Surviving the New Economics of a Post-Agentic World

Welcome to the Practical AI Podcast, where we break down the real-world applications of artificial intelligence and how it's shaping the way we live, work, and create. Our goal is to help make AI technology practical, productive, and accessible to everyone. Whether you're a developer, business leader, or just curious about the tech behind the buzz, you're in the right place. Be sure to connect with us on LinkedIn, X, or Bluesky to stay up to date with episode drops, behind-the-scenes content, and AI insights. You can learn more at practicalai.fm. Now, onto the show.

Welcome to another episode of the Practical AI Podcast. This is Daniel Whitenack. I am CEO at Prediction Guard, and I'm joined as always by my co-host Chris Benson, who is a principal AI and autonomy research engineer. Welcome, Chris. It's, uh, it's one of these episodes where it's just the two of us and we get to talk about whatever's interesting for us. So, I'm excited about this. I missed the interview with you last week. It was a great one. But yeah, excited to be back on. Absolutely.

Welcome back. And I know we both had outages lately with summer vacations and family and things like that that we've been doing. Good to be back together. And yeah, these fully connected episodes as we call them where you and I get to kind of go wherever we want to go. Are, they're always fun for me.

Yeah, for sure.

For guests, it gives Dan and I kind of the chance to freelance and to kind of, instead of just focusing on a particular topic, to kind of go wherever we want to go. And so we have a good time with them. So yeah, a lot's happening right now.

A lot's happening. And yeah, just as a reminder also for our guests, a few things we don't normally share on our shows when we have a guest because we like to get into that, but please do engage with us online. If you didn't know, we are posting videos now on YouTube. So if you haven't got a chance yet, at least go over there, give us a subscribe on YouTube, The Practical AI Show. And of course, you can still listen to us on all the other places as well. And then reminder, just coming up in October, October 15th in Indianapolis, we're going to have another Midwest AI Summit, which was a great experience. Chris and I got to, to jam out a little bit last year, and excited for some really cool speakers that we have on deck this year, and lots of practicality with an AI engineering lounge where you can sit down and talk through architecture and design and, uh, agents and plans and security and whatever you want to talk about with practitioners. So Check it out, just search for Midwest AI Summit and make sure you get registered for that. But yeah, lots of things happening.

Before you get away from that, I wanna point out that that's a fun conference. You and I go to a lot of conferences and that one is a fun conference.

It is.

And everybody is accessible. If you wanna talk to somebody, they're there. It's great. So I just wanted to point that out. I love that one. So yeah, you wanna start us off?

Sure, sure. It's interesting today, this morning, I mean, I guess like most people now, we don't have regular TV, but sometimes my wife and I listen to some sort of like feed of, like on Prime Video, they have CNN news highlights or whatever. So sometimes we'll listen to that in the morning while we're eating our cereal. And this morning on those news highlights, it highlighted that IBM had a major hit in their stock. So 25%. Stock plunge, which is kind of crazy. And it talked about, oh, IBM has this 25% plunge in their stock. It has something to do with AI. And I was obviously paying attention. I don't know that I, based on how they were describing it, it immediately made sense to me, but I did a little bit of research and, you know, am looking at this and also thinking about if it's a wider trend that is going to be happening across tech companies. So is this something you ran across or have been thinking about, Chris?

