes stuff up. Right? And I disagree with the paralegal thing.

And the reason, we reason is because the next thing it has to do is analysis. And it's, and it needs that systems architect, that senior coder, that expert marketing person, that level of analysis to say, is this right or this wrong? Which is exactly what that data is pointing into.

And then the last thing, which is, I appreciate because I'm, I'm in my sixties guys, and that last thing is called wisdom. And you know, as you get into this stuff, you've been, you've, you've hit your head up against enough things to the point that you kind of go, you sit back and you kind of go, okay, seen this one before. Let me not be a stick in the wood here, but let me have an open mind and think about where this goes.

And so those three levels of knowledge, data, expertise, and how we apply to this, the LLM is just coming outta college with data, memory, memorization and recall. And we're asking it to be a 60-year-old, 70-year-old, 80-year-old wisdom G whatever his name is God fault or whatever. Um, listen, kind of wi wi wizard.

And that just is not gonna happen overnight, guys. I'm sorry. Well, you know, that the AI slap issue that assumed we weren't already producing slope.

That's beside the point. Um, without ai, you know, it, it's, it's, it's the, the whole prediction around jobs always entertains me. 'cause it's kinda like, you know, the weather, nobody's doing anything about it.

Nobody's laying off all these people yet, right? They're just pontificating about what's gonna, what's gonna matter. And here we're talking about the disruptive nature of ai, but we're not sure what, where to use it and where it's really gonna be effective.

So it's a little hard to predict that accurately. Um, I think to your point, Kimberly, combining AI with someone who has a great knowledge base is powerful. We see that in vi we see that in coding just to, whether it's vibe or not, Sanjeev can wax on about that, who's got a ton of experience there.

And, uh, but it's also people who have a, a real, really depth in a skill that are used to kind of some structured type of work and, and how to work with a, uh, an LLM or, um, interface like that. Get a lot out of it. I think the other end of the spectrum is, yeah, there's kind of mi folks in the middle of their career.

I think people coming outta college are gonna kick ass with ai. Mm-hmm. They're growing up with it.

They're learning it, they're using it, they're doing it. They're gonna, they're gonna, it doesn't matter. They don't have that huge base.

They're gonna have a huge base of how to use ai. 'cause they're, they're playing around with it all the time. They're using it for all kinds of stuff.

We just have to help the middle kind of figure out their transition too. Yeah. I think some Thoughts on this.

Yeah. Yeah. I had to Go ahead, Alan.

No, No. Sanje. You go first.

Yeah. Uh, uh, two, two points. I wanna make one on Mitch.

I think we are spot on when it comes to what are the college folks going to do? My son just came outta college and has seen work. My daughter's in college and seeing how she's using or not using ai.

I think people panic a little bit saying, Hey, the, the entry level jobs will change or will go away. I don't think they will go away. They will change.

'cause I remember what the entry level jobs was when I started. Right. You know, have people handing me 64 floppy disks to say, go install this system.

You needed to know the, the, the, uh, the, the geometry of your hard drive in order to install Linux. Back then that was entry-level job. Then today it's double click on download now and that's the entry-level job.

And then it's configuring and they'll just get elevated beyond where they were the the second. So, so I, I don't think we need to panic. The second point I wanna make Kimberly, which is, is on what you said.

One of the things I'm recommending as I'm traveling, I just came back from the Bay area, met with a bunch of startups, which are in the AI space, dealing with infrastructure or, or with coding. Most of them in, in one of those two spaces. One of the things I'm recommending all of them do is read the systems thinking book.

Uh, by, I think it's by me. I forget the, I'm, I'm, I'm looking around for it. I can't find it on my desk.

It's probably behind me in the bookshelf. 'cause I'm bad with names. But it's from the, it's a, it's a pretty old book.

It's been around forever. But people need to realize that you are inserting something into a system which is already in place. The system could be a factory, it could be a law firm, it could be a, you know, places where we work could be a household.

It's a system which works. It attaches processes and workflows. When you bring in something new, all the workflows need to change.

And if you don't change them, either the system will fail or whatever you're trying to insert will fail. Which, what the McKenzie report showed us, right. 95% of projects are failing because they cannot get integrated into the system without changing the system itself.

I, I think we have a lot of thinking we need to do and go back to some fundamental systems thinking. That's interesting because I was reading when in that, um, Udacity report, they talked about the thing that it was lacking in. Or was it that one or something else I was reading that was lacking inducing.

I'm sorry, I'm talking about vibe coding is the lack of the architectural design that goes with it. Mm-hmm. I, I know that's another topic.

So I'll drop. Did I ever tell you about, did I ever tell you about my time in the construction industry? No, I don't think we've heard about, I'm about third year.

