The Future of AI in Development and AWS re:Invent Insights | TSG Ep. 979

Hey everyone. You know what a kiro is. You're going to, you're watching Textron Gang.

Hey everyone, good morning to you. It's Alan Hummel for Textron Gang. If you couldn't tell, well, you could see the logo back here, but if you couldn't tell, we are in Las Vegas continuing our AWS reinvent coverage of, uh, reinvent 2025.

It's been a great show. Uh, you are watching this already on Thursday. So we've had two days of coverage here.

We hope you've been able to catch 'em. If you haven't, don't worry. They'll be available on Text Trunk tv or the Text Drunk tv, YouTube channel and OTT channel probably by early next week or so.

But our, our crackerjack video team is busy, busy editing these as we're shooting them. So good for that. Um, we've got a, a good panel of, uh, we've got the core panel here, core Three used to say the Core four, but that would be the Yankees Core three.

We've got Mitch Ashley on, on my far right, Mike Ard in the middle. And of course, I'm Alan Shimmel. Gentlemen, welcome to the gangs.

Great to have you on this morning. So, you know, continuing our, uh, I feel like Sherman marching through Atlanta, continuing our Agent AI campaign through Las Vegas. Shoot, That, Um, I thought we'd start off with a little kiro.

Yeah. So we are definitely on the wrong end of the fire hose here, but a lot of stuff coming down the pike from these folks. Yes.

But Kero is this AI coding tool that AWS created and is now updating to add something called Kero Powers, which are basically a set of specialized agents that have been trained to perform specific tasks, including autonomously with hooks and links to an MCP server to call things like Datadog and whatever else do we want to, uh, invoke as part of a, a workflow for the developer In a lot of ways, I think, you know, what we're seeing is a lot of shifting further left of various DevOps tasks into the developer, I guess what we might call the interloop. And, um, what's interesting about all these new AI agents is that it continues a conversation you and I were having yesterday about the fact that they only load when needed. And so they're reducing the total cost of using these AI agents because there's sort of one AI agent that keeps track of what the developer's doing and then automatically invokes these specialized agents when needed.

But Mitch, you're closer to this than I am, you know, what's your take on what's going on here with AI coding tools? 'cause I feel like they're getting a whole lot smarter, faster than we thought. It's, It's the move, if you want to use the term mag agentic, DevOp, DevOps.

Basically we w we wire these things together now today, whether it's A-C-I-C-B process or plugging in, scanning for vulnerabilities or deploying, whatever it might be, and observability, things like that. And you do see a lot of effort today around hooking in observability security earlier in the development process. And that's part of what they're doing through these agents.

Uh, it's also pre-integrated. I mean, the agent knows how to talk to Dynatrace and, and, uh, you mentioned, uh, Datadog. There's others.

I think they've been launching with about six or seven companies in Oh, Horseman's in there and a bunch of others. Yeah. So wait a second though.

I, I, I got a question here. Yes, gentleman in the far left of the room. So what you're trying to tell me is that we're teaching agen AI to shift left.

Finally, somebody's shifting left. Okay. I thought shift left was passe.

I thought we went through our shift left period. Well, now it does bring up an interesting question because just because you can do something doesn't mean people will. So it may wind up just being too overwhelming for somebody to cognitively keep track of.

'cause you're, all these things now are happening in parallel. So if I'm writing code and suddenly, you know, an AI agent's popping up going, Hey, you know, you want me to run this, or we need to do this, it's kind of like, well, I, I think when you're writing code, it's kind like, even with the help of ai, it's still as much art as it is science. So how much of this stuff winds up just interrupting the flow?

Very, very, you could get pestered just like popups and notifications and et cetera, I think is your point. Absolutely. Well, it's gotta be managed through some kind of a control plane, whether that's inside the IDE, more likely it's probably part of something like, uh, like Agent Core or something's got to orchestrate and control what the agents are doing.

I, I doubt that's gonna start that way. I think it's gonna start by the developers running, doing their thing, leveraging these agents. They'll build 'em into their workflows.

If they're bothersome, they'll get, you'll kick 'em out or make 'em more useful. Mm-hmm. Well, let me give you Shimmy's take on this one.

