AI News for Mid-September 2026

0:07Aaron: Good morning, good evening, wherever you are, and welcome back to the Enterprise AI Show. This is your host Aaron, and we've received some feedback and actually seen in the listener stats that you enjoy the monthly news shows, But a month is often too long to really talk about some of the AI topics since the AI news moves so fast. So we're trying something new. We're moving to two AI news shows, one every two weeks. So let us know what you think. Through all the usual channels, you can send your feedback. And with that, we're going to jump right into the news right after this quick break.

0:49Today's show is sponsored by Nasuni. There's a growing gap in AI right now between what's possible in theory and what successfully works at scale inside an enterprise. The difference comes down to unstructured file data. Many AI initiatives struggle because the file data they depend on is scattered, unstructured, and disconnected from where and how work actually happens. Nasuni changes that. It brings your unstructured file data into a single, secure foundation so AI, both generative and agentic, can access it with the context, governance, and performance it needs in production. Bring AI to where your unstructured data lives.

1:23See what it takes to activate your data for AI and request a demo at nasuni.com slash AI.

1:32Aaron: What does the outside world already know about your company? Cyber criminals could be seeing leaked credentials, compromised session cookies, exposed infrastructure, and even impersonations of your brand and executives. Nord Layer Intelligence by Nordstellar gives security teams visibility into those threats across the deep and dark web. Data breaches, attack surfaces, and brand impersonation all in one platform. Find out what attackers know before they can use it. Visit nordlayer.com slash intelligence slash enterprise AI and use code ENTERPRISE10 for 10 % off your Nordstellar plan.

2:20Aaron: and we're back and i guess you could call this a new spin to an old show i don't i don't know we'll have to make a name for this so we call it a remix that's what the kids call it it's a remix yes here we go remix for for news so here's what we're going to do everyone out there for a while there we did news weekly in the introduction then we moved it to once a month and AI news once a month, it might as well be on see your show. And so we're moving it up. We know we just did a new show two weeks ago and this is what we're going to start doing going forward. I feel like AI news biweekly is a horrible name though.

3:03Aaron: So just like our other naming series, we have to come up with the name for this series eventually. I like your remix, the news remix. Maybe that's what we're going to end up going with Brendan, But it's just Brendan and I for the news. And we're going to hit the major news for the last two weeks or so. Brendan, first of all, though, is there any sports you want to talk about? I feel like the news show always starts with sports unofficially. Well, it should. I mean, you know, there's a lot of new doom and gloom out there. But listen, if you just want to watch an old-fashioned rivalry and just with a lot of emotion, and if you're not even interested in American football, you should watch LSU and Ole Miss play this week.

3:40And for those most of you don't have no idea what's going on, just know that the coach for LSU left Ole Miss about eight months ago in a very dramatic fashion. So it's like it's almost like, you know how they talk about sports is great when like two teams are really similar and like all the people are very similar, but they then play and they hate each other more. So there you go. So you can watch it. And there's going to be a lot of talk this week about him returning to play a game there. So if you're just looking for a distraction and just let's call it sports family drama, find your TV and watch Ole Miss play LSU.

4:16And there'll be a lot of dramatics. I don't know exactly what's going to happen, but there'll be a lot of drama that will go on.

4:22Aaron: I agree. I agree. I love it. All right, man. So let's jump in. We actually have, we haven't done this much recently. We have a cloud news story. And it's all about Oracle, though. I feel like Oracle, they may want to be an AI player, but we're going to put them in the cloud news bucket for now. So multiple things go in there. And part of it is related to AI, but like I said, we're still going to put them in the cloud bucket. So, Brendan, there's a couple things going on here. So first of all, A, cloud revenue up dramatically, 121%. 300 ,000 GPUs to AI customers have a backlog, a really, really good backlog.

5:09I mean, I've just seen the number here,$664 billion. At least this is what this article says. I mean, I guess we need to put that in perspective, but it seems pretty tremendous to just basically, if you will, think. And I guess this is maybe we have to dig into it. It's like maybe that backlog is dependent upon companies that don't really have all of that money right now.

5:30Aaron: So maybe that's the paper tiger mission. Well, that's where I was about to go. That's where I was about to go with all of this is like, what does backlog mean? I know we talked about it recently on one of the other shows. It's like ARR and the all of a sudden shifting definition of revenue. Are we just counting when somebody promises you someday to pay you money, we're calling that money now? and that's what we're reporting on? I don't know. I don't quite understand that. But maybe we should talk about this in a smaller scale because we deal with much smaller sums of money in our various lives and professional careers.