I wasn't at all surprised about this in that way. Like, I, I didn't know that IBM would do this, but so it's something I've been thinking about quite a lot lately. And, uh, it's something that I know I've mentioned this to you. I'll mention it without any names. I'm, I'm in discussions about writing a book that kind of hits this topic, uh, with the CEO of a, of a publishing company, uh, and I'll leave it at that. It may never happen, but they asked me to write some books that they were— a book that they were interested in. And I said, no, I'd rather write this book. And, um, and, and the, the notion of the book, if it, if it ever comes to pass, is kind of a post-agentic world. It's kind of like if you look at what's happening right now in the marketplace and you take all your biases and all the things that you want or are scared of out of the equation to where all your emotions are, are removed a little bit and you say, Well, these are what's happening and you play the events out over months and years. What are the probabilities of various outcomes that are looking like based on today? And I've been going through this exercise and spending quite a lot of time on it. And so that's why the IBM thing didn't surprise me. And I think just to scare the heck out of people, I think that we're going to see a lot of that disruption happening. In the months and years ahead because you're really looking at a situation where you have technology that is replacing exist— both existing technologies and existing human positions and processes that people are engaged in with these new technologies. And as we scale out that in an agentic world, that has quite an impact on the fundamental economics of, of how businesses are, are changing now. We've seen all the mass layoffs in the tech industry, and we're going through a moment where, where companies are really experimenting with hyperproductive alternatives to human labor. And so that doesn't remove all the humans from the equation, but what it does is it changes their roles and it changes the activities that the humans are engaged in. And that's evolving very, very rapidly. And so, that, you know, as a start, you know, it, it, it leaves it as something, you know, I know it sounds very ominous, but I think the, the key point there, rather than just being frightened of it, is really that the rate of change is accelerating exponentially right now. And so the IBM thing is kind of that, I know the, the New York Times referred to it as potentially a canary in the coal mine, and that's one way of looking at it., but we're going through the beginnings of a period of rapid change in the fundamentals of business, not just technology. It's really important that people catch that, but business itself that's using that technology. So, I'll stop there for the moment. How would you react to that?

Yeah, I think there's a lot of things tied up into this discussion, some of which were cited in the IBM case, some of which maybe were not. There's, and maybe we can get into a few of these things. One of them would be like where AI companies are trying to become more sticky and make sure that you stay with them so that they can recover some of these costs that they've put in and just sort of hemorrhaged over time. There's also the kind of, panic buying situation of people trying to make sure they aren't priced out of AI and they get actual hardware that will make them a little bit more resilient to that. There's the, you know, I saw the, I think it was another article that you had posted to me about, or maybe it was something that I saw elsewhere, but Anthropic really doubling down on kind of implementation and services side, which we've also seen with, with OpenAI versus kind of the model side. So, so there's so much tied up into this. Just to give some stats on the IBM case, the kind of canary, uh, 25% single trading session collapse, which is about $70 billion in market value, marking it the, uh, worst single day drop in over 50 years. Outpacing even its losses during the 1987 Black Monday market crash. So, and part of this where I was researching is like, well, what's going on? What are the, at least the cited dynamics that they're talking about in relation to this on from IBM and other analysts? And part of what they're talking about is that companies are diverting their their IT budget or their technology budget away from enterprise software and services. So enterprise software and services to some of these other things like the, like I was mentioning, the panic buying of maybe hardware or infrastructure such that they aren't priced out of maybe what they see as coming as a price hike in the usage of these AI models, the agentic future where you need much more context, you have long-running processes, et cetera. And so there's been this shift, I guess, from the capital expenditures on enterprise software and services, at least to some of these hardware and infrastructure expenses, as kind of, yeah, there's finite budget and there's this tug of war happening. So any thoughts on that piece of the puzzle?

I, so I, all those things I think are contributing and there's something else I don't know if you saw this in the news, it kind of skimmed by pretty quietly, I thought. And that was China is now recognizing the value it has in its open models, which are much cheaper to use than these expensive proprietary models in the US. And they're recognizing that while the US has put export controls around that, that these open source models that we've assumed that the world will kind of fall back to because they're leading the way in that, they may now become restricted going forward. And so there's this political, like, there is one set of capabilities that the US is stronger in and is putting protections around. And now China has recognized that there is another set of capabilities that they are stronger in., you know, of economic value and that they're putting protections around that. So, I think all of these things that we have talked about are, are kind of feeding into, to how people are shaping their use of technology. And I think it's creating, to your point, a certain level of panic, in, in executive-level companies where they're trying to figure out how to mitigate the risk, both in terms of the proprietary US models, the expense associated with that. But the fallback strategy of going to Chinese models, if you're in an industry that allows for that, is now closing off potentially as well. And so that is putting a lot of pressure from multiple sides on companies on what they're going to do. And so they are definitely funneling internal budget to try to mitigate that risk. At the expense of a lot of other things that have historically been kind of key keystone operations, you know, in their organizations. And I think, you know, whether we're talking about IBM's collapse and the shifts that they're making, you're seeing all these influences affect kind of emergency allocation of capital. And so, and I don't expect that process to stop. I don't think this is a blip. So I think those things will continue to evolve, but those reactions will continue to out.