So, You know, when you talk to my friends of Plat in platform engineering, they tell you one of the reasons that platform engineering Rose is because in DevOps, the whole shift left thing just really didn't work, and it didn't work as well as intended because it just put too much on the developer's plate. Hmm. So, by putting too much on the developer's plate, the developer did what developers do, which said, the heck with this, I'm not doing it right.

I, I want to code. And, and so we, we've seen this whole rise of platform engineering to put a platform together to allow the developer to do what he wants to do within the garage, guardrails and policies that we want. So along comes this idea of saying, Hey, instead of letting the developer do it, we're gonna have an autonomous agent, AI do these tasks.

And so the developer won't have to, he could just code, or he or she could just code like they want to. And now we're saying, wait a second, this thing is gonna be API A, it's just gonna pop up and do all of that. You don't think the developer's gonna say the same thing and say this, there's A-P-I-A-I, I'm not gonna pay attention to it.

I think there's gonna be some balance in that equation. But I also think, you know, it goes back to what we were talking about a couple of days ago with AWS. They clearly think that they are gonna be the platform engineering team for the developers.

And so they're more than just an infrastructure provider in their minds. They're saying, we're gonna provide various types of DevOps AI agents, and they are gonna be the manager of the DevOps workflow and the platform engineering team for all these developers out there. And that's how they're starting to think of DevOps as more of a managed service rather than something that, um, you know, a bunch of software engineers are trying to support on behalf of a hundred developers.

They're kind of inserting themselves into that process. Do you, is that how you see it? I, I think so.

I definitely, I do see it that way. And I think you, you skip the step, which is the developers are the people who experiment with things, right? Maybe engineers do, but developers definitely do.

Not all of them, but most are, they'll get on, they'll try these agents, they'll see what's if, what's useful, but they'll get to a point where they say, okay, now the hassle factor, I don't wanna deal with this anymore. Can someone take this? And now that we have platform engineering, I don't think it has to be, oh, the CIS admin who, or the who, who runs the build who can take this.

It's the platform engineering team can work with the development team and say, all right, you guys, try this stuff out. See what's helpful. We will make sure it's wired the right way into our tool set, and we'll keep the things that are helpful to us.

But I think your point also is correct of AWS is to figure out how to compete with GitHub. You know, how do we provide not just the tool to write code, how do we create an environment for creating mm-hmm. Software and agents integrating all these things.

We've got all the rest of the infrastructure from, you know, hardware all the way up to databases and security, and that let's start to build an environment. You can really kind of pre integrate these things together. My, My problem with it is you can't please all the people all the time.

We all can't be right here. One of us is wrong. Well, clearly it's, thanks for stepping up, Alan.

No, I, and I'm not idiot fan. I know. Nothing.

Love It. Well, he's such a giving guy. He's really just sticking.

But here's, Here's the thing. We talked about platforms and platform engineering. You know, when platform engineering first came out, and we talked about platforms, if you managed your Kubernetes, you were a good platform engineer, right?

You had a nice Kubernetes workflow, you had your CICD pipeline cooking, and wow, you're a genuine platform engineer. And then a funny thing happened, platform engineering shifted focus to IDPs, right? Internal developer platforms backstage and, and it's ilk.

Um, are you saying that these hero agents, these agent AI agents that, uh, Amazon's coming are gonna form a new IDP, they're gonna run in the existing IDP backstage or what have you, because is that really where we want them there? Or are they after the fact? 'cause if we want them in the IDP and they're generating the code and they're doing all of these things, we are fundamentally changing the role of the developer.

The developer becomes less the person who develops code and more the person who's gonna manage the agents through develop code. Yes. That is, That is, I'm asking my developers out there, raise your hand if that's what you want your job to be.

So, so hold on a minute before we, we jump fully off the bridge. Let's stop about halfway down. So here, here's Where we're, it's not American, and I won't stand For here.

Let's, let's where we're gonna march if we're, let's, let's protest. I mean, look at what's happened so far. Um, we, we've gone from the, you know, every tech executive in the world saying we don't need developers anymore.

'cause I can build a snake game on my computer, right? Mm-hmm. We don't need people to write software.