6:07And it's like, sometimes people, you'll get to closing a deal, right? And you'll say, listen, I want the money before I do the work, right? And they'll come back and be like, well, we only do net 30, right? Or whatever. And maybe sometimes they'll be like, we do net 60. So you're going to basically do the work, wait 60 days. And it's like, there is, I guess, in all business, right? I guess all of, I mean, in some ways, all of this, even a small thing, even if you're just having like a small project done in your house, right? Like when you pay the contractor, when the contractor does the work, like there is a moment where like, we have to agree to trust each other, right?

6:42Somebody has to go first. That's like, and you can make it as small as you want. Like, well, if you show up tomorrow at this time, I'll pay you. Or like, if you show up and work a day, I'll pay you for that. You know, it's like you, and it's something along those lines. So to be fair about it, right? Like all Oracle, all anyone can do is to say, you know, if I'm the client, like this is what I will need in the future. Right. And I'm willing to pay this. And me as, and you as the provider, Aaron, he's like, all you can say is I will provide this to you. And, you know, and I guess outside parties can look at us and be like, it's almost like when two people are getting married.

7:17and like everyone's like, I don't know if they're going to make it. But like, what can you do? Like these people are making a commitment to each other and it's like, who are we to judge? So part of me is like, I guess that's all you can really do. And so I guess we can always be like, well, it's not going to happen, but what would you have them do, right? Would you have the AI companies be like, well, I don't know. I'll just wait and see. I'll wait until I have the money before I buy the data center or the GPUs or vice versa. Like if you're a hyperscaler, like I'll wait until they give me the money to build the thing because nothing would happen then.

7:48So I don't know. Maybe that's a counter take to like it's all going to collapse. It's all fake money. But there is like, I don't know, to some degree, this is how business is done, right? And there's no way getting around it. What are your thoughts?

7:59Aaron: Well, my big thought here is like how many of these customers are on a bunch of people's backlogs? This is where maybe the marriage analogy falls apart is you're going to Oracle and going, yeah, no, I love you. And then you're going to Azure and going, yep, I love you too. And you're going to all of them. And then whoever comes up first, that's who you love the most. True. But I think the one part of it, it's not even a but. It's just like the one part of it, though, is like we do have pretty good transparency into what's going on here. And maybe this will come up a little bit later when we talk about some other topics.

8:34But, you know, we have financial reports. Investors can ask questions. We see the backlog as it's described. you know investors can ask about like who is the backlog and you know that's the whole idea of equity investments right is that like you're not guaranteed a return right you're basically putting your money in and the customer company is usually making some type of estimated profit based on some type of you know strategy or story if you will and you know i don't know i don't know if we can complain about it so it's one of these things where it's like if you don't like it you know you cannot invest and if you really don't like it you're you can short it i mean there are ways to make money on the downside.

9:12So in this part of the world, it does kind of feel like the system is working as designed. You know, it's like there's like every, you know, everyone can have a position and everyone can kind of, if you will, put skin in the game one way or the other. So I kind of look at it like, I don't know what's going to happen. I'm certainly not giving any investment advice. You know, if you would, you know, it's like, look at my, I'm just sort of an index fund. So I guess I'm riding, riding what the market is doing. But beyond that, it's like, I get it. But now I guess maybe the question around Oracle though is, is always the, the fact that it always comes with this one, two, right?

9:45It's sort of like cloud is killing it. Backlog is great. And then they announced, I guess, another layoff. I don't know exactly what the percentage is, but I think it's in like maybe double digits. And I don't know, that part is just sad, I guess. It's just more like, you know, someone that works in tech and has, you know, seen all of this go on for years. It's just sort of like every time the company is doing better, it seems like it comes back with another layoff. And at some point, it just, if you're an employee there or, you know, like how many rounds can you sustain where the morale just is like, it just becomes so low that it's just like, kind of, what is the point?

10:19I don't know. What are your thoughts about that side of it?

10:21Aaron: And so here's the thing, actually a couple of different vectors on this one. So first of all, when it comes to kind of the layoffs and everything going on in the industry, at least at a time when we pulled all these notes together, I never saw an exact amount, But, I mean, they've done some pretty significant layoffs here recently, and it seems like this one was pretty significant, too. But the other one was, I mean, we're getting really, I don't know, mature, professional, brutal, whatever term you want to use about it. Because basically you got an email at 6 a.m. Like there wasn't even the call with HR that for 15 minutes on a Monday morning or a Friday afternoon, like today's your last day.

