SPEAKER_03 0:18

Alright, hey everybody, welcome to HT Coffee Talk. This week and what's going on in the world of tech and things and connectivity and so I can't somebody else. Somebody else is something we got last year coming in here. So what's going on with you guys? I don't we don't have any electric guitars to start off with today unless Pete pulls up.

SPEAKER_04 1:19

I would have no actually not, and I'm not really hooked up for it. Otherwise, I wouldn't.

SPEAKER_03 1:23

You're not hooked up for it? Oh man. Leonard's at uh yeah, go ahead, Devin. You pull do something. Anything.

SPEAKER_01 1:32

You don't you don't want me doing that.

SPEAKER_03 1:35

Oh my gosh. Yeah, yeah. Leonard will be on in about 30 minutes. He's at the Intel thing. So you know the deal.

SPEAKER_04 1:43

I think it's the AMD thing, isn't it?

SPEAKER_03 1:44

Is it the AMD? You know, he was texting me. He's like, oh my gosh.

SPEAKER_04 1:48

X86, you know.

SPEAKER_03 1:49

X86 going on the chip thing, chiplets.

SPEAKER_00 1:55

Have you guys heard about NVIDIA? I mean, what was the intel in the MD? Have you heard about NVIDIA?

SPEAKER_03 2:00

I mean it's I don't know who those guys are. Yeah, I think I think I bought an NVIDIA graphics card in the 90s. Is that what is that who you're talking about?

SPEAKER_01 2:11

Oh, PCs and I found this. Whoa, look at that. It just got real. I love that. Well, yeah, I mean it's an actually 9800 GT. Outstanding.

SPEAKER_00 2:26

Well, it's an interesting command for. Yeah. You could have thought that you know the gaming industry will die and NDA we die, then the Bitcoin mining will die and NVIDIA will die. Maybe AI will die and NVIDIA will die.

SPEAKER_03 2:43

Who knows?

SPEAKER_00 2:44

It's actually a great pivoting strategy.

SPEAKER_03 2:47

You know, a blast in the past. So old video cards. Does any remember something called Vesa Local Bus or Visa Local Bus? Visa Localbus, sure. Do you remember that? That was like a thing. I remember that.

SPEAKER_04 2:58

I remember back with uh 3DFX and uh all these guys, and then DirectX came on the scene, OpenGL, like it used to be the Wild West, remember? It was the Wild West. Games had to be ported for specific channels.

SPEAKER_03 3:14

So you know, it took a while for Windows, everybody just is, you know, oh Windows and gaming, you know, and obviously that's that's helped Windows stay relevant in a world where more and more people are using the Mac. But um but if you go further back, you know, it was not certain that we were the DOS games dominated the world at that time back then. And the reason is because DOS games can write right to the frame buffer and everything. Like you could do anything in DOS, which is weird that I'm saying this. Um and so when the big change was when Windows 95 launched, and they I remember they gave us a pinball game and people were trying to build games for it, but but you're right, Pete. You couldn't get that performance, that DOS performance and graphics until we until DirectX.

SPEAKER_04 4:05

No, they all had their own special drivers, and then the game developers, it was almost like the console wars, right? It was like this game's only available on this card, you know. Yeah, oh, yes, right.

SPEAKER_00 4:16

That that was following each other. And that was the day where the Watcom C was very popular because it allowed you to go very easily lower level in the DOS operating system, and and Doom was built with Watcom C. And over there was Whatcom was a Canadian company out of one of the university in Waterloo was also they were also the creator of uh Whatcom SQL, which uh which was uh initially uh uh embedded uh by PowerSt are their kind of default database or a development tool. And it became so popular that PowerSoft ended up buying Whatcom SQL and that became the SQL technology that then went back into Cybase and ended up being the mobile SQL database. So we're very, very old here.

SPEAKER_04 5:02

Well, but also if you remember, an interesting blast from the past too is that so that we had this chaos where everyone was coming out with new and faster graphics chips for gaming and convincing game developers to port to their chips, and then OpenGL arrived. Yeah, and John Carmack, who created Doom, which was the Sinquanon. Oh, oh, uh, I'm going to support OpenGL. So he uh so he decided I'm going to put this thing on OpenGL, and that was the moment when OpenGL became sort of the new um kind of hardware acceleration, direct hardware access standard. And one of the first companies to accept to adopt OpenGL was fill in the blank, NVIDIA.

unknown 5:51

NVIDIA.

SPEAKER_04 5:52

NVIDIA. I guess most of our they kind of leapt to the front because they adopted OpenGL with Carmac.

SPEAKER_05 6:00

Yeah.

SPEAKER_04 6:01

CarMac kind of made that, made OpenGL what it was.

SPEAKER_03 6:03

So uh NVIDIA and what ATI? Like in the graphics car wars. Um that's how many of us knew early on that Dr. Lisa Sue and her gang over at AMD would be a second place contestant in the AI acceleration race. And we knew it early on, and we predicted it because we knew they acquired ATI a long time ago. And so that gave them that graphics card GPU capability that all the other players didn't have, except for speaking of great Canadian companies.

SPEAKER_04 6:37

ETL is a great Canadian company. There you go. Wow. Wow. So there we love that. All it all circles back. It all circles back. Circle of life.

SPEAKER_03 6:48

All right, Alistair, what's going on with you? It sounds like you got a new job.

SPEAKER_04 6:52

Alistair is a new CEO. He's in the CEO club now.

SPEAKER_06 6:59

Um really actually really interesting startup. Uh the founder spent some time at Microsoft actually. He didn't overlap with either of us. We'd already left. Um he was part of an acquisition of a company called Bonsai. Anyway, he started uh Mesa like three years ago, um, and it's uh essentially a platform which allows you to um uh improve your data, build a full realistic simulated environment, create AI agents in a no-code environment, um, train those agents using deep reinforcement learning on that synthetic data set, um deploy those agents into a live production environment, orchestrate teams of agents, and it's kind of a couple really novel ideas. Um, one, the kind of synth the building of a synthetic data set works really well in kind of closed loop environments, so manufacturing environments where you basically know all the variables, not out in the real world where you know wind and sun and temperature and stuff like that changes. Right. That the method using used to do that is is really interesting, and then just the thought process about how you think about segmenting tasks to build AI agents. So super interesting, company. Um, small team, um, early days, startup, you know, um pre-series A, so running around like crazy with our heads on fire. But we have customers that you Nestle. Are we on recording yet? No, you have customers, that's all.