What are, where are all the tools that are coming out? They're tools for developers to use AI to build software. So are you gonna write less code?

Hey, I'm happy not to write all the code. If it's for good code, it's, I'm happy not to write every piece of code if I know it's gonna work and not cause me more work. Right?

As AI gets better at doing these tasks, generating some of the code, doing code reviews, now it has to be done in a way, it can't be the intrusion prevention of work system, right? Like we saw with I yeses and I IEPs, but developers do these things anyway already. They already orchestrate work.

They already work on design. They already craft software. They write code, but that's actually a small part of what they do.

The 80, the 80% of what they were doing, that's not writing code, is all this stuff. Yes. So we're trying to automate that while they're sitting there.

But your point is well taken, right? Because on the other side of it, developers are gonna pick or have a lot of influence over which of these AI coding tools that they're gonna use. And some of these decisions may be made by, you know, a CIO who's gonna sit there and say, this is our standard folks and this is what you're gonna use.

And that definitely happens, right? And then there's gonna be folks who are gonna, where the developer has more influence over which tool they're gonna want to use, and they may like open AI or Microsoft or whatever it is, or philanthropics coding tools better for some reason. And then there'll probably be some middle ground between turning on every platform engineering AI agent there is, and me wanting to be comfortable automating a couple of functions, like observability of my code as I'm writing it, versus still relying on a platform engineering team.

So I don't think that this is, you know, an an either or, it's a spectrum of things that will vary widely depending on the nature of the organization using it. So it's a spectrum. It is a spectrum.

Well, We know developers anything to do with vaccines. No. Um, but, but, but, But, but as we have, but as we have noted on this show many times, we are all on the Spectrum.

Yes, we are. So I appreciate the conversation. Appreciation.

Let me, let me bring this home to a business. Mm-hmm. Conclusion.

We spoke yesterday about to a certain degree, and I, I spoke about it last night with Daniel Newman. We record, actually, we, we think we recorded that. Yes.

This reinvent has been a big pivot for AWS because the perception was they were trailing Google and Microsoft and ai. Mm-hmm. And they came out guns blazing with ai, AI all the time.

All ai all the time. You said something before about Microsoft, you haven't mentioned Google. Google had a great week last week in the stock market, right?

Yeah. Because I think the market recognizes that Google may have the best vertically and horizontally integrated stack among the hyperscalers. They have it all.

Microsoft has some great tools though too. They have GitHub. They have the developers in their back pocket, 150 million of their friends, right?

And they always had a great developer channel before GitHub even. Exactly. But they've got GitHub, they've got it.

Azure, they have the open AI relationship as well as their own copilot AI things that they also own the enterprise. And they own the enterprise formidable, formidable competition here. Mm-hmm.

Now, AWS look, developers built the AWS that we see here. There's no doubt they were developer developers love to whip out their credit card and, and open up instances on AWS, but they didn't have a GitHub analog. They don't have Google's stack.

Yep. Agreed. And so what AWS is doing here is this is their claim to a stack, or at least the beginning of an AWS stack that'll run on train and processors as well as on Nvidia.

And, and they'll have an analog to a GitHub or some sort of developer IDP and they love open source. They can take backstage. Yeah.

And they, and they'll go one step further. They'll say that rather than just focusing on Gemini or open ai, they're gonna say, you, you Can plug anything in Bedrock. You can plug anything you want in here.

And we will automatically optimize the right AI model for the job based on the cost and the level of accuracy that you want. And it will just be something I tune rather than something that I decide that I'm gonna be like all in on open ai or I'm gonna be all in on Gemini three. And they're not completely out of the model game.

They have the Novo models. I'm not saying that's where Yeah. They're right.

They have no where The others are, but hey, you know, they're at least attempting to make a play in that. We'll see. But they're also giving you the ability through Novo Forge, forge to make your own models to a certain extent.

But we so crazy like a Fox. Yeah. We talked about that the other day though.

It's gonna be, you know, Nova's kinda like, you know, the generic version of the prescription drug, right? It's does it enough and it's cheaper. And it's, It's the A WSA way.

It's the AWS way. 80% for 20%. Right.