Have, have— Chris, I, I was in a meeting with our engineering team this morning and one of them mentioned the Chinese token black market. Have you heard of this?

I, I have.

Yeah. So I— this is actually— so I'm, I'm all the time learning new things. My engineers are on Reddit more than, more than I am, so I, I don't— I hear about things through them. But yeah, so the— it's related To some of what you're talking about, there's this Chinese token black market, which is kind of a reference to this like underground gray market area where brokers are reselling discounted API access to Western artificial intelligence platforms, AI platforms like OpenAI or Anthropic, because those services are blocked to mainland, you know, Chinese, uh, uh, users. There's a lack of the ability to actually pay for, for that. And so there's this economy or ecosystem of resellers and proxies that's emerging to work around this, this kind of access piece.

And, and those kind of things almost always happen when, when you, uh, when you put in kind of regulation for the purpose of artificial shaping, you know, you know, for political ends and such. So you're going to have black markets that appear. So I'm, you know, that, I don't think that should surprise most people seeing that. So, you know, the question is then, where is everything going, you know, with this?

And what does it mean for us as viewers? Yeah, if you're an enterprise software, well, I guess if you're an enterprise software consumer or you're an enterprise software vendor, right? What is the future, I guess? Yeah, that's so like if people are moving away from investment in enterprise software, where are we going?

Yeah, I mean, one of the things that I think is interesting is that lately I'm hearing a lot more conversations about, you know, not only looking for other sources of open weight models that are out there, you know, that are not specifically Chinese or specifically from the US. Obviously Europe is doing that. We've talked to some of the organizations there, but also I'm starting to hear about the recognition that there may be a need, especially as models are now getting, you know, we're getting smaller models that are used for a lot of specific purposes and edge cases to actually go and do training of models or getting back to fine-tuning. I think a lot of organizations got away from it. Remember we used to talk about fine-tuning, you know, once upon a time, and then it seemed like the models got so good that a lot of organizations said, why would we bother with that? And they would just take a model and it was good enough without fine-tuning, without the cost of that. And so I'm hearing a lot of like, what do we do as we get squeezed from both directions? And how do we approach that? And so some of these things that people had said, eh, let's not, we don't need to do that. We don't need to engage in that, are now coming are now kind of coming back around. It seems like those are conversations that are starting to happen again. And so I found that, that very interesting.

Yeah, I, I think that that also becomes, I mean, there's platforms like Unsloth and others that allow you to fine tune, you know, even on your MacBook and, you know, run things a lot more efficiently. So I think that also gets to some of what this was the episode that I recorded when you were out on vacation, I think, Chris, with the ZenML folks. But as we transition from agents being like my personal assistant on my laptop, which I interact with back and forth, which obviously limits the speed at which and the amount of context that, that agent can consume. And I move that into the cloud as a long-living agent that just works for hours and days and weeks maybe, right? Then all of a sudden the economics do change around like, hey, if I'm using a pay, pay as I go API endpoint, that's gonna be a vastly different economics than a small self-hosted model, right? That, that's powering those agents. So I, I do think that that makes a, makes an impact.

I do too. And I think, you know, but I think what, I think we're moving past that, you know, when we talk about having our agent that's long-lived and stuff like that. And this also depends on industry and need and stuff like that and requirements. But I'm, I'm looking at situations where, loops are incorporating, hundreds of thousands of agents or millions of agents, that are, that are all do, creating, very complex contextual productivity.

Yeah.