11:05Aaron: Like it was just and you just check your email and it's like, hey, you will be locked out or you're already locked out. And so, like, I don't know, like that for me is the more I guess the personal side of this is also I'm like, oh, God, that makes me hurt of like we're down to just you just get an email. but I don't know maybe I'll go the other way on this it's like you like it you want maybe I'm just like too desensitized at this point like I just feel like at this point it's like this is just where we're at right it's like I almost just prefer the 6 a.m email to like the random not random the uh the sudden uh I need to speak with you in the conference room or like can you jump on a zoom right now it's really important and you'll be like and you're like okay and then you get on and someone reads a script to you basically and, and, you know, out you go.

11:52So, I mean, I don't think any of this is good. I'm not like happy about any of it, but I'm also just sort of like, kind of, this is what people want. But I do think, you know, the flip side of this is just like, it can be hard. People want to talk about culture all the time. It's like, these, these are the things, right? Like you are establishing a culture. If you're doing this by email at 6am, there's only so much, you're going to create a level of detachment with your workforce. That's just has to be done because people have to like have a way to function right they can't come to work and be like i believe in the mission i love everybody and knowing tomorrow at 6 a.m you could get an email so you're always going to come with a little bit of skepticism right so those are the real trade-offs that are going i guess the actual process of how it happens it's like well this is the culture you want to create right and people always talk about culture but i'm like these are the moments like

12:39Aaron: this is how you do it you're sending a sign about your culture yeah fair i also wonder so i have I've been on the side of one of those famous HR meetings. I've been on the downside of one of those in my career. Yeah, maybe I do want to receive the email. I don't know. Let us know out there what you want. Do you want the email? What's the preferred way to end your time at a company? All right, man, let's move on. So I would say probably the biggest, I mean, there's a couple of things about models and other things like that, but there's always model launches. But I think the biggest news story in AI was this pacing debate, if I call it that, of, you know, we've been going strong, going strong.

13:18Aaron: Everyone's up and to the right, up and to the right. And now all of a sudden, everyone all at once was like, oh, we need to slow down. And it seemed like it was a little like everyone do it at once. And we're kind of wondering and everyone's like, oh, this is independent. This is independent. I'm kind of combining our first and second stories all together. And one of, like, it turns out, actually, they might have been talking all along behind the scenes. And when I say they, I'm talking about, like, Anthropic, Google, OpenAI. They may have all been – this may be a coordinated thing. And so I guess I'll ask you, Brendan.

13:55Aaron: First of all, A, what's your thoughts on the slowdown? B, what's your thoughts on the coordination of the slowdown? And then we can take it from there. Yeah, I guess thoughts on the slowdown is sort of like – I kind of think of it more around, you know, Because I guess basically it kind of comes down to like, there's the belief that like, if we don't slow down a lot of different proclamations, like, you know, end of humanity in 10 years, three years, depending on what's going on. So I guess we have to kind of start there and just say to ourselves, like, are those like credible actual things that we really need to think about?

14:25And where is that coming from? And I kind of start from kind of a point of view of like, it's not because someone's great at building models. It doesn't mean that they're necessarily great at predicting what's going to happen. Right. So they kind of can't. Now, it doesn't mean they're wrong, but it doesn't necessarily mean they're right. And I think that's the place we need to start. It's like, do these people have information beyond knowing how these models are progressing at a level of detail? Most of us, they don't. Beyond that, is there some set of information that they have that really gives them a view into this that we don't all understand?

14:55And if so, I kind of start with like a transparency argument. I think it's their role to kind of educate the rest of us about like, why do you believe this? What exactly is going to happen? If it's the end of humanity, then like in detail, right? Not just like everyone's heard about the paperclip thing, right? And it's like not just repeating that, but like what specifically do you think is going to lead to this? And if you can't educate us, then it's like anything else. It's like, well, there's risk here and there's risk with lots of things, but we kind of live with it. So I guess that's kind of where I start on the whole thing.

15:27And then I don't know. I tend to just believe incentives and human interest, everyone's just going to do what they is in their best interest. And going back in time that Anthropic only exists because Dario and team left because they felt like OpenAI wasn't taking it seriously enough. And of course, Elon Musk famously started this whole thing or funded it. And then he left because Sam Altman wasn't doing what he wanted. And so I have a hard time believing that these three people could agree on where to have lunch, much less how to do like an orchestrated PR campaign around pushing for regulation.

16:05If anything, it's sort of just like happenstance. It's like the researcher for Anthropic post it. Right. And then the comms team sort of like, hey, this is a great opportunity. Let's let's because they believe it serves their own interests. Right. So I think it's just more along those lines. And then, of course, like, you know, you have the other side of it where, you know, Jensen Wong, you know, kind of representing like, no, like, let's keep, you know, in his economic motivations are very obvious, right? It's like he wants to keep selling chips. And so he doesn't want there to see suddenly be like a slowdown.