And it is the way, And we're gonna, and we're gonna, this Is the way, this is the way, and We're gonna talk a little bit more about NOVA in our next segment. All Right. Let's take a break on text on gang.

We'll be right back. Discover Techron 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. So we are gonna shift the gear here a little bit.

'cause uh, AWS is talking about, um, making it simpler to, uh, retrain AI models and also customize them. So the retraining is a function of bedrock where they're kind of streamlining that whole process so that I don't have to a PhD to go fix the AI model, and I can fix the AI model on two parameters. One is I can make it more accurate, or I can also make it less costly.

And so they're trying to make the whole play, uh, simpler for mere mortals, or I wouldn't call it everybody, but we're getting past the point where I need a data scientist to do all this stuff. And what was interesting to me about this particular capability is that, um, it shows up first on Nova models, oddly enough, and then it's gonna show up on all this other stuff later. Mm-hmm.

So they're starting to send a signal that says that maybe NOVA is not a, just an also ran, it's gonna be like their preferred platform. And then everything else we'll see later on other models at some point. And then they're also talking about, uh, within SageMaker, which is their managed servers for building AI apps, um, that they're gonna add some ability to have a serverless capability for customizing the models.

So you get a little more granular control over today, what is pretty much, you know, a take it or leave it kind of approach to build out an AI model. And I think that's a fair assessment of what they're talking about and thinking about doing Good coverage of it. Um, From your perspective, as you look at all that, are we kind of getting closer to the democratization of the building of AI models so that I don't need to have a mahome bunch of data scientists to do every little thing?

'cause that seems to be the gating factor in my mind. I can't find enough of those people and they're expensive, and, you know, they work for me for six months and somebody steals 'em. Mm-hmm.

Well, there's a duality here, which is bring your ai, bring your models. We will, we'll use whatever you, and you, of course, you can use our, our nova, our AWS Nova models. On the other hand, what, what, what they're doing with Forge is bring your data and train your, our, our models, your instance of it, if you will, on your data, rather than doing everything through A MCP access.

Right. Well, what does that create? I'm not gonna move Nova over to somebody else's service.

I'm gonna run that in. Aw. It's a stickiness right there.

Mm-hmm. I think that that's the brilliant move behind not just giving there is Yes. Making it more alluring, the bar to be able to train models yourself and not have to be a genius.

I love a duality, Mitch. You Know, I'm all about the duality. So I looked that word up.

Speak back to a, it's my word of the day. It was in my, on my tongue. No, No.

We, I learned about duality in St. John's, but in a theology class. Oh, you did?

Yes. Okay. Yeah.

Light and dark and, and all of this eroticism. I wasn't doing the evil Good thing, but, you know. No, but there's a duality.

But, but here's, here's the, I guess for me, the crux of it with this forge tool, I'm not looking to create a model that's gonna take on anthropic or open or any of this. To me, forge, you know, when we did our, um, hackathon almost two years ago mm-hmm. On operationalizing ai, we had all those really smart people, a lot of brain power in that room.

It was clear to me, listening to them that large frontier LLMs were not the answer for everything. They're good for consumers to play around with and, and do a lot of things. And maybe for marketing and analyst people, you know, but not for everything.

And we were going to need small language models, right. SLMs, we were going to, you know, and back then they were, you know, Patrick and John and those folks were doing vector databases to introduce Just the beginning of that for MCP and what became MCP, What became MCP rag stuff like this. Mm-hmm.

So to me, is forge a better mousetrap to build my SLM? And then do I maybe plug that in a bedrock and I still have an open AI or a frontier model behind it? I think we're gonna go back to the word spectrum.

So there's four tiers to this thing, right? I can build a foundational model, I'll, uh, open AI and all those things, then I can use, but you Really can't. I, well, I know, but I'm just saying as the spectrum is that's the top end of the world.

Then you use Forge to build your own foundational model. You're not really building a SLM with that you're building, and it's for global 2000 companies that are gonna build a fairly large model. Then there's all these tools underneath in Bedrock that I can use to, um, I don't have to be as big an expert, and I'm gonna use those to build SLMs, and I'm gonna use those to kind make that more accessible to a broader number of people.

