ERP127 - Navigating the AI hype cycle

Michael, Mark, welcome to the Evolved Radio podcast. Thanks for having us. Thanks Todd. So glad to have you here. You know, there's been a lot of talk in the industry lately about everyone calling for the doom cycle of AI. There's a lot of conflicting opinions about whether or not this is working. So we wanted to kind of get into some of the details of like why that is potentially a debate. And Michael, maybe start with you. You have sort of an interesting perspective on this. Some of the things that you saw in your travels in the medical realm where you were before joining thread, do you want to give us kind of a bit of your backstory and how AI is related? Sure. To your point that where I've been in the past, I've spent the last 10 years or so inside the digital health space which has its own set of challenges outside of the AI piece. But yeah, for the past prior five years before joining thread, I was working at a, in a healthcare organization that was building agents on a kind of a more classic rules based structure versus what is really more commonplace today based on the large language models and etc. But yeah, look, setting aside kind of the unique challenges of operating in the health care space and kind of the different set of rigidity, our clients today in the space that Mark and I operate on have a set of standards and requirements that we need to make. They're just kind of uniquely different in healthcare. But I think to your point about at the beginning where you mentioned this doomsday and is AI really delivering on kind of the value prop? I think it really depends on how you frame the question of that. I think where it may be failing or perceived to be failing is where the promises were ahead of its ability to begin with. And the shiny promise is really easy to sell and that's easy to sell in almost any industry regardless of being healthcare or in more of a core technology space, a place where there's no doubt it's creating value is some of the back end systems, some of the analysis side of things. Where I think some of the challenges have come in both in healthcare and outside of healthcare, is where the rubber meets the road on real specificity of problem solving, solution solving, having the AI or in our case an agent being able to take an action that brings to resolution. And I think if you're able to frame the ability of the product and set the right promise rather than this, you know, North Star. Always looking at this North Star type situation, I think it's in a lot of cases been more successful than maybe we Give it credit or some of the recent articles. And also, no matter how you look at the speed at which it's, at which it's moving, I mean, Mark can certainly speak to, you know, what thread has been able to develop in other agentic companies in a much shorter period of time with a much higher degree of fidelity, a much higher degree of solutioning for the end user than we would have seen five years ago. So kind of to end that rant. Yeah, I think the promises have been too huge and kind of inset the wrong expectation. But the underlying, I think if you, you kind of said before we got on here, you have a little bit different view. And I think I'm with you. If the expectation is set correctly, it is. Has been incredible. The advances and the ability to solve day to day work problems are. It's quite remarkable. Yeah, it's, you know, the, the Gartner hype cycle sort of comes to mind here. We had chatted about this before. We were talking about coming on and like, I don't know that we're necessarily in the trough of disillusionment yet. Right. I think it's maybe on the downward trend. Like I think to your point, there was a lot of sort of unrealistic, unrealistic expectations of what, what these things were capable of. But at the same time, the part that I marry up with that is like how much we've blown past even the most basic expectations of what an AI should be capable of. Like I heard this conversation on a podcast and they were talking about like, hey, remember the Turing Test? Like no one, like that was hot for about five seconds as we ripped past it and no one seemed to really notice. Right. So like, like be in a way, like we're actually beyond, I think what the peak of inflated expectation would have been like. We shot so far past it that the, the, the drop to the trough of disillusionment is sort of like almost passing through the workable cycles, the workable workflows. But we're in this sort of weird, messy middle of. We expect. Maybe it should be capable of more because it's not AGI. I don't know. Like, Mark, what's your take on this? We had a great AI service unleashed weekly meeting today and I love your point about Turing Test because we've, we've now progressed. We are now at the Will Smith eating spaghetti test and it's insane. And we've all been like, oh, okay, it's okay, it's conscious and like spaghetti doesn't crunch and there's steam coming off the pasta. So we've, we've definitely blown past that significantly. What. And I think about this a lot and there's like two big concepts I want to convey. This is not the first AI cycle in human history. This is the second AI cycle. We have had a nuclear winter for AI after the first big realization, right. I forget the guy's name, but