I'm in third year of college and I come home and tell my mom, mom, I think I'm gonna take, we didn't call it gap years then wasn't, I didn't come from that kind of house, but I said, mom, I'm gonna take a year off and make some money. I'm gonna talk to my uncle who's in construction and let him get me a job in the construction industry. Mom, my mom being a Jewish mom, three worst words you could say is I'm not finishing school.

So they get me a job in construction. My construction, I'll tell you about it another time. My construction career lasted two days and then I quit after school.

But here's what I found out about. That's good. 'cause you don't know the difference between three and four words either.

So. Well, exactly. That was a clue.

But I didn't leave Mike. I gotta tell you the truth. I met a lot of your friends.

They were recently over from Ireland. They took me to the Blarney Stone for lunch. It was liquid lunches and it was a whole story.

But that being said, what I found out about construction, it's messy. There's a reason why they wear those hard hats. Mm-hmm.

Because s**t gets blown up and there's stuff falling and they're, they're breaking down to reconstruct and it's such a mess. It's hard to imagine what a beautiful building or what a beautiful thing they're building in the midst of the construction. 'cause you, you literally, you can't see the forest through the trees because you are in it.

You, you, you are working on this little piece and you might be banging and tearing that piece up to make something beautiful, but you don't see it till it's done. That's where we are in ai. That's where we are in ai.

We are in the midst of a major renovation. The likes of which we have not seen before. Not cloud, maybe the internet itself.

Yeah, You are, you are spot on. com era of the, if you want an analogy, right? com, but done, right.

Yeah. Right. com happened during the messy period of the internet.

com is happening now. When the internet is well understood. The systems work.

They know how to unplug things. They don't need to go disrupt the system to make the company work. Yeah.

So drawing that right Now, we need a little hard hat around this AI stuff. But, but it will, it will shake itself out. It will normalize.

People aren't going away. I, I've written about this, I've said this, I know Tracy, you're a big fan of ar you can't wait. But when did that happen?

No, I know, but, but, but here's the thing. I've been saying. AI will never replace the spark that resides in us.

The creativity that humans bring that have allowed us to evolve over these hundreds of thousands depending who you believe, whatever, I'm not gonna get into it, but has made humans. Humans. Is that spark, is that ability to use tools, is that creativity.

AI is perhaps maybe the greatest tool we've ever invented. So let me just chime in for a minute here. Go ahead.

I don't think that AI is one of the biggest changes that we've ever seen. Hmm. I think it's a change that's happening faster than most changes.

But we have gone through far more disruptive, uh, periods in time. Uh, think about going from when everybody had a green screen in the mainframe and we went to open systems talk about breaking down software. That was a massive overload.

It was a, the every bank, every insurance company, they couldn't find enough people to start writing open systems software. Everybody was trying to get off the mainframe. That allowed us to be where we are today.

So Along the mainframe, Mainframe is still around and it's safer and a lot of government or will never get off of the mainframe and a lot of banks. But, and it never went away. But what I'm saying was, I would say, and maybe I could be wrong, 90% of of applications now are not running on ZOS now.

Well, No, very few are running on Xerox that I'll give you Exactly. God bless though. These, these are always destruct.

Everything is always disruptive in software. That's what we do. We, uh, we love to disrupt ourselves and the people on the, the users get freaked out every time we do it.

Every single time we do it. People had to learn to use a mouse talk about difficult, I can remember classes for, for organizations teach, teaching people how to use a mouse. So AI is disruptive, software is disruptive.

The difference here is I think the speed and the control, because I believe that a lot of people see that they're losing control. And that's what freaks 'em out. Even stuff, you know, going through the mainframe to the Computer computer, I, I still think there is, but there is a lot of frustration out there.

And I was just talking to somebody about this this morning. They said, you know, I use AI to do the first draft of all my posts and everything I do, but I, I do it because it just makes me angry. And then I fix it and make it right.

And again, it kind of makes it better. And it, and so he was using it as a motivational tool, but basically he was also saying that this thing doesn't do what it needs to do for him. So I collaborate, I collaborate with Ai.

Don't blame the tool. Blame the tool. Blame just, I, I just wanna pick up a little bit what Tracy was laying down there.

Um, because I actually am, I'm sort of halfway between you all. Uh, I see the revolution, I see the opportunity. Uh, I see it certainly for no knowns.

And I've talked about this on, on Textron before, the idea of known knowns. Uh, the LLM AI especially is essentially just autocorrect, right? It has to, it has to know what's right to be able to do the work.

There's a lot of management by magazine where people are saying, well, AI is like a human, it knows things and can work things out. And this comes back also to what you were saying, Alan, about the spark, the known knowns. It's really good at, if I wanted to write a blog about DevOps and value stream management using my content from the last 15 odd years as input, it's gonna do really well.