And then underneath that is SageMaker, which is kind of this total managed service where I just basically show up and describe what it is that I want done, and SageMaker will automatically generate it for me. And they're all kinda like, on a different set of classes of things and customers. It's Like Skittles in the rainbow, I guess.

Yeah. Taste the Rain. But yeah, it, it gets a little complicated.

But, you know, from their point of view, there are different classes of companies using different levels of ai. So there, I gotta disagree with you. I think companies will use different classes.

Not, it's not this class goes to that company. Yeah. One company, in one instance may want to use the Forge or Nova and, but at another instant, or in another instance, or in another team, or at another juncture that we don't need that right now.

I just need the SageMaker thing. This is true. This, Right.

And I, I think that's what we gotta understand is this is not, as AI continues to develop, this is not a one size fits all. And it's not, I just need a one solution. I need the whole spectrum.

Well, the LJ would use is if you're not a, a frontier model company, which AWS is not a, not not at the, the level of course of a open AI or philanthropic, what do you do? You sell people, you're an arms dealer, you're a picks and shovel dealer. You're selling the tools, right?

And I think that's to your point about the spectrum of depending on what you want to do, how, how invested you want to be in creating your own models versus using models versus customizing access to 'em. That's the strategy here. Yeah.

No, I got a question regarding chip usage in this spectrum, though, as we go to that front on that spectrum, from left to right, right from ultraviolet to infrared, where do I need my GPUs? And where do these traum chips fit in? Mm-hmm.

It, it, it remains to be seen because, you know, right now I think most people are using Traum as an alternative for inference engines. I can though, you know, AWS is talking about Traum four here, which, you know, right now doesn't exist. It's just a piece of paper with some specs on it.

But, uh, you are right. But let's be fair, they came out with Traum two. They said, we're coming out with Trane three, they deliver.

And yeah, So the point is, I think a significant portion of the AI training will move to AI accelerators. And for two simple facts. One is GPUs are more expensive, and two is GPUs have a lot of code in there for graphics that have squat to do with ai.

And they're not the most efficient way to do that. And they only happily, accidentally wound up being the platform for AI because they existed. And somebody went, well, that's the best thing for training in parallel.

So here we go. But if we have processors that are designed for the tasks specifically, and it's not just training 'em, there's a small army of these things starting to show up more and more of those training and inference sets are gonna wind up on things other than GPUs. And they're gonna be less costly to run.

They're gonna consume less energy, and it's gonna have an interesting impact on stock valuations. I think it's all about economics and availability. Mm-hmm.

I mean, that's the, that's the game right now, right? Yeah. If you can provide an, an alternative, maybe it's not as fast, maybe it's not as quite as cutting edge as some things about Nvidia.

Yeah. On the other hand, the other thing they did with, with the, the training environment is they opened source, the SDK, they came out with an SD SDK for that. Uh, Well, that, to, To fight against basically the, you know, the non-open source version of what, uh, NVIDIA does with their SD K.

So there's another track. I thought NVIDIA's, SDK is open. I Don't think it's open source.

I, I'll double check, but I don't believe it's open. I I think it's not within a consortium, but I believe like large aspects of, It's not a Linux Foundation thing. Right?

But it's, it's open. It could, It could be open. It may, maybe it is Because I've see Cuda libraries showing up in other products.

KU is definitely open. Yeah. So there's pieces of that that are probably open.

But, um, it's just a question of who's open, who's writing all the, My, my feeling though is this, if you're AWS every time you gotta write those big checks over to Jensen, you wince a little bit, right? Mm-hmm. That's, You know, Jensen was here doing the thing, man, but They're partners.

Yeah, they're partners. You know, that makes no difference, right? I mean, you know, the, what's his name?

Who wrote the, the Yard of War? Sonya. Not Sonya.

Sun. Sun Sue, Sun Sue. You know, you, your partner could be your competitor too, right?

It's true. So they're partners, but every time they write 'em a check, they don't think, they don't think to themselves, well, I'd like that money staying right here. Yeah.

Mm-hmm. Of course. And so, a a, it, it would, it would be ludicrous and naive, who's naive now to think about, to think that they're not gonna put their thumb on the scale and say, hell no.