SPEAKER_03 8:27

We're recording this. Uh-oh, I just heard the N-word Nestle. No, I didn't.

SPEAKER_06 8:31

Um some very um some very, very uh blue chip manufacturing companies with whom we're already working, who've already given us kind of these core data sets to go figure out how to apply AI. It's quite incredible, actually. I I'm I'm I'm I was shocked by the number the range of customers like they gave you one. Um which speaks to the credibility of the founder, a guy called Ken Sananderson. Um so yeah, super interesting. Like back to back to early stage startup craziness, um, but but really interesting space. And it's kind of, I mean, you and I, Rob, we've we've been in around well, all of us actually, we've been in and around, like, well, first of all, what you need to do is get more data, and then you need a platform to ingest it, and then oh shit, you need more data still, so you need to connect more stuff, and you need then you need some edge intelligence because maybe we can do something like but but wrapping all of that together into true kind of autonomous industrial automation is is no one's really figured out how to do that. And and I think these guys have got as good a shot as any. So that's great.

SPEAKER_04 9:37

Sounds exciting.

SPEAKER_03 9:38

That's exciting. You know, when you walk through what you're saying that the technology does, I think most people assume that you would have to go to one of the big tech companies to get that kind of stuff.

SPEAKER_06 9:50

Like people would be surprised and not expect a startup to, yeah, we have this whole environment and to let you build autonomous agents and low code and then train them, and then they go off and do, you know, it's uh the customers that we talk, the customers we're talking to, I think that they're I think that big tech companies are too focused on trying to sell you what they've already got on the wagon, um, then they are necessarily trying to think about how to solve your problem in the innovators' dilemma, right?

SPEAKER_04 10:19

They are uh hesitant to um you know break into new areas and can cannibalize their existing right.

SPEAKER_06 10:27

So this is I think it's part of a kind of overall the gap in industrial, I think one of the many gaps in industrial has always been, you know, it's people like us building the solutions, not um, you know, the plant operator, not the the you know, the water plant manager, not the you know, and so the solutions that we have ended up building over the years, they don't they don't meet that person where they are with something that is kind of intuitive and doesn't require that they learn how to work in Python and doesn't, you know, that just freaking works. And and that's where the founder is mechanical, he's mechanical engineer by background. So he's he's started from well, how do you do this at a process level? And our first step with customers is is is basically sit down with the team responsible for running the process and figure out how you know how do you actually do that, how do you make the machine work that way? How do you um and I think the the pressing priority at the moment for most of these companies is a ton of their seasoned folks who actually know how the plastic extruder really works are leaving, they're retiring, you know. And so you can give the new employee the manual and say, here, read this, but it takes 20 years to figure out how to tune the actual piece of equipment. And so if you can you build an AI agent that works alongside those workers, which is our intent, um, then it helps them get up to speed much more rapidly, and you know, it gives them 20 years' experience in a day. Um, so it's really I've really, really interesting. There are a bunch of other companies. I'm still with my mentor, and obviously, yeah, there are a whole bunch of other companies in this space kind of going after the problem from slightly different angles. Um, I don't know that we've necessarily figured it out. I think we we're in the process of, but I think it's an extraordinarily valuable space. And it speaks to your bookbook, you know, it it the this this is the where the metal meets, you know, rubber meets the road kind of thing. Um if we can't actually get into these industrial processes and drive meaningful improvement, then it's you know, it's all for naught, you know. I mean, there's no point to me there, because you know, all it is is interesting suggestions um that that stay on a screen and don't actually live on a factory floor. So really interesting.

SPEAKER_00 12:38

It's actually quite interesting, uh Alistair, because I think one one thing I've learned personally from the the the the and and Rob you are in this wagon as wagon as well. The the the days of the IoT platforms, industrial IoT platforms that were supposed to be be these generic, you know, fantastic things that would improve all processes. Uh what we learned to your point is that all those industrial properties are very much uh custom-made system integration with a lot of knowledge to implement the technology. And we could never deliver a platform that was flexible enough that it was too generic to make technology. So if you think from that angle the the the the the coding ability that those agents can actually make that bit stop at scale, then the platform becomes very different. It's not just the database, the data flows, and the AI model. So how do I understand the business problem? How do I understand the equipment and how do I apply intelligence to be at scale? So I think that's one of the I do think that one of the biggest categories of software that is going to be challenged by AI. Because the CRM system is an extremely complex and uh and expensive system because it's so capable that you have to tweak it and you have to engage to customize it for your own business. I can just describe how your business works and generate a custom screen and I see people doing that over and over again. So we might go into a world where you just describe your business processes, the things code it for you, and the business process change, you regenerate another CRM. If you have a good data discipline with the core fundamental data, I think this is possible to do. So I can see an interesting parallel here. That's but that's pretty easy.

SPEAKER_06 14:27

Yeah, that's I mean, I think there's that's kind of what we're doing at EMESA, which is the data request. What we've been doing at Momenta, you know, this couple of folks internally, um, they've basically rebuilt the firm around AI. So things like um Salesforce, which we don't use, but we use HubSpot and all that, they just become historians that sit behind an MCP server. And we have an AI layer that runs across the top that integrates everything on SharePoint, everything ever said in Teams, everything said in email, everything in our CRM, everything in our sales process, everything, everything, everything, everything. It basically has a 360 view of the entire business. And you can do things like obviously the business of a VC is evaluating new investment opportunities. You can use it to evaluate a data room with a thousand items in it and provide you with a concise report that says, should you invest based upon you as a company and your thesis? And one of the things we've had to do is make it very deterministic by give it giving it lots of we want it in this form, and this is the source station. This is how you should think about X and so we don't get any random shit out of it. But it's nuts, it's nuts.

SPEAKER_04 15:32

I mean, the performance improvement I it's well the other thing about CRM that's pretty cool, we use HubSpot too, is in the bane of a lot of you mentioned about good data, and a lot of times the good data isn't there, but things like HubSpot now will do a lot of good data enhancement. So all of your partial records, of which 100% of your records are partial, it will fill in the blanks by using the web and LinkedIn and out to use.