If I want it to create a product vision and roadmap that's gonna create innovative new capabilities in my software, eh, I might give it one pass, but I don't think it's gonna come up with ideas like my team would. And it again, makes me think of a Mara's law, the idea that we overestimate the effect of a technology in the short term term, but we underestimate the impact on the long term. So I think we might all be right here.

I think we are overestimating what it can do right now. And there's a lot of management by magazine. And this is why we don't trust it, because it's creating lop because we're overestimating what it can do.

But the things it can do now are amazing. And we'll probably get to a point where it's doing things we didn't even think it would. Like the internet does, like mobile computing does, like distributed computing did.

It got to a point eventually where it's like, oh wow, I never thought we'd get here. But we're overestimating it in the short term. And that's, and that's causing a lot of these problems.

I think, you know, I Just, and you on, and what Tracy was saying, I would take, I would say that there's two, there's a couple of other major technological innovations that were much bigger than the mainframe to os and I'm not a programmer, so I don't have that perspective Tracy. Um, and that is the industrial revolution. Um, because I look at that as what that changed is going from a farming community, cities, and we actually moved people because of that technologically or, or you personally moving those people.

And that's what we're doing also with ai. We are moving them to other jobs that are not that technology And, and to other ways. I, I mean, I'll tell you.

Right, right. And the other one was when we went from batch to online transactions, because again, when we went to online transactions, we moved people from what they were doing, let's say inventory management. And they were calculating all the numbers.

Now all of a sudden the system is spitting out information about how to manage that, or we're actually doing our, using our own ATM so we don't have to walk into the bank. So that changed with the same thing we started out talking about is how we're disrupting the workplace at the same time. And then there was fear at the same time.

There was a amount of fear at the same time on the people. So to me, those were the two big eras that I see. I think we have to open our apertures a bit.

AI is not just for tech folk. Mm-hmm. AI is going to be civilization wide.

It is humanity wide kind of impact. You know, I I I said at the top of the shy spent last week out in, uh, Napa Valley at Jfr Swamp up. And we were there with people from Nvidia and ServiceNow and Sonar and, and others and Jfr.

And I don't know if I drank too much wine in Napa or the AI Kool-Aid, but between them both. I, I'm all in. I'm all in.

I think I, the things I saw, the things I heard from the Nvidia people, and granted, you know, they, they've got a course in this race, obviously, but it ain't their GPUs, it's their software. It's their software. Um, what ServiceNow is doing, what, what we're seeing every day these giant companies doing.

But Mitch, I'll, I'll, so I'm gonna come back to something you said. The fact of the matter is we've shed over a hundred thousand jobs in the tech sector in the last 18 to 24 months. Well over a hundred thousand jobs.

Mm-hmm. Just like, I'm not, I don't even want to get into it, but just like we revised month to month and quarter to quarter, year to year, our, our labor board, you know, job growth here in the us I think we're going to need a little perspective to look back and say how many of those hundred thousand jobs plus jobs were actually eliminated, at least partially due to AI or betting on ai, I think Right. AI on the cart.

I think Mike said zero. Mike says it giving me a very New York zero. That's how much That's Is.

This is like the, uh, I'm gonna jump on the bandwagon and call my project a transformation project. 'cause that's what we're funding right now. That's what, that's what's happened with these job layoffs that we've claimed, or because of AI is we had an AI strategy, we won't need all these people, whether it's Zuckerberg or it's, you know, x, y, Z person.

They're, they're making those claims. Part of it is just an excuse to, to manage Wall Street and manage wall expectations. Yes.

Those changes will come, will come. But I'm with Mike. I'm on the zero scale of Yeah, none of those are, well, no, You gotta go like this.

Mid zero. You mean they Don't wanna stand up in front of you mean they don't wanna stand up in front of Wall Street and say, we sucked at this and we over hired and you know, We're ai, we're on bandwagon. We're we, we don't need half those people.

Let's, let's do this. So by the way, it really helped our, our, our numbers this quarter. Yeah.

Alright. Hey guys, we can talk about this one all day, but we got more to talk about. I'm gonna pull it, call it on this.

We're gonna take a quick break and we're gonna come back. I'm gonna talk about the reality of AI coding or is it all a hallucination? You're watching Textron get Discover Textron Group, the epicenter of tech innovation.

We are your go-to for reaching IT leaders and practitioners worldwide. Our secret impactful content that sparks awareness, engagement, and top quality leads with us. You'll access editorial websites, streaming videos, virtual events, custom content analyst research and more.

Join our satisfied clients, let's revolutionize your tech journey. Contact us today and tell your story to the world in the most powerful way with Textron Group. Alright folks, we're back.