This is a job for a traum. Mm-hmm. Not a, not, not a, a Blackwell or something like that.

And I, and, and quite frankly, look, Google's gonna do the same thing. Now, interestingly enough, the the thing they're swimming against is that there's a certain amount of, um, good old fashioned marketing mojo behind Nvidia these days. And you go talk to data science teams, and they're like, I absolutely have to have the latest, greatest GPUs.

And frankly, they're not really thinking about like, what are the cost implications of that? They just want the latest greatest, they just want as well, Greatest. They're like, my Career path is being trained on Cisco.

Right? So, but I'm in this case, being trained on Nvidia. Well, everyone want, you know, it's fomo.

It's a little bit of a Boca thing. You all want the, you know, that, that kind of thing. Again, I, I had this conversation last night with Daniel Newman about this, which is, at least for the foreseeable future, Nvidia will probably stay a generation or maybe two in performance out ahead of the trains and the, you know, the Google chips and the Broadcoms and what have you.

But as much as the developer wants to be on that latest and greatest, and the data scientist wants to be on the latest and greatest, if the price gap mm-hmm. Is mm-hmm. Is wide enough, I'm sorry.

Right. And If I can run it on my desktop, I'm even Happier and happier. Exactly.

Right. 'cause ultimately, A CFO and A CEO will sit there and look at the IT guy and go, excuse me, you wanna do what for how much? Yeah.

I, I that, and that's where I think I, I think to a certain extent, AWS may be betting on that. And they're not alone. I think Google is too.

I think you're gonna see Microsoft do it. I, you know, I think meta was, was going in that direction though. Things seem to have gotten a little rocky over there.

I I hear the phrase token sticker shock a lot. Yeah. Mm.

Mm-hmm. Mm-hmm. And that I, and you know, again, though, what, what's interesting is this is all part of this continuing evolution that is, we're going through with this and we'll, and we'll see how it shakes out, but we're gonna shake out with a break right here, come back and talk about our third segment from A-W-A-W-S reinvent on TechOne 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 boardroom.

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 cloak, digital executive protection, defending the new attack surface your personal life.

All right, folks, we're back with our third and final segment of the day from AWS reinvent. And we're talking about Strands, which is this software development kit that AWS open sourced earlier this year. And now they're adding support for TypeScript to this.

And it's basically an SDK for building AI agents. And it makes that whole process a lot simpler. Now, the other thing that they were talking about was related to AI agents too, was they now have a tool that will help automate the building of the UI for the AI agent, which is, you know, the AI agent still has to have something that looks like a user interface on it.

And that's not something that everybody's very good at. So they're providing some tools to automate that process as well. They were talking, and we discussed a couple of days ago, you know, we're talking about billions and billions of Carl Sagan AI agents that were supposed to be building and deploying right now.

That feels like a fairly manual process, maybe, or so is that whole thing gonna become automated? And what does that look like? Well, you said the word TypeScript, which is code, right?

Mm-hmm. So we're coding agents today, end users aren't, I mean, in our Google workspace, right? We're using Flow that we're not writing code to do that.

We're using a UI to do it. But to do what developers need to do, they write code. And it's just that sophisticated.

Um, the, the, the strands, SDK, what it does is basically build around models. So it's all about the prompt and getting the model that you're accessing and making really simple to create an agent to go do that. It's the things that, it's the, it's the, uh, framework that the developers have to go build if they're gonna write it from scratch, right?

So you don't have to do that. If you contrast to like what Microsoft announced, they announced their agent framework, um, about two months ago or so. Mm-hmm.

That's about a orchestration. So companies are tackling different parts of this, right? The, the strands doesn't do orchestration.

That's part of Bedrock and, and what, what we're looking for that to do. So it's, people are working at the pieces of it. But I think think if you back up for a minute and say, oh, there there's a couple of new SDKs that, that, uh, that AWS came out.

Sounds a lot like a development company to me. Company telling developers back to our first conversation. In other words, one more thing saying we're trying to recapture the developer, uh, a persona to come be beyond our platform, use our tools, And, and they made it open source out of the gate.

So they definitely have, so there's no lock this thing as broadly available as possible. Well, that Addresses the lock-in. You can't claim lock-in then.