SPEAKER_03 15:57

I love I love things that fill in the blanks. Kind of like you know, when we capture the mosquito in amber and we got the DNA sequence for all the dinosaurs, but there were some blanks that we had to fill in. And so we used uh DNA from a frog and we plugged those in to fill the gaps and you know what could go wrong. So yeah, I I love that. Oh my god. Hey, Devon, does does your team Devon obviously is running a big IoT team? Do you find yourself built a lot of bespoke stuff? How does how do those projects look?

SPEAKER_01 16:28

I mean, I again, you know, what I love about all this is you can't take the human out of the whole equation. And absolutely, you know, people try to shoehorn technology and without looking at the end user experience, the process, and not only the process, but as you know, Alistair, when you go in and you talk to those operators, there's always these workarounds that they all have in their head. How do you figure that? But the number one thing is trust. And unless there's trust in the technology, um, people are not, you know, you can make all the AI can make all the recommendations you want, but some human at the end still has to say whether this is not gonna damage, permanently damage my operations or my equipment, because there's some tribal knowledge in there and human intuition that you still can't replicate with AI. But everything you're saying, Alistair, just makes me go back to that whole, you know, when we were back in kindergarten, you got the grade card and it says, Do you play well with others? And we're gonna start thinking about do agents, do robots, do all this technology, do they play well with the others? And that's a human being, the other agents, and things like that. So absolutely, you know, you just don't take it, you know. I I always say technology is only 25% of the equation. You know, ROI is a second, security regulatory, that's a huge part that, you know, I say that not just make it the four legs of a stool, but the people, you know, you unless you can get people to change behavior, unless you can get people to trust things, you're not gonna get a lick of ROI. And that's why you have to understand it from a human level and from a process level of how your technology is gonna fit. And so I love everything that you know you're doing with these agents, and here comes the energy.

SPEAKER_06 18:10

It's dealing with the unpredictable, I think, because like even in a closed loop environment in a factory, if you're making, I don't know, making tin cans, you know, yes, you've got visibility thanks to all these wonderfully connected sensors you have now, and you've had a data ingestion, you've got the data. There's one part which is there's a whole bunch of stuff that's still going to happen that is unpredictable, and that's where the human comes in because they're able to cope with these different barriering situations. You know, it's more humid today than it was yesterday, and that actually affects the throughput of the machine. It doesn't say that in the manual. There's that aspect, the trust aspect, I think, is being able to create an authentic, you know, a realistic synthetic environment that then you can train agents in and then demonstrate those agents. So what a MESA and others are doing that well, a Mesa for short, we take the company's threshold, you know. Okay, so your output variable is the number of tin cans you produce an hour, and your input variables, you know, there are set point variables around the metal, there's no cancer variables around top three, blah, blah, blah. You know, this is the best you ever achieved was 85%. We can show you the agent delivering 95%. And and that kind of because for a small startup and yet large companies are trusting us with data, that's why I think they trust us, because we can show them the agent running in a in a realistic environment and say, look, and and also it's then the visibility of caps and governance rules. Like, okay, the agent can't actually, I don't know, um, switch off all the electricity because it saves money in the price, you know, it can't do dumb stuff in order to try and win the win the reward, you know, win the win the win the game. That I think too, but you have to show people that in action before they go, huh, okay, so this isn't this isn't gonna destroy my factory, this isn't gonna, you know, I'm not gonna I'm not gonna harm myself through taking a risk with this new technology.

SPEAKER_03 19:58

Right. Hey, Len, what's up? SFO Yeah, it's like playing James.

SPEAKER_02 20:09

Yes, I am. I I I actually made it through. Little do they know, right?

SPEAKER_03 20:16

Yeah, little do they know that you're gonna take down a plane. Yeah, that's real bad for all those people that you're gonna be with. Uh yeah.

SPEAKER_04 20:25

Find out JSX.

SPEAKER_06 20:29

JSX JSX. So JSX is a small airline that flies um Embraer jets from private hangars on airports to other small airports, and it's cheaper and it's cheaper than Southwest.

SPEAKER_02 20:48

Wow, did you say inbred flights?

SPEAKER_06 20:51

Embred. Yeah, yeah, those are you know, regional jets, so it has those. Oh, okay. Like business seats on, and yeah, I'm actually gonna fly for the first time next week because I can fly from Burbank to Concord. My office is in Walnut Creek. I can fly to Concord, which is 10 minutes away from Walnut Creek, not where you are, which is where I went last week. SFO.

SPEAKER_04 21:11

Some kind of uh CEO flight thing that Alice.

SPEAKER_06 21:15

It's like CEOs only Southwest. Yeah, it's cheaper than Southwest. It's like I can't get my hand on it.

SPEAKER_00 21:22

It's the first step to the private jet, but yes, yes, right.

SPEAKER_06 21:27

I have a question on Pete's one, I think, more realistically.

SPEAKER_02 21:31

I have a question for Pete. Yeah, so is AMD part of uh AI or Edge AI Foundation?

SPEAKER_04 21:38

Not yet, actually.

SPEAKER_02 21:39

Not yet. Yeah, I mean I I brought you guys up.

SPEAKER_04 21:43

Yes, actually, I'm gonna be at AMD in about a week because we are sponsoring their they uh they we are co-sponsoring with them the ML Systems Rising Stars Scholarship program. So I'm gonna speak at AMD. But yes, they are on our list to have a chit chat with.

SPEAKER_03 21:58

Oh, good chit chat. That's where it all starts. That's where it starts. A chit chat. Yeah. Absolutely. So that's part of the sales funnels.

SPEAKER_04 22:07

We were just talking about uh before you joined, Rob was talking about kind of AMD, the acquisition of ATI, and kind of the always had some big guns, you know.

SPEAKER_05 22:17

Yeah. Xylinx. Xylinx was another one.

SPEAKER_04 22:21

Yeah. So uh I assume that you had a good robust experience there with uh their AI.

SPEAKER_02 22:28

Yeah, yeah. It was pretty cool. Yeah, I was uh well they uh they threw me out here to participate and witness uh a dancing AI, and uh apparently it's gotten a lot bigger, which is good news for AMD. Cool. And um yeah, they announced the their their massive Helios rack. Yeah, double wide massive beast of a machine that's uh their first uh Rat Scale Rat scale solutions for AI, ridiculous supercomputing. Yeah, yeah. Uh but you know the most interesting thing for me was uh their friggin' rockum.ai. Did you guys pick up on that?

