Episode Summary
David Cady joins Ian Medley to break down the world of voice AI, including what makes it work, why so many implementations fail, and how Netfor is approaching it differently. They discuss the difference between workflow-driven and knowledge-based AI, the importance of having a strong knowledge base, how AI models generate responses, and why voice personas are more than just choosing an accent. David also shares how Netfor is building voice AI for specialty medical clinics to handle tasks like appointment scheduling and intelligently escalate calls to human agents when needed. The conversation also covers the changing cost and pricing models of AI, how different generations feel about interacting with AI, and even takes a quick detour into the future of data centers.
Episode Summary
Ian Medley: All right, David, welcome back.
David Cady: I see we didn't...
Ian Medley: We didn't quite coordinate on our outfits today, since we're totally matching. I didn't see what color shoes you're wearing, brown?
David Cady: Yeah.
Ian Medley: All right, cool, so at least we have some sort of differentiator today.
David Cady: Yeah.
Ian Medley: But you just got back from a big trip, I understand, a big road trip around the US. So tell me about that, I'm really excited to hear about it.
David Cady: My wife and I drove out to Mammoth Lakes, California, but we took the old Route 66. We stopped along a lot of different places, like Winslow, Arizona, which is famous for the Eagles song, and we had lunch right there, they have a whole thing set up, you can take pictures and stuff, it's a whole tourist attraction.
Ian Medley: How many states did you hit?
David Cady: Like nine or ten.
Ian Medley: Any states you hadn't been to before?
David Cady: New Mexico, I'd never driven through it before, which was stunning, by the way, absolutely beautiful. I can see where people would be attracted to living in a place like that.
Ian Medley: I've been to Phoenix, haven't been to New Mexico, but I'm not a big heat person, so I don't think I'd be able to handle that.
David Cady: I much prefer heat.
Ian Medley: Yeah.
David Cady: So one of my daughters lives in Mammoth Lakes, and the other lives in Lake Tahoe. They work on the mountains, they ski, they do those things.
Ian Medley: Ski in New Mexico?
David Cady: No, in California.
Ian Medley: Oh.
David Cady: Sierra Nevada mountains. They're real big into winter sports, and I'm very proud of them and everything, but I have no interest in doing any of that kind of stuff.
Ian Medley: Yeah, I've tried my fair share of snowboarding, I'm not very good at it, so I'm going to stick to the corporate life.
David Cady: Yeah.
Ian Medley: My corporate hobbies. All right, so today's podcast is going to be primarily surrounding voice AI, what it is. I know we've been in development of a voice AI product for quite some time. I've had a little background understanding what you guys are doing on your team, it's still, everything we talk about is over my head, so I wanted to get you on here to discuss what CX leaders in the space can expect from voice AI this year, next year, and the coming years, what we're doing to further that development within the industry, and take a step back and compare apples to oranges when we're talking about human support versus AI voice support, and how those two can marry together.
Ian Medley: So I'll let you take it away, just a high-level overview again. I know I started the podcast out last time with a high-level overview, but, voice AI, what it is, how it's possible, and what have you guys done to develop this?
David Cady: From a marketing perspective, the first thing I'd want to tell everybody is that it's here, it's probably not going away. It may not be functioning exactly like we'd ideally like right now, but it's getting better every day, and I don't see a time where leaders give up on this because of its potential value. When you're talking about voice AI, it really boils down into several different categories. If you have a workflow, say I need a voice AI that books a car rental, I need to interact with the client, gather some information, and then execute some action on the back end.
David Cady: That's one style of voice AI on the market right now, and honestly it's probably performing the best out of all of them. The next kind is, I need to answer questions, I need to provide some source callers can call into, there's some knowledge base it's tied to, it understands the context of what the caller is asking, searches the knowledge base, finds the answer, and responds. That's probably not performing as well as the first workflow. It's much less deterministic, and it's 100% dependent on how good your guardrails are and how good your knowledge management system is.
Ian Medley: Yeah, because last time we talked about that voice product only being able to pull from the knowledge base you give it, it can't go out and source something else. We talked about that in the last episode, but I have a fact here: 70% of voice AI implementations fail within the first year. Is that shocking to you at all?