You know, in or outcomes, I would say, you know, in terms of what you're trying to do. So if you are, if you are just assigning agents a bunch of very discrete responsibilities, and that there's kind of teams responsible for each of those and they're interacting a lot, but, and I think that's the kind of thing as we're moving toward, and that becomes more common across industries that you're, you're looking at a set of capabilities that, you know, going back to the IBM thing right here, where that level of agentic implementation and agents running agents, which is happening more and more, is really taking us into a new world. Like, it changes, it completely changes the value proposition for a lot of enterprise software out there because you suddenly have a new capability that in many cases can do things that traditional enterprise software just can't touch. And I think that goes to the, you know, when we talk about IBM stock today, I think that's the threat. But I mean, IBM is just one significant company in multiple industries with lots and lots of companies that are essentially taking that enterprise software approach, and they are all at risk to some degree. And I think that's going to continue to play out in the marketplace. So it's, you know, the world is shifting. It, it, it, the paradigm of operation is shifting significantly right now. Uh, and the economics are going, the economics that that implies are going to, to play out. So I think, I think, you know, I think that's why, and you're seeing some recognition, maybe not of the full picture at the executive level, but enough to where they're saying, ooh, we really need to spend money. You know, we need to move capital from here to there. And, and so, and that's impacting, that's impacting employees. Such.

Yeah. And as a, as a founder of a software company, obviously I believe there's a place for software in, in the future. But I do wanna validate your point of like, like with, with our vision and what I'm trying to paint at PredictionGuard, we're working towards that future that you mentioned of the agentic workforce, the, uh, thousands, hundreds of thousands of agents that need to run, run securely, run governed, run with identity. And just to validate your point, like even in the last couple of weeks, we've talked to one, one company that already has 70,000 agents running, one that has, I think it was like 6,000 or something. So this is not, yes, it is future, but it's not like 5 years future. This is, this is very rapidly what we're, what companies are, are. So I think there are many people out there that are maybe still on a one-to-one basis with their cloud with code or with their Hermes agent or whatever it might be. And so this picture of thousands of agents running in enterprise infrastructure might seem far-fetched, but I think it is not. It is very much, we're getting there very rapidly and people will start really experiencing this, which I think stresses all of us to think of like, you know, in our In my world, it's very much like, hey, as a software vendor, how do we prepare for that? What is your thought, Chris, in relation to, I guess, if you have enterprise software that people have been using for some time, let's say like IBM or even a smaller midsize or a niche vertical, software vendor, because I've talked to a lot of those vertical SaaS software vendors, right? How do you, like, how do you lead your company ahead into this world and what, what becomes important? What strategy is, is important as you move into this world? I've seen some take the approach of, hey, well, you know, like a NetSuite or something like that. They're saying to some degree saying, well, we're gonna, provide a connector into the AI world, right? Which is often MCP, right? So their value is maybe they're assuming their value is in their data platform, how they organize data, the functionality that you're, that they provide, but surfacing that in an agentic way. So that's, you know, through the MCP side. So that's one take that you have. You have others then, you know, maybe more vertical. Software companies that are saying, no, we're not going to expose that, but we're going to create our own proprietary set of agents that are our agents and are going to run in, you know, what you need. And we're not going to tie into the more generic kind of agentic ecosystem. Do you have any thoughts on that kind of like, or maybe there's like other options within those strategies, right?