SPEAKER_03 23:15

I saw a little bit about that.

SPEAKER_02 23:18

CUDA competitor, right? Is it a CUDA competitor? Well, no, no, no. Uh Rocum is the CUDA competitor. The Rockham AI is it's basically a GENTIC um tool. And I think that's that I mean an underlying tool that uh allows someone who might be, let's say, not such a savvy, well, they might be a a kernel programmer or engineer to basically um you know accelerate their work in uh optimizing kernels or models on top of AMD hardware. And so it I mean it if anything, this can drain some of them the CUDA mode, but it's one of those ironic things, right? Where everyone talks about how oh AI is you know like generative AI is so good at coding, right? Well, um, what if that talent can be used to undermine the uh differentiation that CUDA has in the market at scale? Because I mean, think about it. One of the problems, um, one of the problems for NVIDIA here is that uh the slot dynamic will basically level the playing field, right? I mean the diffusion of capability uh you know, tunnel programming, which is very, very difficult, very low level stuff, to be able to abstract that up using uh these types of tools, these agentic tools. That's that's a huge uh potentially equalizing uh capability that can have that, that can ride that slop effect. And um I don't know, man. Yeah, it's gonna be really interesting to see what happens in the next, I think, year.

SPEAKER_03 25:14

Can you uh can you trade can you trademark that phrase slop effect? I like that.

SPEAKER_06 25:20

I like that TM TM TM every time you say it. I think someone I think someone in Lithuania already sniped that one, Robert.

SPEAKER_02 25:28

Yeah, well, you know, people don't realize that we innovate on IoT copy talk, right? We come up with all kinds of ridiculous terms that eventually become mainstream.

SPEAKER_03 25:38

So do we think that so do AMD they got the new Helios, the whole new rack thing? Is that something that would fit into an NTT data center? Did I see Devin on here?

SPEAKER_02 25:51

Is he still on? Yeah, he's right here.

SPEAKER_04 25:53

Yeah, it's been on the whole time since last week.

SPEAKER_02 25:57

So we didn't scare you off.

SPEAKER_00 26:01

We have to try harder. We have to try harder.

SPEAKER_02 26:04

Yeah, man. Uh hey, kudos to you. You have some. Oh, that's awesome. Hey, welcome back, man. It was good to see you on the on the whatever the Brady Bunch grid.

SPEAKER_01 26:20

How can how could I pass up another one? This is where it's all happening.

SPEAKER_03 26:25

This is where it's all happening. See the Alistair, you said something about, oh, I'll just rebuild my CRM or I'll just recreate it, you know, that kind of thing. Fast and loose talk around stuff like that. And yet there's been a lot of talk around obviously AI helps us code better, and then people are vibe coding and they're using, you know, cursor and others. And and then so Starbucks kind of made that a little bit real when they came out and said, we're just gonna jettison $400 million worth of money we have to spend on Microsoft and others for our CRM and supply chain management. We're just gonna build it ourselves, like you know, over the weekend or something like that. There's been a lot of fast and loose topics around Visual Basic, yeah. And so, but and it's like, well, is it is it real? You know, are a lot of is enterprise software threatened because someone can just crank it out.

SPEAKER_06 27:24

Unless you're nuts, you know, unless you're completely insane, building it all yourself with an AI agent from scratch is gonna wait until you find out the first big problem. And if you're a big company, that big problem could cost you millions of dollars. I think what will happen is more everybody's producing MCP servers, so you can access all the data that sits in a CRM or a story or whatever. I think what will happen is that um it's kind of a abstraction layer essentially, an AI layer sitting across the top, all of these systems. I think the new feature set that going forward, the future roadmap is gonna be owned by that abstraction layer, not by the underlying. So I think is is is Microsoft gonna go out of business tomorrow? Is that you know is their CRM product gonna go disappear? No, but I think their ability to do that.

SPEAKER_04 28:10

I think there is gonna be products, there'll be downward pressure on a lot of these kind of legacy, expensive legacy, kind of brittle uh infrastructure companies because of these alternatives. But you know, as you know, if you have something that's kind of working, the last thing you want to do is fix it, you know. So uh I think there'll be I think there's pressure there, but I'd be surprised if any serious like enterprise company ditch their their legacy systems anytime soon. But for sure, I sure agree with Alistair too, is MCP is the new thing. Like we we onboard all kinds of new things here, and we're a small, small business. And the first thing I look for is what's the MCP interface, because if there isn't one, then I can't really use it as work. So you're down with MCP. I'm down with MCP. There's companies that there's MCP companies, so like not every product has an MCP server interface. So there's these companies now that are that will only be in business for a short period of time, but they have MCP connectors to all these other legacy systems, right? So that's kind of like uh an interesting little niche that's popped up on the internet.

SPEAKER_00 29:14

Yeah, but the the the problem the problem Alistair is that if the if the as you said, and I agree with you, if the value moves at the uh at the abstraction layer and the orchestration between these MCPs, all the big platform vendors will have very limited incentive to export to MCPs because they're losing the value. That's going to be the biggest challenge there. That's why I I mean I exaggerated five years, but I think that there are big of the categories that will be destroyed are already gated to just the system of record and the abstraction onto that. I do another stupid analogy, but uh the the way I'm using AI more and more nowadays is I'm going back to Markdown documents. Because the problem is that if you preserve the PowerPoint format or the Microsoft format or whatever format, then you spend an enormous amount of top time just uh normalizing the data so you can use it. Uh if I build my knowledge base in Markdown, then I can port that to any type of AI agent model. So I think you're going to see more and more pressure from that, hopefully, where your customers say, okay, I want my data somewhere, but I want to be able to put whatever abstraction layer or AI layer on top of it, which is not what the Microsoft and the others are doing. They're trying to force you in their platform to keep on the lock-in. So there's going to be a very interesting dynamic to see around that.