the first learning of neural networks came from a biologist that was also mathematician and he figured out the neural network and we made a massive leap and then it just went quiet. Right. And you know, to Michael's point, like we had to do a lot of rule based conversation flows and it went quiet and then the transformer model from Google came out. Hilarious that they did not monetize or productize that and OpenAI did. So I think we're on the cusp of another big, big uphill and just like this explosive growth with gen AI now there's a lot here. Humans always overspend. If anyone was around during.com bubble, man, did we overspend for pets.com to do something that was probably not worth it, right? But what, what was left behind that was infrastructure. What happened was we went from a thousand dollar T1 lines to 50 DSL lines and that infrastructure is why Google exists, is why Amazon exists. So we are definitely overspending. The question is what's going to be the infrastructure that's left behind for us to build upon? Yeah, no, I think that's a really good prediction that like, you know, maybe we're kind of misunderstanding what the point is right now. Right. And this kind of goes to like, I, I kind of bring this up every time we're, we're talking about AI on the podcast, but you know, I won't do the full detail of this, but my, my AI sort of prediction timeline is now three or four years into this. And I predicted kind of three to five years was usable workflows and we're well into that. Like there's a lot of very useful creations, you know, thread certainly being one of them is a very reusable work workflow use case. And it provides value. Like those things absolutely exist and it moves on to sort of later on it becomes something that starts to potentially eat service desks after five to 10 years. And I think there's a lot of things that you guys are cooking up that lend to that vision. And it's not a doom and gloom type thing of like everyone's going to lose their jobs because AI is taking over. It Definitely pushes people to higher value work, right? It's commoditization of rote work within those ecosystems. So if that's the case, like that's part of what I'm curious about here and I have, I have some, some opinions I can definitely lend to this. But I'm sure your guys's perspective on, you know, like the HBR study that came out that said 81% of AI projects in business are considered a failure, right? Like, like why, if we're sort of seeing the early stages of this golden age of AI, why is everyone predicting that, you know, we've overstated this and you know, now everyone's like well crap, maybe we shouldn't have fired all those people at sales and Microsoft because now we have to hire 80% of them back, right? Like what is the tension between those, like is it, is it oversold or you know, is, are we sort of missing the mark on what we're actually supposed to be achieving right now? I think it's probably a little bit of both and we have to be careful to not be a little bit of the self fulfilling in that. Right? If you're a big major company, leave them nameless. But any of them, or any company, large or small, who you make a commitment and you go out and you get ahead of your understanding of the practical application of the technology in this case, you know, agentic that we're talking about, then there's this pressure to live up to. Oh, you said it was going to replace, pick a department. You said it was going to replace your legal department. Well, you know, if you're a public company, you better show some evidence that that occurred because you've suggested it happened rather than, because I think because of the, some of the stuff we talked about, the promise is so huge and the reality is some of the stuff is so amazing that it's really easy to take a look and say look, it fundamentally can do this. Well, we can jump all the way to there. It's like, it's like almost, it's, it's almost human nature to kind of jump to the end of the, right to the end of the rainbow or, or whatever it is rather than understanding that it's not understanding, you know, being methodical that everything is a step process, right? You've got to, your legal team's got to know how to use it and where the value creation is before you could not have one or, or to the point you guys made, you said earlier, how do you up level skill set, how do you get These people in any department doing higher value creation things rather than kind of time, workforce burning less optimal things. And I love that. Can I, can I add on? I don't know if you. Man, there's two big things that are happening economically. We went crazy on cryptocurrency. We went through Covid. We've inflated the market. We needed a way to generate more value. I don't know what would have happened if we didn't get the gen inflection point. So I think big business needed a soft landing. Right. Or they needed a way of, of progressing forward. And I think, I think we've saved all the crypto farms because I'm pretty sure all the GPUs are now used for AI. But also there's a lot of inflation in the economy. There's a lot. And so we're trying to figure out unemployment is low, salaries are high. So what do we do? How do we, how do we fix the labor market? And I think there's this big demand for us right now to figure out the