16:34Right. So he's very much like we got to keep going and things like that. So I kind of come back to like, I think everyone's kind of like, if you will, trying to serve their best interest and there isn't any coordination. But I'm open to new information. What do you think? Do you have like the tinfoil hat on? So first of all, two things about this.

16:53Aaron: So I agree with you when it comes to we need more information because here's the thing. Dario, I feel like has been at this point is the boy who cried wolf. And we just never know to believe him because I mean, all the way since what GPT-2, he's been like, nope, this is too dangerous. and so he's been crying wolf for so long it's hard to know when we're like oh okay now we should listen to him and so the problem with it is is if he's going to be the spokesman for this it's just not a good look now all of them getting together and saying it i feel like yes absolutely that had to have happened it was way too coincidental but then there's the argument of hey if if the Americans don't go fast, the Chinese will win.

17:41Aaron: So there's two sides to that. One is, hey, we all need to keep going fast because of that. But the flip side is, and this is the tinfoil hat conspiracy part, well, what if they're just distilling from the American models? And well, if there's no American models to distill from, they're going to slow down too. So it almost would be an interesting test to see what happens now i'm not saying like i want the americans to let out and then the chinese models keep going i'm just saying it'd be interesting to see the trends because there's so many theories of who's going to win and what's going on and all these paces and i feel like no one has enough information and no one will ever have enough information to actually realistically talk about this other than two chuckleheads on with an opinion on a podcast like us that's right i'm admitting that well do you ever just kind of a little bit of a side here do you ever think we're kind of living in like one of those like uh like sci-fi uh movies that has time travel in it where it's like the hero is like sense like listen this bad thing's gonna happen and you're the one that's gonna have to save us and we know about it's happening so here's what's gonna happen but then inevitably and this is about to ruin the plot of every movie like this it's like the hero is the one that causes the catastrophe right you know and it's and and it's you know again that's you can pick many like terminator there's like many of other movies that do the same kind of thing and it's always the irony and at the end it's like right as he right as the hero is uh either trying to save humanity he realizes oh no like i'm the one that actually caused this whole event right so i think there is to your point like when you're saying that i'm always thinking it's like we have to build the models because if we don't build the models then something bad is gonna happen it's like no but like we are the people like you know it's like we are the ones like you said you said it perfectly because you're like well if we if china and other people were using our models anyway if we didn't build the models there wouldn't be any miles so anyway that's just a long it's just a weird long aside that sometimes i just feel like i'm in a sci-fi movie but your point is just like i just think it's just more human human nature humanity it's just like you know there's this technology that exists in whether it's the americans or some group like some group of people is going to continue to pursue this area of research and build it like regardless of what we do and of course like we can be more positive we can be like and that may lead to some significant improvements in lives whether it's you know disease diseases being solved or like new mathematical theories or like better sources of energy whatever you know pick your you know big human problem and maybe it unlocks that right so i just kind of look at like Like, I mean, part of me just comes back to like, I don't, it's just a stupid take.

20:26It's like, you just bet on humanity. There's a writer I like that writes a lot about like financial stuff. And he's just talking, he's always talking about like, you know, every generations or every five or 10 years, there's like, it's like, should you pull all of your money out of the stock market? And he just kind of shows like, you know, it was like there was a world war and there was atomic, atomic bombs and there was, you know, diseases. And he goes on and on and on. And he says like, at every stage, people kind of felt like maybe they should take it out. But for the most part, humanity, if a long enough, given a long enough horizon, like things generally get a little bit better over time.

21:00So part of me is just like pattern matching that way and just being like, I don't know what will happen, but I just humanity will figure it out. Right. That seems to be what we do. But maybe that's overly optimistic. What's your take?

21:11Aaron: It reminds me of like, if you're the CISO of the organization, like all it takes is one time and you're done. like you have to prevent against all the security incidents and you have to be perfect and it's almost exactly that like it's like well what happens if we're wrong though so i feel like this is such a weird fomo thing of like or to actually fud more than forma fomo but you know it's like everyone always falls back to but what if we're wrong what if this time is the time but i think that's a good one to like think about like two things i think of this week right One was, and maybe we'll talk about it, Salesforce had their big conference, right?

21:50And they're talking about headless agents and using chat, using Claude. It's like, first of all, welcome to an idea we all had like two years ago. But anyway, let's put that aside. But during that conference, Salesforce went down. I don't know if it's up now, but it was like intermittently down. It was just down actually during conferences.