And yes, we are continuing on a theme. There's another report talking about, well, just how are folks using AI specifically when it comes to software engineering? And it kind of suggests that while everybody's all into Alan's point, but they are running into what might be called uhy, systemic issues, things that are bottlenecks and DevOps workflows that AI has not yet addressed and may never address.

And as a result, the amount of software being built isn't all that much faster than it was before. Whereas one of the wags put in, you know, people are running harder into the same walls. They always ran into Mitch.

Um, you know, what's your take here? What's real Well the low hanging fruit of, of AI news is I can generate code with ai. No s**t.

Sherlock, you can generate code with ai. That's not really the big benefit of, of ai. Yes, it is.

It's, it, it is helping us generate code. Sometimes a lot of code, sometimes all the code, but the real benefit that that's one person doing one task. And that's what this survey was really showing us is people are thinking beyond just generating code of where AI can help them.

Now it was interesting kinda juxtaposition. This, this, uh, survey was, I think done by Uplevel, if I remember the organization, but 87% said that their organization's prepared or very prepared to adopt ai, AI as part of their strategy in, in their software engineering process. But where they're looking for help, where they're not seeing it is what I, where I take this from the survey is helping them fix technical debt.

There's a great place for AI having a clear strategy for how they're going to use ai, um, data security, privacy, kind of the things we would know and expect to be issues. But they also want to use AI to drive operational efficiency, not just development efficiency. They want it also to help them do not just work faster, but uh, innovate faster.

Do do really meaningful things that are gonna be valuable to customers and to the business. So I, I liked this survey both because it wasn't just somebody that had a product to sell. It was yes they do, but it it's also that it's looking beyond just the developer and the task that a developer does and saying, how does this help us with the entire software development lifecycle, DevOps operations, you name it.

And let's look at where we can get value. com, like while you were at, uh, swamp Up Allen. And that's what it's, this is all about is ai.

You know, that was every part of the announcement with Splunk, but it was all about operational efficiency and ways that they can are starting to use ai. But of course it's also about the data that drives it. Anybody I, I'll jump in.

So a factoid I picked up while I was out there last week, according to Gartner, according to Gartner, if you believe him or not, Microsoft and Google acknowledge that up to 30% of the software they're pushing right now internally, externally is AI derived code. 40%. 40% of CIOs are asking for more budget because they're being pressured by their boards to do more ai.

Mm-hmm. Mm-hmm. Right?

This is, this is not a bottom up, this is a top down, right? The discussion we used to have around DevOps a lot, but it's a top down kind of thing. Um, I, I don't know this particular survey, but other surveys I've seen, 70 upwards of 75% or higher of organizations are already using AI in coding.

So trains left the station guys, if you didn't buy your ticket, you better be able to run real fast. That's Low. I think everybody is in some form.

Some way they're using AI today it's, it's 99 9. Even if they don't have a AI coding tool, they're putting, you know, using chat GPT or perplexity or something to get that Shadow shadow AI Question for Out there, uh, on what you said. So sorry Kimberly, I, I want to, uh, uh, Sunar pitch I for Google from Google appeared on the Lex Friedman show and Lex asked him, how is AI coding going and sun?

There was very candid, he said 30% of the code being generated is from ai, but the caveat that has increased their velocity only by 10%. Mm-hmm. Because the rest of the pipeline is still slow.

So is it empty calories? What, what's the story? My wife hates empty calories.

Well, well, but, but it gets worse than that because a lot of the code being generated is very verbose and very, uh, large and it winds up increasing the amount of technical debt you gotta service. And then, you know, as Tracy has noted on previous shows, the people who are supposed to debug that code never saw that code in the first place. And so they don't really know how to debug it and there's no AI agent for them to debug it with.

So they're kind of like, what am I supposed to do with this thing? So I've got a question for you Dev people. How do you define what is technical debt when it comes to the DevOps side of this?

I, I kind of know technical debt on the infrastructure side, but that's you of it. And how will, how do you think or how could AI assist in fixing technical debt? Think of it as that.

There's gotta be analysis on that. Yeah. Yeah.

There's, There's a lot of studies around this in, and a lot of it falls into maintenance just kind of keeping things current, upgrading this version of Linux and, and all the things that it, it impacts. It's also the, the, the features of the bugs that you don't get to that people have asked for that kind of stack up in the back of the room. That we just, other things have taken priority.

Um, and, and a lot of times it doesn't get addressed until something else needs it to change. So in order to get this new feature in, I've gotta upgrade the database. Okay, now I've gotta go back and take the time to do the technical debt work to be able to then go to add this new feature to the database or to the, You know, where else we're seeing technical debt build building up in the transition from to platform engineering and internal developer platforms.

Man, there's a lot of debt sitting on the books right there. Yeah. Right.

And as we, as large organizations, especially technical debts usually a large company problem. Not that smaller companies don't have it, every Team has it. It's just Right.