digital work because if we can do will heal our economy, at least in the United States, it will heal our economy. So I think with Harvard Business Review, they're obviously going after the large companies. So I think that's why we're seeing that pressure there. Yeah, I think, look, some of the challenges right now, are you looking at a very high level. There's a couple dozen companies driving this perceived. If you look at the market, this market explos. Whereas the reality for those major companies that have their point, that have their AI plan mapped out and they're building the data centers and they're building the hardware and they're building the processing power. Those are very clear. The vast majority of the companies in the US around the world, and certainly our customers, they're figuring out how to deploy it really effectively. And everyone uses it and touches it on their computer, they do searches with it, they do whatever. But from practical workflow utilization. What do you mean? We're taking my wedding photos and making them anime is not productive to the economy. Yeah, it's my favorite activity. We're still like the average company. The business model isn't as clear as an Nvidia or somebody who's building the data centers. In addition to that, we're still settling in. Mark mentioned the broader economy. Like what is it doing? Right. As these data centers absorb the electricity and power, there'll be other investments in new sources of power at the same time. Until those are there, we all might see our Electric bills go up. Right? Because supply and demand, right? So if there's that, or terraforming the. Planet to be a compositional brain. It's wild. It's insane. I do have one thing I really like. I need to get off my chest. Let him do it. There's a big difference between a proof of concept and production rate. There has not been that big of a difference in traditional software development. Like, sure, you haven't thought about all the workflows. You haven't thought about all the use cases, obviously. However, this is the first time we're talking about indeterministic code. This is code that's making its decisions. And it's a completely different game. It might look great, but every time when we first started to QA our voice agent, every time we'd pick up, what would we say? I have a printer problem. I have a printer problem. I have a printer. I was like, this is awesome. It's so good. But I never thought maybe it's just good at printer problems. Because when you get into production and it's not a printer problem, it's a lack of connectivity, it's a broken dll. It just acts completely differently. And so I think the combination of the, the. The pressure economically to find the next way to heal, but also seeing those little POCs, they're like, yeah, I'm Clara, baby. I'm gonna fire 700 people. And it's not going so well. Maybe we should reverse out of that. Because people actually have hard, hard questions. And it's not all about, like, I need a refund. Yeah. And this, this sort of gets to, like, my feeling. I have two things that I think are sort of at the core of. Of those. Those numbers of the percentages of failed AI projects in business. And one, I think is. Is sort of the majority of it is the historical percentage of failed IT projects in general is. Is catastrophically high. Right. So, like, sure to say 81% fail. But if we're starting from a base baseline of 64% of IT projects fail, which is. Is a general stat, then we're really not that far off the mark. So I don't feel like this is as. As sort of as diminishing to AI as a project as a whole. In business, it's more to the point of AI, of IT is messy, and the implementation often doesn't go the way that we think because of poor project management, because of scope creep, because of people issues. And I think where that gets married. Up with it's always people issues. Project Management is as simple as who does what by when. And then you add people and things get complicated, right? So like I think the other core of this is that a lot of the AI, like well, I'm using air quotes here for the audio listeners. Like AI deployments and projects were largely just sort of, hey, we turned on Copilot for you guys. Go forth and prosper, right? And it was, or you know, we subscribed to Gemini. You know, you guys, you guys now have access to these tools and then you go back and you check with people who have not had any level of training and have no concept of how to utilize these things in their day to day operations. Like, hey, is that, that, how's that been? It's been amazing, right? Like it's totally revolutionized your work. And they're like, I got nothing. Like I didn't find it that useful. They're like, failed project, I guess. AI didn't save our business, right? So I think those are the things that are largely at work. And I've been even kind of amazed. You know, I've been working on some, some AI projects and as a part of this I've been interviewing a lot of MSPs and technologists and saying if you had an AI army to build some workflows for you to automate some things, to insert some agentic capabilities in your day to day operations to save you money and time, what would you do? And the level of sort of sophistication and creativity in those responses from highly technical people that are