22:05Aaron: And not just a little, like down hard. Yeah, right. And it's like, well, one, first of all, I was like, talk about a great recess. I was like, thank God, Salesforce. There's something we all just could live without. Salesforce is kind of on my list. But anyway, putting that aside, it's like, of course, if we had the meeting at Salesforce, we can never go down. It can't happen. Like, you know, life, the company will be at a serious peril. And it's like, it's fine. It went down. It was annoying. It came back. Everyone's like, it's like, maybe there will be someone at Salesforce now that personally gets held accountable and potentially fired.

22:35And I feel bad for them. It's like, these things get worked out. Amazon, same thing, right? Like how many times has a US East one, like shit the bed and like half the internet goes down. I would say it happens like, I don't know, anywhere from like six to 18 months. And it's like, it's fine. It comes back. I'm sure someone at Amazon gets in trouble and maybe there's some repercussions for them. So I always just think it's like all these things. It's like, I think something bad probably will likely come off AI, right? And we will have to deal with it. And I hope that some good things will come with AI and that we will all benefit from those things.

23:08Agreed. Agreed.

23:09Aaron: I don't know. I just feel like it's a bit of a technology and we all just keep moving forward. But let's talk about Salesforce more because I think the Salesforce one is super interesting because, A, they did Agent Force a while back. So, you know, the whole idea of agents. And now they're releasing their own models. So they have an in-house reasoning model built off of NVIDIA, Neomotron. Was it Super? I'm assuming it's Super because it's probably a big model. But the whole idea there was, okay, they're building their own agents. They're building their own models. i'm one to like hey when the sass apocalypse came everyone was they were one of the prime ones i would say them and service now it's like oh they're going under they're going under no one's ever you know they might as well just close and i was i was always of the opinion well first of all you know salesforce is most people's system of record and if anything the layers on top of it or how you get to it may change but if anything they'll just change and make it themselves and And to kind of go back to Kote and, you know, your other podcasts, like the default lifestyle kicks in of Salesforce has a model, it's good enough, and they have an agent and it's good enough.

24:22Aaron: I feel like there's the moat, right? I feel like Salesforce will continue to go on. And I'm not, I actually think it's interesting. And probably the more interesting thing to me is the whole idea of they didn't build the model. They went into NVIDIA and got Nemetron and then customized it from there. And I see that happening more and more in the industry going forward as well. Like I was kind of talking about this with somebody else recently, this idea of, hey, we really need a fully open model released to the Linux foundation or released to the, you know, somebody else where it was like not just open weights, but actual open source.

25:04Aaron: and because you know people aren't going to necessarily trust the chinese models and all these other things and right now this is an interesting trend to me of people are kind of realizing hey i don't want to build a model that's expensive and i don't want to be doing everything it takes to do the models so hey i'm going to call my buddies at nvidia they're already probably selling me the chips so i don't know i i'm more interested in that was the fascinating piece for all of me. What did you take away from kind of everything going on Salesforce?

26:01Aaron: follow the clues unravel the mystery as civilization collapses 2029 the war for our world a science fiction news series from the future premiering october 30th follow wherever you get your podcasts well i guess i'm probably one level above you there sort of like the model is less interesting to me more than they're just embracing sort of the headless lifestyle so i think they have a whole new like licensing model for like basically you're like you're clawed agent or clawed force. It's like, I don't know, it's like$500 per seat. So that's one interesting that they're kind of carrying over the licensing models of like virtual agents.

26:38But two, I just think this is a realization, like lots of people have written about it. I think we've talked about in the podcast and Ben Thompson's written about headless. It's like, once you start actually interacting through systems, through like natural language and the language comes back and starts answering your questions, like you kind of never want to go back. It's almost like all this point and click GUI stuff that we end up doing is sort of like, yeah, this is very inefficient. This isn't really the way that we think about those things. And something, I don't know if it's in the run now today, but I actually downloaded and tried out Meta's Muse, which is kind of funny to say that.

27:10That's their new, if you will, chatbot agent, personal agent kind of thing. And, you know, obviously there's lots of issues with meta and privacy, and we'll save those topics for another day. But the idea of like, they're kind of getting in that too. It's like, here's how a quote-unquote consumer would start to interact with their systems, right? The idea that you would just kind of talk to it and it'll create feeds and it'll look up things for you. And it's sort of like, I think what we're just kind of hitting on here is like, this is what we want to see going forward. It's almost like moving away from the GUI, natural language interfaces.