David Cady: No, no. Because what we've discovered, I don't know if you have any experience working in a call center, but when you work in one, there's a lot of tribal knowledge that happens.
Ian Medley: Yes.
David Cady: Right, so there's nothing really documented, it's just trained from one person to the next. If I'm on a call and I don't know the answer to a question, I can literally mute my caller and ask you, you know the answer, you tell me, then I tell the caller, and I remember it next time. That's traditionally worked extremely well in call centers, that's how virtually every call center on the planet works. An AI, however, doesn't have the freedom to do that. It can only search the information you give it, and if it doesn't find an answer, or finds conflicting answers, and you don't have an exit methodology...
Ian Medley: A guardrail.
David Cady: Yeah, some kind of escalation path for it to take, and make sure it actually executes that escalation path. Without that, it's just going to make something up or give a nonsensical answer, and result in a really bad customer experience. We've seen hundreds of demos of exactly this happening in the market, where improper guardrails are placed on a product and it's allowed to just do whatever it wants. This goes back to something I've said before, a lot of companies are trying to treat these as autonomous workers, and that is not what AI is.
Ian Medley: Yeah, they're AI human-in-the-loop in a lot of those instances.
David Cady: Exactly. Whenever you leave it to just autonomously do whatever it wants, that's when it's almost certainly going to fail.
Ian Medley: Yeah, okay.
David Cady: And the whole concept of guardrails has been around a while, but it's progressing over time, it's maturing. The more interactions we're able to gather, the better we're able to predict where we need to place those guardrails.
Ian Medley: Okay, yeah, that makes sense. So that covers the 70% of failures. Where are you seeing the 30% where AI voice is working? Is there a specific area where it's working perfectly? I know you mentioned the rental car.
David Cady: Yeah, the workflow. When it needs to execute some workflow, even if it needs to gather information first, it has a conversation with you in normal English, gathers your name, date of birth, driver's license number, whatever's required, and then on the back end there's an API. We've basically removed the need for a knowledge base, all we need is a script for it to follow. Sometimes you don't even need a script, you just tell it, here's the information you need to gather, and depending on how well you set it up, or how trusting you are as a company, you can let it create its own script on the fly.
David Cady: That requires a lot more testing and refinement. But you can also literally just give it a script, ask these questions, get these answers, it stores those answers as an object in the background, and when it comes time to make the API call to execute that action, it has those things it can place in there. It's much more deterministic, do X, Y, and Z, then 1, 2, and 3, and once that's done it can confirm it to the caller and everybody walks away happy.
Ian Medley: Okay, and is there a way for these voice AIs to start interacting with LLMs and getting context from outside instead of just the knowledge base you're feeding it? Is that on the horizon, is that happening?
David Cady: I don't see very many companies doing that, ever. If you train your own model, I suppose, Bloomberg famously trained their own model because they have very specific information that can't be shared, so they realized the only way to really use AI was to build their own model. But for 99.99999% of companies out there, that's not feasible. So you're relying on these models, which are great, Claude just came out with a brand new model today or yesterday called Fable 5.0, and we explore those models. But the point is, I have no idea what Claude was trained on.
Ian Medley: Right.
David Cady: I can't get access to its training data, so I have no way of knowing what it's going to say. It can literally say any nonsensical thing you can come up with. So I don't see a time where, unless, say, for whatever reason your company is just reporting sports scores to people who call in, something very generalized that has nothing to do with your specific company, where I need a way to answer generalized questions, then I could see letting it use its training data, or the internet.
David Cady: But the minute you cross into needing it to answer questions specific to your company, that's when you no longer want to let it use its own training data or the internet.
Ian Medley: Yeah, I think that...
David Cady: You know, many companies have critics. What you don't want is your AI searching the internet for information about your company and suddenly stumbling on a Reddit thread where somebody's trash-talking your company, and now it just tells the caller all that stuff.
Ian Medley: Yeah, "guess what, you'll never guess what this company did back in 2005." I have another fact here: 81% of business leaders plan to invest in AI customer experience technologies, and 91% of customer service leaders are under pressure to implement those AI solutions. That's a pretty high number, that means most companies' CX leaders are under pressure to figure out a way to implement AI, and even cut costs. I think AI can be looked at as a cost-saving center in a bunch of different ways, but when you still need the human in the loop, it doesn't fully deter the cost savings, but I don't think it's ever going to be fully autonomous.