But the large companies, they're just wallowing. We've had Their systems, the bigger the data, let's put it. Yeah.

I mean they just, you know, Large Companies did not become large organically. They usually acquired other companies. They acquired systems, they had somebody else build a system and none of those projects ever finished.

Right. None of those integrations ever finished. There were bandaids and bubble gums sticking them together.

And that's the technical debt which needs to be fixed because it just, You've never seen that sanji any of your stocks at IBM or truist or anything like that. Yeah. And Forget all these known vulnerabilities that we shipped anywhere.

Yep. I, I would personally posit the technical debt is anything I'm not still working on. Right.

Um, anything that's in production is effectively technical debt. You always need to upgrade or you know, to your point, you're patching and, you know, getting rid of, of, of, of, you know, libraries that have, you know, got vulnerabilities in them or something like that. But also one of the areas where I see in terms of technical debt, AI could be really good and I've had some personal success with this, is porting to a different platform, a different language.

Um, you know, there's a whole bunch of old stuff written in a bunch, you know, we talked about mainframe in the earlier block. You know, the idea that we've got a bunch of stuff in cobalt and we wanna port that onto a distributed system or onto a cloud system language. It's about language.

LLMs are large language models. It's actually a perfect use case for LLM styled AI to do translation from an old language to a new language from an old platform to a new platform. So that's one example of where technical debt could actually be sold really well by ai.

These known knowns again, come up that it knows this language, that language, this platform, that platform porting it would be quite a good use of ai. I think in terms of result. You'd Probably do it on the new AirPods three.

Oh, the, the in translation. Yeah. That's all.

Now I'm interested to see whether that, I want them to go to Northern Scotland and see if they can translate Anything. Andy, I was gonna use it for when you and I talk, Just go to Brooklyn. That that's, so Then more of the issue is prioritizing which technical debt to tax.

Yes. Well that's, that's the issue. You can't do it all For the purpose of, you know, driving forward the new application environments that the organization says is critical for us to be competitive with.

So there's a, so that's kind of part of is that is a management issue about prioritizing those visas. So let's move, right? We go in some ways we're making this, we're making this a bigger problem.

And it will grow to be a bigger problem than it already is for two reasons. One is, and I'm curious your perspective on this Sanjeev use, if you use these tools, whether it's cursor or co-pilot, where it might be to develop code with, it's very anxious to generate, to modify a lot of code for you and generate more code. It's always offering to, would you like me to do this?

Would you like me to do this? It sort of like that pesky little brother that wants to, you know, wash your car. So you take him to the store and buy some can, would you like kinda kind, kind Can I literally it is, and you have to stop it from, I actually don't want you to change that.

'cause that's more than I want to change. So one is we're having, I think AI introduces not just a lot of code generated, but a lot of code change, which puts more pressure on better testing quality of code. The other is, um, it it, it isn't generating secure code.

Some people say it's less secure, more vulnerabilities, whatever. Even if it's the same as what we do today. If you believe the amount of code we generate, let's say it is 30% more code, A code is being generated by ai, then that means AI is generating 30% more vulnerabilities.

Or if it's only 10% productivity improvement, that means 10%. You know that we've now carved off to do other things, but we're still generating vulnerabilities and we need somebody, AI people to review that, to fix it. So you will get to a point where if a human has to be in the loop to fix what AI does, that's, that's, that's a point where now you can't pass that.

You can't go pass, go and get $200. 'cause there's not enough people entry level or not to fix what I, I think right now humans have to be in the loop. Do we do they do?

Yeah, a hundred percent we do. And you'll see it. You do today.

We do. We don't wanna get Tracy started on the, the problems of LLM. Oh, I was, no, I was just about to turn it up, Tracy.

That's why that I couldn't get to you. LLMs are technical debt themselves. Imagine the amount of data that has to be maintained in those large language models.

The sooner we stop building these, the better off we will be. I am gonna be on record for saying that we need domain specific small language models to get the job done. That's more, that's more accurate, thus hallucinations and less crap on your screen.

Tracy, do you have an AI dooms clock? And are we like three minutes to build that here? I listen, I love technology and I love ai, but I hate the fact that money has gotten so deep into this particular technology that we can't stop ourselves from writing something that's not gonna work in the long term.

And we keep running down these LLM models and we think they're the best solution, but they're not. We need domain experts. We need domain experts, and we need small language models that are domain experts.

That's what's gonna make us better at u at using ai. That's what's gonna improve the trust. I would love to have a DevOps small language model.

Uh, I mean imagine what we, what you could do to build that. You should have a small language model for medicine, a small language model for coding. The, the, the way we're implementing.

Um, I mean, I, I use, I use chat GPT when I'm writing, and I, I say I collaborate with it. Sometimes it makes me laugh. I'll crack up with the stuff that it, it comes up with.