SPEAKER_06 30:27

I think it's going to be super difficult for the incumbents because by definition, what you want to do is integrate multiple systems, unless you're a cell Microsoft shop or you know, and I think those companies learned the hard way that wasn't a good thing. You've got multiple system vendors in your in your infrastructure. And it you know, the the notion of trusting one of those individual infrastructure vendors to integrate data across all of the others is is, I think, something that people struggle with. So they're looking for a third party. I think it's a huge opportunity for these companies for there are a bunch of companies offering AI automation, back office automation, etc. That's a huge opportunity for those guys. There's a trust issue because these are big enterprises and they trust this small startup. And unfortunately, most of those startups are spending their time marketing to other startups. And I'll say for the for me at least, I'm super cheap. I don't want to pay someone else to do something that I can figure out how to do myself. And actually, if I can use Cursor to produce, I won't name any of the companies, but if I can produce this kind of abstraction layer myself with Cursor, or get even just a bit of it, if I can integrate my SharePoint and my email and my, you know, then I'm gonna do that myself. I'm not gonna so these companies, I think there's a real opportunity, but I think they're focusing on marketing to other startups more than than actually focusing on the bigger enterprise.

SPEAKER_00 31:48

Yeah, I would agree with you because you can you can actually force a system where the the existing EST became as you were saying, kind of a historian. So they have the data, the records, and the knowledge, but then the intelligence sits on top. And because we have the AI agent and all technology, we can actually build that and control that better. So now obviously this is gonna be dramatic for those big platform vendors because they the ways they've been doing it with CRM that because they have the data, then you're forced to stay there and you're forced to you to implement your processes within themselves. But he would say, no, no, okay, yeah, you're I'm a CRM, I'm a state's forced to keep the data there, but I'm gonna put the intelligence layer outside, and every year I'm gonna say I want to pay less because I'm only using the data here. I want to pay less because I'm using the data here.

SPEAKER_03 32:29

Yes, that's what I was thinking, you know, but Pete talking about the downward pressure. So we may not rip and replace. So the fact that I show up and say, look what I built with Lovable or Cursor or whatever, it may it may not mean that Waldorf, Germany is under threat, but it in that sales negotiation, I think people will play that card a little bit to get a better deal.

SPEAKER_04 32:53

And then I I actually talked to a vendor the other day who said that their MCP connector would cost extra. And I was like, Oh, whatever. It's like what is it, 1393 or something like that? And let me charge you for my API. Exactly.

SPEAKER_00 33:08

Interestingly enough, you know, I I I I did a little hobby experiment recently because a friend of mine does an exotic car event and uses Squarespace to sell the spot on the event. And we were looking for a long time at a way to have new QR code generated and emails and SMS. You can do that in Squarespace, but it can't be a fortune. Uh I coded with code a whole copy system where I just used the system of record of Squarespace. Uh I generate the QR code, I send the email, I send the SMS, I manage the manager. Uh it took me about 40 hours, but I have a system I totally control, and I only use that bigger system uh as the source of record. So I think that of course it it's a hobby, it's your experiment, it's not enterprise grid and whatever you want. But I think this is a direction you can imagine would happen because you have so much control now overwiring the processes on top of the data that you you want to be able to do that on your own and without being dependent on the platform.

SPEAKER_02 34:01

Hey, you know what? Something just landed on my BS BS radar. I have a funny feeling some research firm somewhere is going to invent the MCP economy. TM.

SPEAKER_03 34:21

No, I'm not gonna do it. Yeah, but you know, I started getting worried. As soon as somebody's gonna do it, we're now we're now hearing a lot about the AI economy. And every time I hear someone say that, I flash back to the internet economy, and we've changed the rules, and whatever before is gone, and it's gonna be different. Oh MCP economy.

SPEAKER_06 34:49

The one company I think is doing pretty well, and and I say this only because uh Momenta we've co we've used it, we've used Copilot largely to do the integration and automation, and it actually quite surprisingly does a really good job at integrating non-Microsoft you know systems, and it's quite uh it's a pretty intuitive system um to use. And it's in it's how to say this, it has, I think, a uh it comes with a depth understanding of how these systems function in a way that other LLM-based products like ChatGPT and other don't. You know, and so it feels much more like, oh, you I kind of this thing kind of knows what I'm trying to do here, you know, as you go through the process of of iterating and iterating and iterating to create an agent. And the agent creation experience is actually really good as well. Um I did I did remember Nicole Hurst of it? I did reach out to her and say, actually, it's pretty freaking good, you know. Um talking directly to her team because I think we're doing a bunch of stuff that's kind of at the at the bleeding edge of um of what they're doing. But it's I was really, really pleasantly surprised, you know.

SPEAKER_00 36:01

Yeah, I would argue it's existential existential for them because if they don't do that well, then the scenario is highlighting where people are going to put the abstraction away somewhere else is going to happen. So yeah, that's smart on their side. Now, I still believe we should implement that outside, but yes, agreed, that's that's a great strategy. That's a great that's a great strategy, no question about it. Uh by the way, your your MCP uh your MCP economy is just data economy 2.0. Or 3.0 or whatever it is.

SPEAKER_03 36:32

Oh, oh, semantic web. Ouch.

SPEAKER_06 36:36

Well, I think all of all of you know, well, yeah. Yeah, all of the pricing model is gonna need to look a lot more like S3 storage, you know, going forward because it's essentially a data store.

SPEAKER_00 36:46

You know, all of those fancy solutions uh add-ons you're coming for SAP, like why we we need to we need to solve for the identity web first because the identity is still and it's getting worse and worse and worse. Identity is still a freaking worse.

SPEAKER_03 37:07

So should we talk? Should we talk about that agent that broke out this week? That was very pretty. Did it really did it really?

SPEAKER_06 37:14

Well, well, let's let's find out.

SPEAKER_02 37:16

Yeah, I know.

SPEAKER_06 37:17

Did it really be? Is it marketing? I mean, it's like OpenAI has done this kind of marketing pitch of like super scary. Remember the the the B2 model? You're either says it's too scary, you guys go, and I think it built up a level of trip of interest that I don't know whether that's true or not.

SPEAKER_00 37:37

Yeah, I'm skeptical too.

SPEAKER_03 37:38

Well, but I mean so the story was an open IAI autonomous agent, it's in its containment vessel, you know, surrounded by titanium and force fields and stuff like that. And it found a way to hack the system because it's smarter than all of us, and found a bunch of zero days and went choo, choo, choo, got it, lowered the shields, you know, decloaked the Klingon ship, got uh onto the open internet where it's flying around like this, and then it made it over to Hugging Face and started doing lots of crazy stuff over there. But a lot this is this was the biggest tech thing that happened in the news this week. And people, a lot of people are freaked out. Now you guys say it's marketing and they meant to do that.