27:42And that's what I think Salesforce is kind of, you know, to their credit, right? Because they are one of the biggest systems of records. Like, I think they're trying to embrace that. And I think that's what people want. Now, whether that agent talks to like their model or my model, or I set up some type of like MCP or behind the scenes, some kind of connections, It's like, it doesn't really matter to me. It's just like, whatever gives me the right answer, right? Whatever tells me like, cause I think a lot of people have done this. I've done this in my own little personal stuff. It's like, I have a little pipeline where I can just tell me like, show me the pipeline or who do I need to contact today, right?

28:11It's like, cause that's what the sales reps want. It's like, just tell me like, who do I need to follow up with today? And then who have I not reached out to? And they help me reach out to them. And then, you know what? Just tell my boss or the person managing the pipeline, like here, here's all the emails I've sent this week. update all of the records, update all of the system to reflect what I've done so that my, my boss and his boss can see what's going on. It's like that, that's a very natural way to work versus like, and workday is the same kind of thing. It's like, here's all the stuff I've done.

28:41Why don't you write up a short summary and put this in workday that some executive can read about what I'm doing if they want to give me a review. Like, I don't want to spend time doing that. That's a waste of time. So that's more like what is interesting to me. So it's not so much sass apocalypse it's sort of like let's call it embracing the headless uh way of thinking i like

28:59Aaron: that and by the way too um you know i did some research into this recently the evolving maybe it's in response to all of this the evolving agent licensing models because the whole reason for sass apocalypse apocalypse by the way was like per seed pricing is dead because there's no seats headless right but what we're what we've really did and it's like in hindsight it's like well And there's almost three tiers because there's always three tiers. There's always good, better, best. I mean, it's your basic pricing strategy. There is three pricing tiers for agents. There's a free tier slash pay as you go tier.

29:37Aaron: There's a here's how much you get for this bundle tier. And then, of course, there's always the enterprise licensing white glove. How much are you willing to pay and negotiate it tier? And guess what? It's not that different. Like all we did is took the old pricing model because, hey, guess what? It works and moved it into an agent world. And so it's not necessarily per seat anymore. But the the way the consumption happens is pretty much still the same. And I think that's going to continue going forward. I think that's going to be the big thing. And then the other thing I was going to add, too, I almost feel like the rise of agents and these agent pricing models may or may not kill off MCP.

30:25Aaron: I feel like MCP kind of rose to prominence because, you know, we all needed to solve a problem. But I'm kind of asking you when I say this, I haven't heard as much about MCP lately. And I almost feel like MCP reminds me of like the open API days. And then everyone realized if you give away your API, like they're like Twitter and all these others over time, you know, close down their APIs. And so I don't know, like I almost feel like MCP is APIs all over again. What's your thoughts on something like that? Yeah, I mean, I just think of it just as a structured way to like, you know, have the two systems.

30:59I don't think so much has like gone away. It's a sort of just it's like I think people now understand what it is. And it really just comes down to like what you personally want to do. So if you're using, I like Cloud Code, but I've been playing around now with Codex as well. It's like if you want to integrate with another system, it's just a tool that's sitting there, right? So they have an MCP server and it's like all built in. It's just easier, right? You can just say like add it and it does it. And if it doesn't, it's also usually fine too. It's just like you have to go get an API key and then you have to point it to usually what LLMS, LLMS.txt, right?

31:30Usually has all the information. So to me, it's just like, how do you want to communicate with these things? And I like doing it through chatbots. So however they want me to do it, I don't care if it's MTP. I don't care if it's an API. Just like make it happen for me.

31:43Aaron: Yeah. Yeah. Agreed. Agreed. That's almost one step back from the interface. It's a little bit behind the scenes and how you glue things together. And as long as it's glued together. Yeah. And I think, you know, and what you hit there was like, I think as long as the other side of those services feel like they have a way to charge for what they're doing, then they're going to be fine with it. Like I think like Cloudflare is a phenomenal site to interact with through, if you will. And I know they're rewriting their, I think it's called Wrangler. They're rewriting it. And like, but it really is the, I mean, the Cloudflare interface, like all these interfaces have always been complicated, but like if you don't have to interact with it and they have such a generous free tier, you can do so many things.

32:21And you're like, and the reason you can do it is because you don't have to go to the UI. So I think they're happy to do that because they know on their side, there's consumption based pricing, right? That's really what we're kind of getting at. So I think that's the future. And then, you know, people talk about the SaaS apocalypse. It's like, listen, if Claude and OpenAI can rewrite every system of record, Workday, Salesforce, SAP, the United States, have at it. I mean, chances are those are pretty complicated systems and you're going to need other people working on them. So you're probably going to be better off just interfacing with them.