It's like, boy, you really are, you really like to beef things up and make things. You're a marketing genius in that brain of yours. And I'll say that to it.

Yeah, but do you But let me, do you actually say it to it or you type it to it? I type it back. So, you know, I had a revelation.

I was in the car with Bonnie last week, last weekend in Napa. And she had, we went hiking, you know, two, two kids, well not kids, but two people from Brooklyn going hiking, you know, this is in Prospect Park, but we we're hiking in Napa and we had a question about, you know, proper things to do while hiking. And so I said, Bonnie, let me, let me chat GPT.

And she said, you're gonna sit here and start typing? I said, no, let me show you how it works. And I, you know, I said, chat, GPT.

And I asked it the question and it started, and we got into a conversation. She had never heard anyone converse with AI before. And someone said before, Kimberly, you are talking to people who are not techie people.

The first time they have a conversation with this thing. It, I, I, I forgot how I take it for granted now. Mm-hmm.

She was stunned. And even that she started laughing and, and the, and and, and the AI said, I'm glad you think that's funny. Well, you Know what hiking means to someone from Brooklyn, Alan.

It's, um, going out to the 10th row at the parking lot parking lot, right? No, this was seriously hiking. I think I need knee replacement.

But anyway, um, but you know, we forget how mind-numbingly kind of earth shattering that kind of, and maybe it's a cheap parlor trick. Maybe it's not, but it, it, it, it makes a huge impact on people. Huge impact.

We, We, we do need to get to the next topic first. Alright, We're off. Take a break here.

Let me ask that GPT what we should do. We'll be back in a second. We're we're gonna talk more about AI vibe coding.

Great. com is the leading resource for news analysis and education on challenges facing the cybersecurity industry. com covers all aspects of cybersecurity, including data security, DevSecOps, cloud security, application security, network security, security threats, and more.

com has the largest selection of security content featuring breaking news, blog posts, podcasts, and more. com to learn more. com.

Home of security bloggers network. Hey folks, we're back in continuing the theme. We're gonna talk a little bit about these vibe coding tools.

ServiceNow came out with a set that they're adding to their platform. And let's start this off with Sanjeev, but, uh, you know, I'll, I'll look at this as the glass is maybe less than half full, but, so we had low-code, no-code tools and end users use those to build applications that, well, they were ugly, they were insecure and they didn't scale. So now we're gonna give them these vibe coding tools that are easier for them to create those same applications and more of them.

And this is a good thing, right? Sanjeev, where are we with this stuff? Yeah, I, I I, I think the, the hype, uh, let's put it this way.

I used to have a boss back in the day used to say that your PowerPoint is way ahead of your code. I think that's where we are. The copy from the marketing people is way ahead of reality, right?

I mean, these wide coding tools are excellent for building a prototype, for building an MVP to for Wing and giving it to product people to say, Hey, what should this feature look like? And it reduces the cost of experimentation, right? I can build a hundred prototypes or a hundred versions of a new feature, uh, and then experiment with them, you know, uh, instead of just ab coding, I can do a to zz, uh, you know, testing and figure out what really works.

But I wouldn't, I, I don't think we are aware. I can white code my way to a production ready application and run banking transactions through it, forget banking transactions. I wouldn't trust it to run webcast.

So I think we are raised off from there, but it's getting there, right? I mean, I think people are real is what we were talking about earlier, right? Long term is going to be very important because long term, we, I think the end vision of this is to have a software factory, which you inputs going to electrons going to and tokens come out, which is production ready software.

We are not there yet. The differe really software fact, an AI factory that produces software and the current model is in a factory. If it produces a bad widget, you change the factory, you change the production system, you don't fix the widget.

Today what we do, we generate code. If it's bad code or if it has bugs in it or vulnerabilities in it, we fix the code. We need to go back and get to a state where we can go fix the models and the processes and the workflows which produce the code.

That's the whole concept of an AI factory, right? Uh, but we are not there yet, but it's progress. What, uh, you know, ServiceNow has brought out what a few other companies, and I met with a few startups who, which I cannot talk about yet.

They're still in semi stealth mode. Uh, what, what's coming out is impressive. I can't wait to talk about those things and can't wait to use them myself.

I got demos and I got to see a few very interesting things. I like what you're, I like how the way you put that, you don't fix the widget. You fix fix the problem that created the bad widget.

Right? Exactly. Which is very much why we need to get, yeah.

We need to generate secure code or have agents or whatever fixing that code before it ever comes to us, right? So it has those vulnerabilities taken out of it. Uh, it, it's interesting.

ServiceNow calls this vibe coding because of, you know, it's like DevOps, it's got a million definitions, right? Started as this idea of just leaning into AI to do everything for you to create software like as much as you possibly could use AI for all of it. And that's not really what we're talking about with ServiceNow.