very aware of what their business is have been very routine. It's like really like we need to scratch a little deeper and think about some things that like that we could really sort of leverage and blow the doors off of the capabilities and allow people to do more. So if you assume that the IT people that are immersed in this all day and have a pretty good sense of what this stuff does, they can't come up with great use cases for how to deploy it and how to you how to leverage it for workflows. How are you going to expect that like Sarah from accounting and Jim in the marketing group are any further ahead in figuring out like exactly what they need from a workflow standpoint. And this is where I think like the core of this really comes down to is a lot of this kind of needs to be programmatic and determined because just to say to people you have AI, it will be amazing for your work ends up as a dead end, right? So I'll sort of end my rant there and leave you guys to sort of jump back in on this. What do you think, Mark? I haven't earned. We haven't earned secret here. The reason for our success is because we pair workflows with AI. It's not about telling people to use AI when they need it. You need to put it into the workflow. Right. The difference between telling accounts payable person, hey, I got you a Claude subscription. It's awesome. And telling them that every invoice that comes in will process added to the database. And I'll tell you at the end of the day what is the breakdown of invoices from vendors. Let me help you prioritize. So the key to success here is a workflow that's deterministic. Like this is a real use case. I need to figure like msps, man. Reconciling vendor invoices to billing to your customers is insane. Yep. And of course you can give them Claude, but you need to ingrain yourself in that workflow. You need to get every invoice and automatically process it and read the invoice and break it down. Like that's been, that's been so instrumental and I think that's what a lot of people aren't doing. You're right. It's so, it's hilarious. Co pilot. Great, Claude. Awesome. But it's not part of a workflow. And what Mark's talking about, when it really becomes just peripheral through society. Right. To where it's just, it's when somebody's used it all day and they don't realize they've used it because it's just culturally part where it is not. I'm taking a specific action that will engage this AI thing. Somebody said to me from my prior career, healthcare moves in decades, not in quarters or not in years. In a lot of ways. Everybody we talk to is still early adopter. Right. I mean to the average 330 plus some a million Americans out there, our. Industry is an early adopter. Yeah. And the subset of people. Yeah. And if you stick to the decade is what creates change. I mean I think just about my kids, they're adults now and simple things like they want to know why I'd call instead of just text them. I mean it was only 15 years that that wouldn't have been an option. So maybe not quite the decade. Mark's kids will not know a world where AI isn't influencing almost every piece of technology they touch. They won't. For good and for bad both. But again, kind of to my early where I started that it's about kind of to where Mark was going in that it's embedding the solution using AI solution in your day to day workflow. So it's just part of what you're doing. And that's when those big promises that maybe that article was suggesting the 81% aren't achieving. That's when we start to. We see the reverse of that and it's 80% of them are achieving and maybe 20% or not. It's also last thing I'll say on that. If we go really long term and we talk about Mark, you were talking about the way people think about planning a workflow or planning a solution. His kids at their age, they won't know that that's just how they will think. So everyone knows my kids are four and five. Oh. So yes. Minor adults. Yeah, but right. You're absolutely right. You're absolutely right. My kids like okay, you're tired of telling me stories. Can you have the AI tell me a story about Sonic? I'm like. That'S crazy. Like how did you come to that conclusion? You know, because it just. But I would say like this is really important. Like this is really important. It's not just AI, you need to get into workflow. If you don't get it into workflow, you will fail. It needs to be part of a workflow. So let's build on that because like, like, I mean I understand what I. Well maybe I'll preface that. I think I understand what you mean. Right. Of like, like we actually map out a full process. We understand what instigates a certain action or some. Something, something. An activity and then it leads through a bunch of actions and creates in some type of endpoint. Right. Like is that sort of all we're thinking about in a workflow? And understanding that enough is being more detailed and understanding what are the actions we take on initiation of something towards an end goal. Is that kind of what. What you're thinking about when you're talking about workflows? Yes. Very high level. Right? Yeah. Business process mining. Yes. Yeah. Right. Like there's a trigger. It's typically, you know, someone reaching out. There's a communication layer. But I think what the most important part that people aren't