32:49So it's just like there's just too much domain knowledge that like you can just go replace it all. I mean, and if you can, like have at it, man. And I'm talking about like, you know, you know, is there sort of like cold fusion? And there's like, can we in the United States ever transfer money instantly between two banks? Right. It's like, I don't know which is going to be discovered first. Right. Like, why can't we get these problems solved? So go ahead. Go fix all these problems if you can.

33:13Aaron: Agreed. Agreed. All right. So let's move on to because even though we talked about a model slowdown, there has been maybe this is the reasons. There's two things going on. So A, there's yet another big model, GPT-6 Astra. And so I want to ask you about that here in a second. But then also there is the bigger the models, I feel like it's the bigger the cybersecurity problems, which hence leads to probably the slowdown reasons. And so we just have this combination of, yes, we have bigger models. And yes, we always have more and more security headaches. But let's break them apart real quick. First of all, have you tried 6 Astra?

33:50I have not. I don't think so. I think the closest I got was sold.

33:53Aaron: Are we to the point we don't care? Well, we're to the point that you have to put it in code. You have to make it available in one of my subscription plans. Because now the, quote, best models are always like Fable and stuff. You have to go use the API key. So it makes it more expensive. And then two, I find myself, it's like I've become very conscious of like, I actually do think a lot about like, am I doing something that I think requires a lot of thinking? Because the credits go so fast when you use them. Or can I just, am I doing Sonnet level work? or one of the more advanced opus level work.

34:25And I think the same thing here. And now the part, and I feel like for like regular people, that's where like most of us are that are using it. The part that like I, and you're closer to this, so maybe you're the one that can see it. It's like when they start talking about spinning up agent swarms, you know, hundreds of agents. And I guess this was back to the, you know, I think I think we talked, you know, the idea of the thing that broke into hugging face. It's like spinning up, I don't know, hundreds of agents over like some period of time and having these like wave of them. It's like, I don't know.

34:56I've never done anything like that. I don't even know what I would do. So I'm always, I guess I'm honest enough to say like, I don't know what people are doing with that stuff. And maybe, and I don't know how I would use it. And so when they talk about Astra and those other things, that seems to be what they're talking about. They're like doing the, or like, well, maybe the same thing with that math problem, right? The millennial math problem they solved where they were just like, all these math agents were working for three weeks doing it. And it's like, okay, I don't know. I don't really relate to what's happening there.

35:27So I don't know. What do you see? I mean, you're in this world close. Are agents swarms? Is there some practical use other than creating some giant cyber attack? What do people do with it?

35:37Aaron: So here's the problem. And for those out there that have listened to the podcast a long time, I don't have time as much for it anymore. But I'm a gamer. I grew up as a gamer. Brian hates it when I talk about gaming because he wasn't a gamer. But I'm going to bring in some gaming analogies here. Agent Swarms, Astra 6, all of these other things. It reminds me of back in my MMO days, so playing World of Warcraft. And you would be playing it. And, hey, you know what? At one point I was pretty serious and I did put a lot of time in and was going through the progression ranks. And I was one of those hardcore people.

36:12Aaron: And it lasted all of, I don't know, six months to a year because, holy shit, that's hard. The problem with what we're doing now is all of this stuff is so top end. It's so long tail. Like when you're talking about agent swarms with thousands of agents and you're talking about folks that could really push an astro level model, the percentage is so small, like it's 1 % or less. and but it's what gets the headlines and it's what people talk about and it goes back to like my gaming days where when i was doing world of warcraft the new raid would come out well guess what one percent of the people actually ever got to the content and went through the content and you were constantly like that was the thing and everyone be like oh i'm so bored and it's like well did you get through those it's like no no i didn't it's like well okay you got work to do And so it's for me, this is such a long tail niche case, high end, but it's what everyone cares about.

37:13Aaron: And I guess it's because I mean, again, look at us talking about it. It's what is the news and it's what people talk about. But at the same time, I don't think it's what most people do day to day. I think if anything, like you and I are probably power users now, but we're power users on a small business individual level. Right. I, you know, I don't think even at the enterprise level, there is a lot of the swarms going on because I don't think the trust is there yet. And I don't think, you know, all of the rights, gosh, what's the words I'm looking for here? Like human in the loop is in place for stuff like that.

37:53Like, I just don't think most places are doing that.