They're talking about adding AI to an existing, uh, no code tool. And I've seen the demo of it, of the ServiceNow announcement, and it's, instead of you laying out the logic of this, great, let's have an agent do that. Let's give it a prompt of what we're gonna do.

So it's in, it's inserting, um, you know, LLMs and, and kind of beginning levels of agents. So is it vibe coding? I wouldn't call it vibe coding, but is it adding AI to a workflow tool?

A service management tool? Yeah, absolutely. And that's, it makes a lot of sense.

That's how to introduce it. You know, you're not gonna, to your point, Sanjeev, you're not gonna throw everything out you've done with ServiceNow today and say, let's do it all this way now. No, you're gonna add it and experiment with it and say, that's really good at that task and not so great at that.

I don't know if I trust that yet. Let's do more of this over here. You know, it, it'll help people get an introduction to weight, introduce AI into their processes, Talk about technical debt.

You just added a whole lot of technical debt when you're using something like vibe coding. Because not only are you paying for that, you're also having to deal with the maintenance of something you didn't write. You have no idea what it's doing.

And then if you wanna change it, are you gonna go back to your vibe coding platform or are you going to, uh, use humans to do that? And how long it will take for those developers to figure out what the heck it did. I, I mean, some of this code is so spaghetti looking, it's, it, it's not, it's not there yet.

Maybe ServiceNow is fixing that and I hope they can. So I was, as I was reading through that, again, not being a DevOps person, and Alan going back to your construction discussion, I kind of, and maybe Sanji versus Tracy, maybe you can visualize this for me, but what I saw this as is, is somebody that's drawing a design of a very attractive office building museum or whatever from an architect standpoint. And after you give that design, you've gotta put all the, the engineering functionality behind that in order to support this thing that is very different looking.

It's not the square box or whatever. It's, you know, it has, there's a lot of things to think about as you structure this building in the way that you wanna design it and whether or not it's gonna sustain and hold up cool heat appropriately, et cetera. So am I thinking it in the right way?

I, I, I, I think you are, Kimberly, that the, the weird analogy falls apart. And I remember when I was doing architecture work for IBMI remember walk, walking up to a real architect, somebody I knew who was, uh, architected buildings, cities, uh, cityscapes and all, and talking to him about, Hey, let's, why don't we write a blog post about the architecture analogies between real architecture, physical architecture, and, you know, software architecture. The reason we never did it was because the analogy falls apart because there are no laws of physics governing software.

When an architect is working and he is saying, oh, let's draw this, let's create this overhanging, you know, pool. He is like, well, wait a minute, do you know how much that'll cost? Because you know how much, how strong that glass and thick will need to be to have this pool that judge out over the balcony, right?

What the tensile strength of the steel needs to be. Those are all driven by physics with software, we don't have that, those laws, people try to write anything and we can technically write anything. We can write complex software which can drive, you know, uh, you know, quantum mechanic systems.

We can write software, which does very complex things. Heck, we, we, we put a man on the moon, you know, 55, 56 years ago, right? Doing, uh, using a computer, uh, with, with a similar code.

So we can do very complex things, but if they are not architected and thought through properly in a systems way, and these systems are very fluid and software unlike, you know, physical buildings. I, I think because we don't have those laws of software, we have principles and patterns and, you know, people take liberties and take shortcuts and introduce technical depth. I think that's why it is, the analogy falls apart after a while and I would love to still write that blog post one day, but, uh, my collaborator ran away when he said how complex and messy, um, writing software was.

Companies like, I'm going to that, to the world of construction. Man, that's much simpler. Well, I think that is a fabulous analogy, and I wanna point out that the laws of physics require safety.

Yep. Right? That is the whole point of it.

And we don't think about security and safety too much in software. Yeah. So please, we need to do that more often.

We need to think, pretend, let's pretend that software has physics that's, that's, that's governing its safety, right? How would we change? And that would've been interesting to put into a foundation, a model for co co you know, for, um, coding.

If you're gonna create a language model for that, it starts, could start with those disciplines and principles about systems thinking that you're talking about Sanjeev. And the second part of that is this is what are the principles, the physics principles for coding that should be in place. Kimberly, we'll be allowed to wait 56 years for all of us to agree upon what those principles are.

That's the problem. You know what I, listening, listening to you guys, I'm reminded of Ronald Reagan, right? And I wanted to say, there you go again.

There you go again. Just look, just look at what it does already. The fact that it's allowing you to prototype in a fraction of the time that it used to take you to prototype and, you know, it's doing, it's getting you 50%, 60% of the way there.