realizing is that if you map out, if you just forget AI if you just map out the whole business process, it starts to branch heavily towards the end. Right. Always it becomes wild. Oh, this customer is on this plan. Oh, this customer. We promised them something. We always promise. Right. Like everyone, everyone promises a customer something. And so what I have found is that the first two, three, four steps of a business process are typically linear, but then what happens is you hire a middle manager to be that, that branching point, like, oh, this needs to go here, or oh, we need to do this. What we're finding is you can, you can automate 1, 2, 3, 4 and 5 is where you put your AI. 5 is where you say, oh, we promised this to a customer. Do not bill them for this. 5 is where you say, this is a security issue. Get it over to the security team as soon as humanly possible. And I love William. I will give him a shout out forever. William from Integratech, he said a great line. AI is here. And we want to use AI to make service more human. It's not about responding back to the other person that's triggered the workflow, but it's about getting it to the right person in the right context with the right information. So like, that's been like our, our pillar. So. And this kind of goes to like, I think the way that maybe people are misusing the technology in a lot of ways. Like, I see this as very similar to sort of the other tools that get used and how people migrate away from them. We talked about sort of the generational changes of how people utilize technology. Right. And you know, I use Perplexity all day and I barely use Google anymore, and that's a fairly recent transition for me. But I know a lot of people, you know, their, their, their default is to, to still Google stuff. You know, I have these conversations with, with my wife and instead of her doing. I had the same conversation with my wife, by the way. It's hilarious. Yeah, like she'll Google stuff and then ask me to ask the AI. It's like, well, you know, you can do this too, right? Like, so I think there's a lot of that of people are somewhat comfortable with the way that they operate, the way that they deploy these technologies, the way that they think about the activities that they need. And honestly, one of the biggest shifts that I've had with AI was, you know, for years I was still using it like a search engine. Like I was sort of one or three shot prompting everything. And then a person made a very sort of simple suggestion to me. They're like, treat it like a co worker. And that was a huge unlock for me. And just the way that I interacted with AI and made it a lot more capable because the expectation was just slightly different about how I engaged with it and as I sort of explained this and we've had this conversation, I think I'm more, maybe more convinced, like, you guys opinion on this. Like, maybe we are on the slope of enlightenment and the Gartner hype cycle because now we have a better understanding of how to interact with and how to prescribe the technology's use, right? To just sort of say, here's a tool and you know you're going to do amazing work, right? Like, if you think of it in a very sort of, sort of pedestrian way of like, you know, here's a, here's a wrench, right, To a person who needs a hammer. And they're like, great, I don't really understand what this does for me. So I'm just going to go back and do what I was doing before. And like, they're like, he doesn't understand, doesn't understand how to use this tool, right? Like, this has been a total failure. Why wasn't this, you know, creating massive outcomes the way that we saw it? I think there's a lot of similarity to that. So I think your point about like, the definition of understanding how the work actually flows through the organization and where are the intersections where we can actually start to insert some of these agentic actions and utilize AI more as a co worker or a team member than just sort of a tool unto itself. Any thoughts on that? Yes. I have thoughts. Michael, you want to go first or should I? You go, I'll do cleanup on this one. Yeah, that's absolutely right. And I think, you know, we just had our keynote and we get this question a lot because Thread is becoming a little notorious for automating, dispatching work. We are for better or for worse. And folks are asking us like, well, are you scared to talk to MSPs? Like, aren't their dispatchers scared to get Thread? And we had this deep reflection where we realized where AI is today. We're not replacing 100% of any job, of any role. What we're seeing is that we are moving to a supervisor relationship with the AI. It is a coworker, but it's very important to understand it's a supervisor relationship. And when I found out that Waymo was doing, it's all, it's, you know, autonomous driving, but behind all those cars, there's a person with an Xbox controller looking at nine cars. And if any of them get stuck, they use an Xbox controller to get out. Now, is it better than hiring nine drivers? Absolutely, absolutely. So I think the role, what's, what's Happening. And again, like I said, like, we'll see how many, how many more inflection points we have. But the relationship is a co working relationship, but it's a supervisory relationship. We're not going to get rid