37:57Aaron: And I really feel like this is just research top end, the folks always being like, look at us, look at all these things you could do. But most enterprises today aren't doing it anyway. right now what about the research side because this seems if any place it seems it's going to be used it would be the researchers building the models to like achieve whatever they call it like you know basically the recursive self-building model where ai makes itself bigger and bigger and it's like you know this is the part where like it really is a black box to me because i'm always like they always talk about well this is like a reinforcement like research scientist and i'm always like what are they doing like what are they just looking at the vectors and the and the math And I guess that's the question I would be interested in someone kind of explaining like, okay, how are they like one, what are they doing to make the models better themselves?

38:48And there's a bunch of stuff we could talk about that I do know about. But then two, it's like, what are they then like teaching the models so that the model like learns how to do it itself? Because that, I guess, would be the place that if any place it would work, it would be there. It would be like, oh, we figured out how to teach an AI, how to teach itself. And yeah. And then the AI can figure out like it does make sense for us to have like whatever. 100 ,000 agents doing this because it leads to a better model faster. So I don't know. I know you're like, you have like relationships with researchers and stuff.

39:19Like, like, what are they doing? Like, are they like, yeah, I'm ready to fire up 100 ,000 agents and like do research? Like what's going on there?

39:26Aaron: No, because again, I just, I don't know outside of like the Google DeepMinds of the world. I don't know that many independent places where they have the resources in the budget to go do stuff like this. And so what I think is going on, and maybe that is also why to kind of go, we were talking about the security angle of all of this. A, this is why security comes up so much because we've reached a point, the models are cracking the benchmarks and then they keep coming out with new benchmarks like SWE Reel now, which is the newest one, or that we unleashed it onto the math problems, the unsolvable math problems.

40:04Aaron: And we won't go into the drama behind that one that happened recently but but i did kind of i did kind of find it funny though that basically it was going and cracking all these things and doing all these things and everyone's like this is amazing this is awesome and then the second we cracked a math room all the math people were like whoa whoa whoa whoa wait a second this isn't human like and so you have this just weird reactions of we've reached a point where maybe honestly the researchers are getting a point They're the same way we talk about the models have been trained on basically all of the data available.

40:39Aaron: The researchers have started to point it at all the things available, right? Like, okay, security, it's like it's going through things like Swiss cheese. It's it's solving math problems. And so it's interesting to me because, again, I feel like this is all such long tail stuff. This is all stuff that the enterprise doesn't need or care about, quite frankly. Right. It's sort of just like metrics. Well, I always wondered, this is the part I always wondered too, is like I guess the answer, I'm going to answer my question before I say it, but I guess it's because they know they can't. But I'm always like, well, that's cool they solve all these math problems, but why don't you just sit down and give the task, like design a very inexpensive solar panel that uses a superconducting material such that we can all have like energy for a very low cost.

41:29Like, why don't we just put an agent swarm on that? Right. Because like if out popped like the design and everyone's like, great. You know what I mean? Like, but who doesn't want that? Who doesn't want like, you know, near infinite cheap energy. Right. It's like, so it's like, I don't, I mean, I guess obviously that's a hard problem. When I just said there, it was like probably impossible, but it's like, I don't know. It's always like, you know, it's always these other things, right. Where it's like, they're kind of like the fear based things, right. It's like, oh, all math is done. Cybersecurity.

41:53And it's like, well, why doesn't someone try one of these other problems that we clearly could benefit from? So I guess, like I said, I know the answer is probably too hard, but it would be great if somebody at the AI labs would be like, let's work on one of those. And let's announce that, right? And be like, here it is. Here's the cheap solar panel anyone can build, and everyone's going to have a lot of energy. That would be great for us.

42:13Aaron: Yeah, I agree. I agree. All right, man. Well, you know what? I think we're about at time, but I didn't know if there's anything like you wanted to do any quick rounds or anything in the backlog that you wanted to mention that you thought was interesting here in the last two weeks. I think we've went over tons of stuff. I would just say everyone go out and, you know, enjoy your weekend and time with your family. It's like, listen, there's plenty of time for doom and gloom. But, you know, it's also time for college football or whatever your pastime of choices. Like, just go have some fun with the family.

42:44Aaron: Yeah, absolutely. And I will say this. So, you know, as we mentioned, we're moving this to every two weeks. And we promise every article or every episode won't be doom and gloom because guess what? In two weeks, everything will be up again and everything will be, you know, back on positives. So, you know, this one may have been a little bit of a downer show, but there's going to be, you know, sunshine right around the corner. And, yeah, definitely echo everyone out there. Thank you very much for listening. Thank you very much for your feedback as well. We certainly appreciate it. And we will talk to everyone next week.

43:19Thanks for listening. Check us out at theenterpriseaishow.com for past shows, newsletters, and all things Enterprise AI.