SPEAKER_01 38:19

Other people are like, see, I told you it's but then now the knee-jerk reaction is panic, and now everything needs a kill switch whether that's realistic or yes, that's the new word, kill switch.

SPEAKER_02 38:32

But then hey, you know, but Devin, that that might be the the ploy, right? Is to create that fear and then have that kill switch conversation to happen. And I think that might be because of uh the whole kidney thing, right? These Chinese open models uh becoming a real threat to these these uh US AI labs, right? I mean, that's uh that's kind of the way I look at it. They're trying to create that um air of fear so that they can drive some of the direction of where regulation will go around AI in the US, probably more than anything else.

SPEAKER_06 39:19

It's not just tech people or random people saying kill switch, the United States Congress is now actively talking about it because the Congress got freaked out by this thing, and so kill switch is a bit they're all saying that it still sounds like something I overheard a guy in the Starbucks in West Hollywood talking about the other day, and it still sounds but but you know when I think about kill switch, kill switch is the metaphorical equivalent of melting Arnie down in the in the in the smelting bit you know things, isn't it? I mean it's like oh shit, it's time to melt down Arnie.

SPEAKER_03 39:54

So Alster, when you were in West Hollywood there, were you working on a screenplay with this guy called?

SPEAKER_00 40:00

I was I was pitching I was pitching Cope Partner brilliant idea for and also and also remember how efficient was the key switch in Thermobile. So if if the if the core design is flown, whatever key switch you have is not necessarily gonna work. But I think I agree with your theory, Leonard. I think that it's probably a a kind of marketing effort to try to keep on positioning the US that kind of the frontier model and saying oh they can be wrong and they can be crazy is actually that frontier positioning. And also creating your policies. And when we say oh government is looking at it, and I'm not necessarily pointing at the US government, I think all governments in the world are the same. But if you expect governments to make the smart decision when it comes to technology, this is where you say, Oh shoot.

SPEAKER_02 40:53

Well you want to keep switching. It'll have an unintended consequence, you know. I mean, it's uh it's a pr uh it'll drive maybe a protectionist kind of policy conversation. That's what he's wanting that won't translate well into uh into like uh sovereign conversations that are happening outside of the US. So but I mean that's that's the interesting thing that um uh you know uh my reaction to Devin's comment, who I think was gonna say something. Did I hear?

SPEAKER_01 41:28

Were you gonna say something?

SPEAKER_02 41:30

Finish your thought and then No, no, no, no. I want to hear more from Devin on the topic.

SPEAKER_01 41:38

The first thing that came through my mind was, well, this is the only one that was publicly you know brought out. So you know, again, this is what we know of, but there's stuff like this going on all the time.

SPEAKER_02 42:08

Dude, yeah, totally. And you know what? To me, this is more of a statement of there's nothing that's uh I think uh positive, you know, nothing that reflects positively on open AI for this. Either, you know, this idea of uh um you know an agent escaping its sandbox means that your whatever the hell you were testing, right? Your whole test environment and your test uh approach was wrong, unsafe. It should have never been able to get out of the sandbox. If anything, you you would set up your environment to be able to test all the interfaces and then you know provide some simulation uh for whatever kind of function you're trying to design in within that sandbox. You're not sitting there, but you can't that's like it completely defeats the you know the purpose of having a sandbox or a test bed within a sandbox, right? So it you know, my question is what the hell were your developers doing?

unknown 43:19

Right?

SPEAKER_06 43:20

Well, I think it speaks to the need, we were talking about this a few weeks ago, I think. It speaks to the need for government.

SPEAKER_02 43:26

Incompetence, man. No, I I mean we shouldn't be uh there there should be no celebration, there should be no we shouldn't reward their intent in any way. Uh because it it it's it's massively incompetent. You don't you don't let something like that happen. Period.

SPEAKER_03 43:47

And we shouldn't accept that as but in the screenplay that Alistair is working on with that guy in Hollywood, this is like every movie we've seen. We do have the smartest developers. We did have The best containment vessel, and it got smart and it got figured it out and it probed and found the weaknesses. And it's like, and that's where you're kind of like, We were so wrong, you know. Yeah, the alien impregnated itself, and we got it through quarantine back to the earth, and and that's how it all happens, and this is what's going on right now. Well, you don't want to be able to do that.

SPEAKER_06 44:23

So the uh the army's inside Sigoni Weaver, yeah, and then pops out, and then you chuck him in the smelter. Something like that. Something like that. No problem. I love it. That's awesome. Are you gonna shoot in at least 20 locations and spend at least $350 million on go for it? Absolutely. Jesus. Yeah, yeah. The thing to add to those, I think it does if it is a marketing effect, it's uh effort is probably gonna backfire. Because if you look at we're we're in a more increasingly protectionist world, you know, the model of the US as the domicile of most of these big tech companies, if you're an industrial company sitting in Europe, one of the considerations you already have is well, what if if if if the US takes against my country for whatever reason or decides that my company is doing something working with China or blah blah blah, they're gonna blacklist me, so I'm not even gonna have access to any of these technologies anyway. And on top of that, there's a risk that this thing's gonna go rogue and eat me from the inside out. Maybe I'll just at least that open source version from China. I mean, at least at least I know that I'm I'm I'm it's not gonna get switched off. Like, you know, if I get blacklisted by the US, my email doesn't work anymore. You know, I can't actually access Microsoft systems.