of all the dispatchers. We're going to make the dispatchers that are the best, the supervisors. We're not going to get rid of the technicians. We're going to make the best technicians supervise their little army, their digital workforce. Because you're right, they can reason, they can talk, they can't innovate. And the world has entropy. So how do we handle entropy? Humans are great at entropy. So I believe it's really important for us to recognize this. I believe that jobs will be automated, like jobs to be done, like prioritizing, categorizing or driving someone a mile. But I don't think roles will be fully destroyed. I think roles are going to become supervisory roles. I like that vision. I think, look to your point about where are we on the curve? If we started that curve, we're, we're really early because even the really early adopters still, I'll still get on an engineering stand up with Mark and the team and there'll be a new release from one of the large models. And we're still saying that group who live and breathe it. Holy crap. Look at, look at what this one does. You know, just the release three months ago didn't do X in 18 months. Before that, you know, seemed like an, you know, it's the difference between an infant to a teenager and kind of its, its capabilities. And that is, that's people like Mark's engineering and product team and the company's team living and breathing it. Whereas look, I, I think about, you know, the masses of people think about, just go through my family. I've got a lawyer and my brother's a school teacher and one's a lawyer and one's a works in retail. And like they are not, they are. Not medical industry radiologists. Yeah. They're just not what a radiologist organization standpoint. Like they're not, they don't, they don't yet have the opportunity to. And if they do, it's it's out of human curiosity. Right. You know what I mean? And to your point about. Yeah. Tinkering and to point about, about specific workflows like my brother who has been a schoolteacher for years, you know, we all know about the challenges of teachers. Like did the paper get written? Did they write themselves? And you know, so there's a whole other set of challenges. So certainly he is. He is teaching himself and educating himself on those practical applications, on how it's impacting his environment and being sure that like, you know, kids still have license to learn and that kind of thing. But again, it's. It's fairly isolated to the more not the advanced stuff. So to your point of the book, I Super early I think in that. In that curve maybe not as early for the marks of the world even for me. Todd, I want to flip it back on you. I have a question for you. I have. I want to know your thoughts on this, if that's okay. Sure. You know, Jensen is obviously the CEO of Nvidia is. Yeah. Fixing shovels. Win any gold rush. Yep. And he said a quote and I was like, yes, we are not forgotten. The IT department of every company is going to be the HR department of AI agents. Ooh, I like that. Yeah. TNHR have been new user onboarding termination. I. I just. I would love to hear your thoughts about that. Yeah. In Jensen's a brilliant dude. There's a lot I disagree with about Jensen's management style, which is a separate podcast seems to have worked for him. But holy hell, the. That image I think is really smart because if I pull that back to something that's a bit more familiar right now, and that being the extrapolation from IT is outsourcing has become a lot more practical for IT organizations post Covid because it just removed all the geographic barriers on where people work from. And one of the things that I insisted very early on when I encouraged people to consider outsourcing was that they said, oh, I've tried that. It doesn't work. I'm like, yeah, but you probably abdicated a lot of the responsibility. And you know, hey, we hired this person. They should work remotely. They're good at what they do. And you know, none of that happened. But you never spent any time with them. You didn't treat them like a staff member. So of course the results were different. So the extrapolation of that I think is now that we'll have an AI workforce, we have to understand the. The workflow analysis and kind of figuring out like what the business patterns are and therefore how we can insert AI into it. So I think there's a very natural marriage there of like the business consulting aspect of it is becoming more prevalent because the, you know, the just the break fix and the support requirements are lower. And you know, once the AI start to live inside the. The business's analysis gets complete and figure out sort of where the agentic capabilities are and you know, building on those workflows, there still needs someone to kind of manage and operate those like from that supervisory position. Right. To hire them. Yeah, exactly. Like we, we understand sort of like which agent is particularly good at this type of role. And you know, we manage some of the exceptions based on sort of a decision tree because stuff happens, right. Work and people are messy and you know, how do we step in when, when someone requires a bit of intervention on that when it's dealing with, you know, an AI team member. So I, I think that's, I think that's a cool vision. I really like that as an, as an