And, and now they're also talking about using that tool to build AI agents that will be deployed at the edge, including for robotics applications, which, you know, has been the hot topic of AI as of Physical AI you're Talking about. That's Right. That's the one that's going to, uh, eliminate poverty jobs Among other things.

Yeah. Yeah. And this all Into Star Trek And, and might someday be the, the final killer app for Wasm that we've been waiting for now for about four or five years.

Akamai will be happy, right? Well, that's, that's what's gonna do. Did, did you know, uh, Akamai bought, uh, yeah, yeah, yeah.

It was interesting. They made a bet on Wasm. Yeah.

So let me ask you a question and, and forgive me. You know, I'm, I, I am not quite as up on this, but are we saying that AI agents will create AI agents? Yes, Absolutely.

AI is gonna write its own code at some point. And It, at some point in the short term, it's still gonna, most of 'em are gonna be built by good old fashioned human developers. But eventually an AI agent will be able.

So would that be some sort of self-replication? Uh, It'd be polymorphic code, basically. Yeah.

Polymorphic self-replicating code, code Writing code. Could be, could be changing its own code, could be writing other agents. Yeah.

In real time. I mean, Odd, like, to me it's, Sounds, sounds like a worm to me, but what do I know or About this definitely could do That. I guess it all depends on your outlook.

Me, I'm looking for God, they're looking for bacteria and viruses and Worm. It's a spectrum. It's a Spectrum.

There's the spectrum for you right there. But, but you know, there is theoretically, you know, something could go wrong where something starts spawning AI agents out of, you know, hundreds of these things and then you gotta go clean them up somewhere. But I imagine they'll probably be an AI agent for cleanup too.

It'd be the mop-up agent, or, you know, Well, don't they have those in Zion? Oh, in, uh, in The, the matrix. Matrix.

Matrix. Yeah. What do they call them?

I can't remember. Like the octopus looking things that The Sentry agents. Yeah.

And, and the agents may even have a set time life. Right. And they're just gonna self-destruct after six months and be ought to replace by another AI agent.

I gotta tell you, I'm just so damn proud of myself that I was able to get a matrix one in here. Mm-hmm. Thank you Agent Smith.

Well, that's 'cause you think we already in, in matrix, right? Well, Well, I, I won't lie to you. There are days I do red pill, blue pill, guys.

It, you know what? Let, let's tie a bow on this. Um, it's been a really eventful reinvent, right?

Look, usually, you know, you come here to reinvent and AWS is just regurgitating out one release and announcement after the next. And it, and it probably takes you three months to digest all of it and kind of make heads or tails. This isn't that different.

They put out a ton of stuff, but it is so focused and it make it so obvious right? That this is where their, their attention is right now. Yeah.

And they probably have a thousand other things going on, but what they chose to highlight here shows a narrative. And there's a thought process as to how all this is playing out. I'm not sure that, you know, other companies aren't doing the same thing.

And in fact, I suspect that they all are. But it's interesting to me that, um, AWS is getting better at articulating the mission and how all the piece parts come together and doing it, frankly, in a way that is better than the other companies that I've seen so far. There.

There are some, like trying to weave it together. What does this mean? How does this work together?

'cause it isn't all being introduced at once, right? It's put a lot of parts. So sometimes it's obvious like, okay, this is clearly what they're doing.

Other times it's, Hmm. Not quite sure whether, whether it's AWS, Microsoft, Google, whoever it is. Agreed.

Hey, we've got one more, uh, Textron gang that we're gonna record out here for tomorrow. Mm-hmm. So very excited by that.

We'll continue the conversation, but for now, I think we're going to wrap it up. We've got Mitch, I know you've got a ton of briefings to do, dude. Yeah.

Mm-hmm. All day we got stuff happening. Uh, so until later or until tomorrow, that's gonna wrap up our Textron gang coverage.

But hey, we've got a full day of live coverage I'll be doing here at our, uh, Textron Studios at AWS Reinvent Hyatt top the tower here at the Winn. And, um, stay tuned for that. But until then, on behalf of Mitchell and Mike, and myself, have a great day.

Everyone. Enjoy our coverage. We're out.