So I think so. Yes, I have some thoughts on that. I think The world, so the way interactions are going forward is not going to be in traditional interfaces, you know, the GUIs and the web interfaces that people are used to, you know, and that's dominated all the way through kind of the early internet era. And then as we got into the cloud era and then even the beginning of the AI era, as we've gone to, you know, app interfaces where we've interacted, but going— as we look at agents doing all these different things and collaborating, the way interactions will occur is agentically in a direct, you know, if you might think of it as B2B in a sense, but think of it as agent to agent. And the vast majority of interactions across different systems are going to occur agentically without a human directly involved in most of those processes. So, with those processes being automated,, through agentics, um, you have to be thinking that way. So if you're one of these companies, you know, I know you mentioned NetSuite and obviously there's the IBM concern and lots of others. You have to be thinking, how do I move from where I'm at right now, uh, into a world in which agentics are incorporated from a, uh, a business, uh, a business transaction consideration. You know, how that it's going to happen without anyone going into your GUI and, and without that, you're going to set the permissions, you're going to set, you know, what they have access to, um, and how the MCP servers are, are configured for, for the different parts. And that's quite complex. And I think there's a whole industry, uh, right there of how do you manage, uh, Agentix at massive scale with resources and MCPs. And so like, if you're an entrepreneur out there and you haven't, you know, that's, that's an area that is wide open. And I know your company is already addressing a lot of those things. And so it's, you know, that's the kind of thinking. And I think going back to your point, this is happening really, really fast now. And so if you're thinking we're a few years out, then you're gonna get overtaken quite quickly as IBM discovered this morning with their stock. So not trying to create panic, but people in these organizations that are stepping into kind of AI and now agentics, they need to be thinking not where are things to, like tomorrow, fairly big leaps are happening. And, you know, like right now as we're talking, you know, loop engineering, you know, we were agentic engineering was, you know, kind of the buzzword a few months ago. And then lately it's loops because now your agents are tied up in loops, but tomorrow it's going to be past loops. Looper is a very transitory thing because this is evolving so quickly.. But before the year is out, I'm pretty sure the notion of loop engineering is going to be kind of antiquated by what follows and the evolution of where things are going. And so if you're a business owner, get on top of what's here now and start thinking, how am I going to get ready for those agentic interfaces of the future? And that's how I would start addressing the problem from any given business perspective.

Yeah. Just to make sure we're not painting or just taking one data point. This is much more widespread, as you mentioned, and is spreading quickly. I was just looking while you were talking and on the same day today as IBM crashed, it looks like, unless my agent is hallucinating, Workday slid down 10%, Salesforce 9%, ServiceNow 8%, Adobe 6%. %. All, you know, similar dynamics going on. What did gain was the, you know, Nvidia, Intel chip providers rising as people, again, you know, panic buy some of these hardware solutions. And there is a lot of, I think, distraction related to the cybersecurity side of this as well, which is partially related to the hardware, but more related to maybe some some of this distraction around mythos and cybersecurity and how agents violate cybersecurity. And so there's a focus on that rather than kind of the generic SaaS software and enterprise software as well. So just wanted to bring in a couple of those points. I was hoping my additional agentic research would back up our points and it seems like it does.

So I think so. And I think, I think those are good callouts in terms of you know, you're, you're looking at enterprise software companies struggling, uh, because people are starting to realize what's happening, uh, and not, not going to happen happening. And, and so I, I think to your point, there's also needs to be a rethink. Uh, and I think this is really important and, and, and I, I'm not claiming to have all the answers by, by any stretch, but there needs to be a rethink about how humans are fitting into this equation. I, I do think that there is a place for them. I think it's very different going forward. I think trying to hold on to old roles is not the best strategy. I think being creative and I get into these conversations with people all the time where they're kind of holding on to what was. They don't want to let it go. It's what they know. It's what they're comfortable with. I saw a quote earlier today from George Lucas saying, if you're not really on top of this with agentic, you're kind of holding on to the heart, the the cart and horse in a day where the automobile is happening. And I think that still happens a lot because I get into all these conversations, especially with older, you know, people closer to my age that are seeing the world change out from under them very quickly. I think the right strategy is to recognize what agentics is really good at doing. And also, you know, like we're looking at things like mythos and fable from which are the latest generation, able to really do things that no human is able to do, Mythos especially in terms of cybersecurity. And I think you embrace that. I think you recognize that's not going to change and you embrace those change and look for places where you can plug in on that. And I'm going to do a lot more exploration of these ideas in the days ahead, especially like I said, if I end up writing anything about it. Yeah, it's, the world's changing really fa— faster than it ever has right now. It's speeding up much faster than it was a year ago. And so it's a good moment for people to self-reflect a bit.

Yeah.

Yeah.