idea. I couldn't agree more. Right. And there's so much opportunity for us. This is why there's so much excitement in the MSP industry. Like we are going to be the HR for, for agents. And I, I just want to, you know, Todd, please, let's keep going. But I just love the relationship between IT and hr. Who's working and what tools do they have to get leverage in their work. Where the tools where the fire. Currently. I agree. Yeah. But look what's happening in the industry. Rippling deal just works. They're all starting to launch little like device management, onboarding and off board. I think that's going to be apart from the AI curve. I think that the marriage between IT and HR is going to be an interesting one. So I mean, obviously I could talk with you guys for hours on this. I want to look. To sort of close it up. I'll sort of drop one more here. That is really outside of sort of the basics of what we're talking about here. But Mark, I'm sure you've put some thought to this because I know that you live, eat and breathe this stuff. What are your thoughts on RL AI like, especially as we start to embody AI into RL real life. Like, like robots basically being human AI and starting to interact with, with the real world. You know, I, I've been thinking a lot about this recently. You have any, any thoughts on, on sort of the, the putting AI out into the world and having it interact in that fashion? We've, we've run out of training data. There's a big problem. That's why I'm saying the last AI winter perhaps may not be the only one we've run out of training data because the LLMs that we have today have learned everything they possibly can from all of our Reddit posts and all the junk we've put online. It's petabytes. It's great, but we're running out of data. I think there's two angles here. I think that AI needs eyes. It needs to, it needs to. You know, our eyesight consumes more information per second than any other of our senses. So I think there's, there will be an effort to get more information. So I think it's a natural progression of, of giving it senses in a certain sense. And I think that's really interesting. Now what I'm really excited about is I think, manufacturing. And I'll go back like after we had a disaster in New York, you know, the MSP I worked for, we bought the Sprint Data center in Purchase, New York, and we ran a data center and there's like predictive hardware failures and you just need to swap out a hard drive from a SATA drive. It's like we could figure it out. We just need someone to do it. So I, I think that on the manufacturing side and on data center side, I think we're going to see some of those, some of those things automated that are not as important as, for example, figuring out the energy problem or figuring out how to get smaller in our silicon manufacturing or get to quantum computing. But I, I think there's a desire to feed more information into the AI because we've run out. We have a big problem. We've run out. We're now using synthetic data, but also to help on that manufacturing side. Michael, what do you think? I'm gonna go more human, philosophical. That's what we work together, Michael. Perfect. I think if those robots, humanoids, real life, play a role in where Mark went, where the agent, where it's inserting and improving, obviously huge value there. I think on a human level, I think we learned a little bit from COVID in those years. We as people get a lot from interaction and being around each other. So if it over a decade or more would create an environment where we're doing less interaction with each other, I think that's not great for us just as people and as relationship. I've kept referring to kids in the generational. I mean, I'm not that old, but I mean, I can remember when my daytime activity was go outside and play with your friends. Like that was what after school was. And then one generation back there, the video games and all, which is, which is great. And it's developed our minds in different ways. But on the human level, I hope we deploy them where it makes us better and not the other way around. Yeah, I share that sentiment. As a technologist and someone who is very excited about this stuff, I learned in Covid that despite the fact that I consider myself an introvert once it was no longer an option to interact with people, it affected me. I think you're absolutely right that there. There needs to be. Still needs to be humanity in this. And I think that'll be a tricky. A tricky tightrope for us as well as we start. I love how. I love the angle that Michael took. And Michael knows this. I don't have TVs. My kids have zero screen time. I understand. Considering how nerdy you are. Yeah, no, I understand. I understand the problem. Yeah. But I also, you know, like, maybe we can manufacture stuff easier without people getting their fingers cut off. That's where I'm like, that's important. Or a humanoid going down and just sort of, like, pulling dead GPUs, slamming in a new one to the array and moving down the line, basically. Definitely. Yeah, definitely. Cool. Well, this has been really fun, you guys. Appreciate you coming on. We'll link to everything thread in the show notes and link to you guys on LinkedIn if anyone wants to reach out, but appreciate your time. Thank you. Thanks so much. All right, people.