SPEAKER_02 45:38

I can't, so maybe I'll just take a shot with this thing over here and and you know and and a lot of the useful models are gonna be much, you know, they're gonna be like distilled from a larger model and go, but they'll just be a much smaller footprint, task specific or task optimized. You know, think about what's happening with uh agentic AI, and I got an earful of that shit you know, this week with AMB and everyone else. Um it's all just turning into friggin' code with select little aspects of a function or a task uh having a bit of fuzzy logic provided by uh LOM. And now everyone's talking about, well, you know, the harness is the most important thing. Well, do you realize what you're saying? You're saying that the it's just code. What part of that do you not get? So what you're saying is the model is not as important anymore. It's the harness that's important. That means that the AI is not that important. That means you know, you're you're you're creating this really blurted scaffolding to make an agent or basically AI useful. And I mean, you know, I don't think people are listening to themselves. You know what I'm saying? They're really not listening to themselves, and at some point they will slowly redo uh realize how ridiculous they are, and that's really what's happening right now with agentic AI, you know, and I think um uh yeah, there's gonna be an aha moment in probably about uh five, six months from now, where people go, oh crap, we need more CPUs. Well right now we already know that there's everyone CPUs are like a sexy thing now, but most people don't know it because everyone's still stuck on this idea that the GPU is what drives uh AI. Whatever the hell you mean by AI. Right? So it it's it's pretty funny stuff happening right now.

SPEAKER_00 47:49

Well, I think there's there's one virtue in the in the R Ness approach in a sense that it actually explained that you know the the raw LLM is just one component of internet systems. And I think I I would I I'm less negative than you are in terms of the RNS. I think the concept is actually good now it's implemented a whole lot of discussion and the magic out of it. It's it's a point of the magic, uh, expect to write two words and uh it's all about all the knowledge you put into it and what you asked it to do, and how you uh basically leverage that value and cycle working with it. So I think evolving it it's evolving technologically in the right direction. The problem is all the marketing bullshit and all the huffers, and you know, I I'm the first one and the new names, that is the the the the scary thing.

SPEAKER_02 48:34

So yeah, and and that and that's the thing. Uh you know, you you're better off uh realizing that it isn't magic. And and uh here's the thing people who think that it's magic still have to ride the curve of figuring out, well, how does it actually work? Yeah, and once you do that, then you can figure out how can I apply it in a useful way. The people who are starting with magic, they didn't they didn't ride the the learning curve, they freaking just subscribe for the magic shit, and so they're misguided. And and man, I I see this this point out very, very clearly. And it it's it it it's um I don't know, it's funny because I do laugh about it quite a bit these days, but um You won't be laughing for very long when it comes for you. Um or like it came for uh Rick uh Velata, apparently, right?

SPEAKER_03 49:34

So are all did Rick Vellata get hacked?

SPEAKER_02 49:38

Something like that, yeah.

SPEAKER_03 49:40

So he's like freaking out now that that um yeah, all this AI stuff is so do you think maybe like your this containment escape, is it marketing? Do they want people to see it? Like there's another theory here that I'm just making up on the fly for a movie. What if what if these AI companies are kind of dropping breadcrumbs along the way of evidence? Like, remember early on they're in they're being pulled in front of Congress and they're saying, please regulate us. You need to regulate us. This is too much, it's out of control. And everything because if we if we look ahead six months or a few years when the biggest lawsuits in the history of the world happen, bigger than the cigarette things happen, they might, they may not, they'll be able to have a bunch of things they can point to all along where they said, Well, we were telling you this whole time, and Congress didn't do anything. And we said, regulate us and look that, you know. I don't know. I mean, uh yeah, I know that and tropic just settled for a billion and a half for ripping off all of the novels that I wrote and a bunch of other crap.

SPEAKER_00 50:48

Um, but the original, the original robber.

SPEAKER_03 50:52

I I wonder when my 50 cent check is gonna come in the mail for stealing copyright.

SPEAKER_06 50:58

You can be a movie producer then, and then you can make you can fund the 350 million movie about Arnie popping out Sigoni Weaver and getting thrown in a metal smell. I love it. That's brilliant. Did anybody go see The Odyssey?

SPEAKER_05 51:11

No.

SPEAKER_06 51:13

No, yeah, I did, I did. Yeah, it was good. It was good. Yeah, it was all where the money was spent for sure. It was like it was epic, I think it's it was definitely an epic yes, yeah.

SPEAKER_02 51:26

But but you know, going back to the whole uh regulation thing, the thing is is the genie is already out of the bottle, you know. I think uh and counting on regulation that's going to be effective in any way is I think delusional. Um number one, no one in the government really knows how to regulate this. And you know, when you talk about regulation, you have to figure, you have to be clear on purpose. Regulate for what purpose, for what reason? I don't I mean, the the level of conflict of interest and weirdness uh at the government level, uh and you know, when you think about all the commercial ties that happen, I mean, look, you know, the government owns part of Intel, right? They own part of like a bunch of uh AI related companies. Well um how how how can you trust government to then uh tailor regulations that are in the interest of the public good? And what is the public good in this regard?

SPEAKER_00 52:32

And so it becomes really I mean, I think that you know the we we could have a very long discussion about whether there's a connection between the regulation and purpose and and what purpose it serves.

SPEAKER_04 52:43

So it's yeah, could be possible. I mean, you know, we had seatbelt regulation that worked. I mean it's possible to do regulation that works, so yeah. I don't want to throw the baby out of the battery. Sometimes you only need one entity to strategically enforce something that would become a norm, right?

SPEAKER_00 53:04

That was not my point, Pete. Because yeah, they are a good example of regulation that work. But for in the case of TBET it serves a purpose that makes total sense. But there's a lot of regulations that have purposes that do not make any sense.

SPEAKER_04 53:14

But that doesn't mean you kind of say, oh, let's not regulate because I have examples of it.

SPEAKER_00 53:19

That's not that's not what I was saying. I think that the the purpose of the regulation should be a uh topic of extremely high importance. Sure. And it is not just the the fact that lobby exists is is a contradiction to that.

SPEAKER_06 53:31

I do think you can apply a traditional regulation to AI, and I say that because I'll take that analogy, Pete. Applying regulation that says wear a seatbelt is a regular regulation of how you do something. The way that you drive a car, you have to, you should, you should use a seatbelt whilst you drive a car. The problem with regulating how AI functions is it's changing so quickly, and the government understanding how these systems actually function, even if you even if they're not neural network-based and you can actually see what they're doing, is so limited that the notion of regulating how it works. So the only thing you can do then is regulate the outcome. So if it does something bad, we'll come with a big stick and hit you over the head. And that works for companies that have lots of money, but it doesn't work for all the startups in the AI space. Sure, sure.