And that gets, this is maybe something for a complete other episode once we do the full research on it, Chris. But to your point, in terms of how things are advancing, Anthropic did just release a paper I forget which day recently, now, whenever it was, verbalizable representations form a global workspace in language models. This is basically, that's a lot of words, but part of this is talking about, you know, what forms of consciousness access emergent behavior, or like how understanding is represented in these models. And there's some interesting things there, obviously. There's always a wide range of opinions once you start thinking about, or once you start proposing what is cognition, what is consciousness, what types of consciousness are there, what can be included in these models, what do they understand. But anyway, there is continued thinking there. And I think one of the points of the paper, if I'm understanding it right, whatever your take on sort of consciousness and cognition is that there is this functional capability of these models to, an improving functional capability of these models to route and report information. So this is not like they're feeling or they, like they have emotions necessarily or other types of consciousness, but certainly there is this capability of routing and reporting information, which is obviously key to the, uh, key to the agentic transformation that we're seeing. So there may be more on that in the future.

Yeah, I think, you know, one of the things in the paper that you're referencing that was pointed out was that Anthropic had noticed that models are creating what they're calling workspaces, where they're essentially— the notion of the workspace is fulfilling the same function as working memory in a human brain. And, and that, you know, and so I think one of the big questions, and you have both sides of people on the consciousness side, some people are saying that's not what consciousness is. And other people are saying, what if? What I would say is, sadly, there's never been a unified definition of what consciousness is that is widely accepted. So that needs to be put out there. Um, but I think that there's also the consideration of if, if you have a model kind of achieving an outcome, an apparent outcome based on what, based on what it's doing, um, but it achieves it in a way that's very different from what we traditionally would associate with. Um, and like, so if you're really trying to model consciousness around what a mammalian brain does and how it achieves that, Or are you willing to say there are alternative paths that if you get to the same kind of an outcome for a given task associated with that, is that legitimate? And I think that's kind of, when I look at the argument, that's how I perceive that is there are people who are, who have a very strict and narrow definition that are probably traditional neuroscientists in terms of how they're doing it. And then I'm seeing other people that are kind of going, going, but if it's getting to the same place in function, you know, not full consciousness, business, but some of the things that would contribute toward that, does that count? And so there's a little bit of a philosophical debate on, on, um, what's legitimate at this point. Um, I think it's interesting. I, I, I, and it wouldn't surprise me if, um, some of these things arise eventually, emergent qualities arise from vastly different ways from what we had anticipated. So, um, I'm pretty open-minded in terms of how things can, can something that we never would've expected to happen contribute towards something that we were ultimately an outcome that we were looking for, Seth?

Yeah.

Yeah.

I like, um, actually it was a suggestion from one of our customers and, and when we were discussing things, uh, they were, they were thinking about the rather, uh, you know, obviously there's an architecture associated with agents that each company is trying to enable, but thinking about things at the level of outcome. And how, how do we achieve these outcomes and what's the necessary human input? What is the possibilities with the agentic systems that we, that we can deploy? I think it is very useful to think at that outcome level. And Chris, I'm, I'm pretty, pretty happy with the, the outcome of some of this, this discussion. I think it was a good one. Enjoy.

This, these, for folks that are watching or listening, these are completely unscripted. This is just us having fun. So, uh, yeah, that's a good one today, and, uh, it gives, giving me a lot of, uh, food for thought, uh, for going forward. Uh, we look, we hope folks will engage us on the social, social media channels, uh, YouTube and the others that they find us on. Um, give us your feedback, let us know what you what you think. We'd love to, uh, we'd love to get your insights into these items.

Yes, for sure. Have a good day, Chris. We'll see you soon.

Take care.

All right, that's our show for this week. If you haven't checked out our website, head to practicalai.fm and be sure to connect with us on LinkedIn, X, or Bluesky. You'll see us posting insights related to the latest AI developments, and we would love for you to join the conversation. Thanks to our partner PredictionGuard for providing operational support for the show. Check them out at predictionguard.com. Also, thanks to BrakeMaster Cylinder for the beats, and to you for listening. That's all for now, but you'll hear from us again next week.