Ian Medley: So that's a huge percentage of companies under pressure to implement AI, what do you...
David Cady: In my view, you have to broadly separate those into two categories. One is if you're strictly focused on delivering good customer experience. The other is if you're trying to replace your engineering or coding staff with AI doing work on the back end, that's a whole other broad category. Recently we've seen a lot of stories about how some leaders are finding that replacing their staff with AI is actually more expensive than just keeping the staff. It depends on the scenario, but there was a pretty famous story of somebody spending $500 million in one month on Claude. I cannot imagine their employees cost anywhere near that amount.
David Cady: That's obviously a very extreme example, but if you're just turning AI on and saying "please write all this code and do all this work we'd normally have engineers doing," pretty autonomously, it's going to basically run 24/7 and build costs very quickly. If you're delivering customer experience, that's a totally different subject. What you're really concerned with there are called tokens, that's how most of these AIs end up charging you. Our token cost for our total workflow architecture is less than 10%, because we only call the AI when we need to, we try to be deterministic always, unless it's a situation where non-determinism is called for.
David Cady: What I mean is, I would never use AI to create and then execute code to call an API. I can write that code myself, and it will happen the exact same way every time, and I don't have to worry about it. There are companies out there, though, that have the AI doing things like that in their customer experience product, and their token costs explode. If the only time I call the LLM is when I want to interact with the caller, I'm only calling it a couple of times per phone call, and it's a very modest expense, not something that even needs to be considered in the grand scheme of things.
Ian Medley: Yeah.
David Cady: So breaking it into those two categories, delivering customer experience versus turning AI on and trying to make it a full-time employee working 24/7, those are two totally different worlds. And that 91% figure, I'm wondering how much of that is on this side over here, because this is the side that's going to continue to struggle. I don't think what the market wanted to happen with AI 18 months ago is going to happen. I don't think it's ever going to be a fully autonomous worker.
Ian Medley: Okay.
David Cady: It's not really functionally designed to do that. I don't think that was ever the intent of AI or LLMs, or even when you get to reasoning models and all these other things, it's always going to require very strict guardrails. If I hired Einstein to answer my call center questions, I still wouldn't just say "have fun," I'd still monitor what he's saying, still provide coaching, still put guardrails around him. We have to treat an LLM the exact same way. Whatever expectations you have for your human staff, you need to have those exact same expectations for the AI.
Ian Medley: Yeah, and you've mentioned guardrails plenty of times already in this podcast and the previous one. So for those leaders under pressure to start implementing AI, guardrails are probably one of, if not the most important thing to establish before shopping for a generative voice AI platform. What else do you think that group needs to be looking out for?
David Cady: Easily the number one thing is their...
Ian Medley: Knowledge management, knowledge...
David Cady: Without question, everything, guardrails matter too, but they're a just-in-case thing, a protection method. Your knowledge base is going to determine whether or not the product fails. Simple as that. If you have some storage drive somewhere that people have just been randomly doing things in for years with no actual knowledge management applied, and you point an AI at it, it's 100% of the time going to fail.
Ian Medley: Okay, so I want to understand this at a very high level. Somebody calls in, and voice AI, am I saying that correctly?
David Cady: That's fine.
Ian Medley: Voice AI answers, then the caller prompts something, and it pulls from knowledge and either escalates or solves it more or less?
David Cady: You know, that's...
Ian Medley: How does that work? I don't even know how computers work.
David Cady: I don't even know how wifi works, first we made this rock and shoved electricity in it, and then it does magic things. But more or less, an LLM, the big modern innovation in LLMs, because we've had models for years, 30 years, I think the first true machine learning models were 35 years ago. But when we were doing translations, we could translate a specific word very accurately, we could not translate the context of an entire sentence for decades. Eventually we figured out how to, with a very famous white paper called "Attention Is All You Need," we figured out how to make it pay attention to the context of the entire text rather than just the next word.
David Cady: It doesn't specifically work exactly like this, but for understanding purposes, every word is turned into a string of numbers.