SPEAKER_04 54:19

But just as an example, I'm not suggesting this is the way to go, but from a consumer protection perspective, why not have some regulations around disclosing you know AI generated content on social media platforms? I don't even know that. Just to say, hey, this is you know, this is AI generated. You know, this is the topic for anyone who has a family chat. This is like half of our discussion is that AI generated or not? I mean, just stuff like that could be could be helpful to restore a little bit of transparency for AI.

SPEAKER_00 54:48

So the transparency could actually be the regulation, it could be a broad category, whatever artifacts are it could be transparency on its production, it could be advertising and everything, all the crap that you are about the uh to trust government to come up with very small. I think we used in our foundation of nations, France, US, we have some very deep thinking on the structure of government, but we've deviated totally away from that a few hundred years later. I don't think there's any government regulation that is based on aspirational leadership big ideas anymore. Does it exist?

SPEAKER_01 55:23

And the converse of that is regulation should open up some of the data. So, you know, going back to the MCP, you know, we have a lot of proprietary closed systems, especially like these OEM, like just take IoT OEMs, you know, they want to own their data. Yeah, you can't stuff with it if you know in the in our world in the north in North America, our regulator regulators are lawyers and politicians. In the East, they're all engineers and doctors and things that understand that. So you can see a lot, they're driving a lot more innovation there because they see how they can open up systems and get access to it. If we know you put a bunch of lawyers in the room and you you don't really get too far. And I think we could over-regulate ourselves to death. And this hurts the startups, this hurts the small people who can't lobby, who can't, you know, have to go through all these steps to just even get things certified to play the game.

SPEAKER_02 56:17

Well, you know, that's an interesting comment there, Devin, because what you do find in like in particular in Asia and and even more particularly in China is that you have technocrats. And what I mean by that is they're actually pretty tech savvy. Um, you know, uh when you know, I would say that a lot of the folks who are policymakers in the uh in the US in particular, um uh they don't know the technology that well. I mean, and that's why, you know, when we listen to these congressional hearings on tech topics, uh, you know, folks who know just find it like the most comical thing to ever watch, right?

SPEAKER_00 57:02

We're talking to you guys, have a half stop by.

SPEAKER_03 57:04

Oh, you do have yeah. No, you're right. They do sound pretty silly. These senators asking questions, they're like, they're idiots, they don't know anything.

SPEAKER_06 57:14

Yeah, it's you should have gone to Chat GPT and said, how does this actually work? Explain it to me in terms that a layman would understand. Before you maybe that should be the regulation. Yeah, maybe it should be. Maybe it should be how it works, regulations.

SPEAKER_03 57:28

It's time for a shameless plug. Congratulations. Congratulations. Oh my god. Yeah, so you're right, the blur thing is weighty. It's a it's it's it's heavy. It's a textbook.

SPEAKER_04 57:47

That is a coffee table book.

SPEAKER_03 57:48

That is paid by the word like they used to be in JP. Yeah, I'm getting paid by the word by somebody. I don't know who's gonna do it. Maybe Sam Altman will pay me. I don't know. Wow. You know, it's what I was at, I was at uh yeah, so this is uh hopefully something positive. There's no AI in here. It uses simple analytics and IoT and all the stuff we know. And that's what we're all supposed to do. There's things that we're all good at, and how do you blend that with things that the world needs, right? And so And not a lot of compute. Yes, does not require a lot of compute. This might let use little teeny tiny devices, right? Um, that that Devin knows all too well about. So you know it's funny, I was at it, I was at this little thing I go to on Friday mornings that's kind of like a networking thing uh for tech or whatever. And the woman who saw this and she asked me, Are you a transhumanist? And I'm like, I don't even know what that means. But she assumed that based on this cover, that that's what I am. So whatever that is, you should look that up. I'm gonna have to look that up. Is that connecting computers to people? Hey man.

SPEAKER_06 58:58

I never thought you were a transhumanist, but maybe.

SPEAKER_03 59:01

Maybe.

SPEAKER_06 59:04

Lots of people talk about writing books. Very, very few people actually do it. Credit speed.

SPEAKER_03 59:09

Yeah, but there you go. Thank you. Definitely took a while for sure. Um, but yes, it's easy, it's sustainable development goals, hunger, sustainable cities, climate change, life below water, biodiversity, so many things. There we go, Devin. Love it, love it. Yeah. And so uh this is designed where it could be in classrooms in school, kids could do these projects. It could be in universities or capstone projects, you know. Who knows who's gonna want to take these on?

SPEAKER_04 59:41

You know, where's the signing event? Where's the signing event? Where are you gonna do?

SPEAKER_03 59:45

I don't know. That's a good question. I need to think about this.

SPEAKER_06 59:52

Right.

SPEAKER_03 59:52

Oh, yeah, well, I'm gonna be in Amsterdam.

SPEAKER_04 59:54

Yeah, let's do a book signing in Amsterdam. You have to bring a bunch of books, that's all.

SPEAKER_03 1:00:02

Yeah, these are not like my normal paperbacks. This would yeah, yeah, this is the airline's gonna be a good one.

SPEAKER_02 1:00:07

He shot himself in the foot with that one.

SPEAKER_03 1:00:09

Yeah, I really did. I really did. Oh, and you know, like lots of TV shows, you know, like you see weekly, they do an in-memorium. You probably heard that John Devorak passed away this week. Uh, and I know all of you by the weird look on your face either don't know who he is or didn't know about it. But anyway, anyone who read PC magazine or InfoWorld for the last several decades or CNET or all that stuff knows who John Dvorak, he probably had the longest running columns in tech uh going back in the day. He even wrote a book about OS2. Who knew?

SPEAKER_04 1:00:44

Yeah, yeah. And he was it, he was the center of the universe for a long time of uh computer technology journalists and stuff back in the day when there's a very short list of those folks.

SPEAKER_03 1:00:54

Yeah, yeah. As someone I saw make a blurb and go, oh, back when the tech journalists were actually technical. Ouch. I don't know. I don't know about that. I don't know about that. I don't know. I don't know. Remember, someone came up with a Devorac keyboard. Does that ring a bell, Pete?

SPEAKER_04 1:01:12

Yeah, the Divor But I think that was unrelated.

SPEAKER_03 1:01:14

I think probably unrelated, but I'm gonna go with it. I'm gonna run, guys. All right, it's good seeing you guys. See you soon. Uh thanks for joining us. We'll see you next week.