And, and here we are spoiled little rotten scoundrel saying, well, I want my 95%, I, I want it to work a hundred percent today. Think about that. You know, and, and, and I, and I get it, we all want it to go that last mile, but even if it just stopped where it is today, it's already a tremendous boon.

It's already a tremendous help in in how we approach these things, right? I I I think sometimes we lose sight of that. And it may not, it may never get to the point Sanjeev, where it's gonna incorporate the laws of physics, physics and it's gonna write Tracy to your satisfaction, the greatest enterprise code apps that we've ever seen.

It may not. Oh. What I do wanna make regarding, uh, what Kimberly, you said of putting these laws in the, in the LLM, I think the LLM is the wrong architecture to put laws in it because LLMs, as we stated earlier, and in fact there was a article come, came out of ThoughtWorks a couple of weeks ago, and the quote in it was, all LMS LLMs do is hallucinate.

We gotta to figure out which one, which hallucinations are useful, right? Or, Or maybe we all limit the hallucination red pill blue fill. Very true Uri, I'm actually not here.

This is, you know, and AI bot speaking to you. Uh, but I think, I think where the architecture we are seeing come from AgTech systems where you can now spawn off thousands of agents, which are challenging all the thousand paths your LLM could take and figuring out where which, and, and, you know, coming basically converging on a solution which is closer to a law of physics level of, of reality is a better solution than saying, let's put it in the LLM because the LLM is still a probability engine, which is going to ignore the law of physics when it doesn't want to pay attention to it because it believes it's gone down a path probabilistically, which ignores the flaws of physics and the probability engine is taking it there. So the agen way of saying, let's look at all the probability distribution of just not take the highest probability output, but look at the top 10 or top 20% and then converge upon one is a much better solution.

Architecturally, in my, in my opinion, cut based on current technology, we might come with appropriately our different architecture a few years from today. Uh, which where you can incorporate those laws of physics, uh, equivalent in in software. You know, I'd like Andy, I'd like to bring it back to, to what ServiceNow is doing.

'cause I think there's something really valuable to learn about this. This isn't a vibe coding tool. Here you go.

Create whatever you'd like with it. Yeah. The world is your canvas.

This is adding AI to a workflow management system of which major, major enterprises use every day to drive their business. And you're not gonna stick AI into it in a way that is gonna screw all that up. That that'll be the last time they use that product, right?

They're moving on to somebody that's not gonna tank their business. 'cause they got all experimental with ai. So look, look at what a, what ServiceNow announced.

Yes, they announced an agent, but they also a announced a way to control agents and know what agents are running and what they're doing. They also announced what the security model for AI and the machine identity that agents or the AI will have in that. So they're, they're thinking about this very purposely and and I talked to them before, you know, sometime before this, before this came out, and, and it's very conscious of we're not just taking what we have and throwing it away.

We've gotta figure out a way that AI fits into this because you're not just gonna build new things with it. We want people to use AI in the stuff they're doing now. So it's gotta be added in a way that's safe, it's secure.

Now, yes, you could probably put a prompt in there that's not gonna be safe or secure, but you've got some, some guardrails around to control it to your point. So your, to your point, Sanji, it's not the model that's gonna save your butt, it's, it's Guardrails and other things is gonna do that. We say agents, right?

And the, those guardrails can be provided by agents and there'll be a planner agent, a designer agent, a validator agent, a judge agent, you know, a quality agent. So I think, I think that's the way the, where the world is headed and that model is gonna be much is the future. So, you know, ServiceNow is ought on to do that.

And there are many other we Working on. I wanna give you the last word and then we gotta wrap, wrap up. Look, I'm just gonna come back to where we all started.

Well, a lot of us were started as DevOps. We're talking a lot about dev. Where's the ops?

Who's running this slot? Where's the operability, the scalability, the manageability, the supportability? Where are all the illities?

Do we have an LLM for the illities yet? Right? Um, so look, I come at DevOps from an ops perspective and what I see is a lot of crap getting thrown over the fence again.

Did we not solve this problem team? Oh my God, I'm having a deja vu. It's a hell of a way into the segment, guys.

I'm sorry, I gotta pull the plug though, 'cause we've got more to do, I'm sure. But people have other stuff to do at their work today. What an invigorating, stimulating panel discussion.

I love all of you for coming on here and talking about it. I hope you've enjoyed this as much as we have. Um, as usual, we've got a full boat of tech, strong TV following the gang today, so check that out.

If you're not watching this live, uh, on Monday, you, wherever you're watching it from, we've got a lot more tech strong content there for you as well. We'll be back tomorrow with another great gang panel. Some of us will be the same, some of us will change.

We protect the names to, or we change the names to protect the innocent. But, uh, Sanjivan. Andy, Tracy, Kimberly, Mitch, Mike, thank you.

Thank you for watching. I'm Alan Humma. We're out.