Ian Medley: Yes.
David Cady: And it's always zero point, and then a bunch of decimals. The best analogy I've come up with, and I find it very effective, is think of every word in the English language as a star in the sky. Some stars are going to be closer together, like "baby" and "diaper," because in the English language we use them very frequently close to each other. "Baby" and "nuclear explosion" probably aren't going to be anywhere near each other.
Ian Medley: Right.
David Cady: Because in natural English language you just don't hear those words together. So what we've created is called vector space, a mathematical space where all words have been converted into numbers and placed next to each other based on how frequently they come up against each other in normal English. So what's literally happening when you ask an LLM a question is it just needs to predict what the next token, the next word, is going to be.
Ian Medley: Is there a prediction success rate? Like is it right most of the time?
David Cady: Well actually, sometimes it's wrong on purpose. If I say "father in," you're expecting the next word to almost always be "law," father-in-law, almost always, 99.999% of the time. But sometimes you want the LLM to not use the most obvious next word, because it sounds more natural. So going back to the star analogy, you draw a circle in that space, and if the previous word is "baby" and it needs to come up with the next word, there's a circle it's allowed to look within, and it can only choose from the stars in that circle.
David Cady: There are things we can do, like top-p and top-k, where effectively some stars are brighter because they're used more often, some are dimmer because they're used less. You can widen that circle with something called temperature. The lower the temperature, the smaller the circle, and the more robotic it ends up sounding. Sometimes that's what you want, very professional, very robotic. Sometimes you want it a little more casual, so you raise the temperature, which broadens the circle, so there are more stars it can choose from.
Ian Medley: This is getting crazy.
David Cady: So there are things you can adjust that basically influence the math, but in the end all that's really happening is some really fancy math, and then it produces the results of that math, which turns into a specific token. What ChatGPT, or OpenAI more specifically, did was create the interface by which that token is transferred into text and shown to you on the screen.
Ian Medley: Okay, that's a great analogy. I can visualize it a bit better now, the vectors, the temperature, the widening. So that's how voice AI is picking each item to respond with, correct?
David Cady: Correct. There are also tons of other systems that go into an LLM, things that control what language it speaks in, and other things as well. But at the base of what's happening, it's what I just described.
Ian Medley: And the language is one thing, but can you also start picking and choosing, like whether it's male or female, high pitch, low pitch, deep voice, monotone? Give it a silly little voice, I don't know.
David Cady: What we've started doing is calling those personas, they're not really just voices because it goes way beyond the voice. You can clone my voice if you want, but getting it to talk like I talk, use the inflection I use, and everything else, is a totally different, much more difficult thing. So it's not simply, hey, here's a bunch of voices you can choose from, it's here's a bunch of personas you can choose from. I can make it professional yet witty, I can give it a Southern accent...
Ian Medley: Yeah.
David Cady: You can give it an Australian accent, literally any accent you want. There are tons of companies that do this specifically for a living, creating these voice personas. Keep in mind, that's different from what I just described about predicting the next token, those are two separate systems.
Ian Medley: Yes.
David Cady: One system is generating the text, the other system is reading it aloud, two completely different things.
Ian Medley: Okay, and one thing everybody can relate to, even this year or last year, Alexa came out with a new voice you could opt into. Have you done that?
David Cady: I have not.
Ian Medley: Do you have any interest in it?
David Cady: I do not.
Ian Medley: Oh, I love Alexa.
David Cady: Yeah.
Ian Medley: It went from the same voice it had for years and years to something a little more flowy and crisp, I'd say. That's one instance I think people can start to visualize firsthand.
David Cady: I'm surprised they didn't offer the ability to record your own voice at this point, that's such an old-hat thing now.
Ian Medley: I wouldn't want my own voice talking to me.
David Cady: But it might be funny to do somebody else's, there's probably some legal reasons you might not be able to do that kind of thing. But at this point, creating voice personas is a very mature technology, we're really good at doing that.
Ian Medley: We, as in everybody, or...
David Cady: Yeah, as in, well, the industry is good at doing that.
Ian Medley: Sweet, I want to hop into Netfor. What's your team working on right now, up in the lab cooking?
David Cady: Right now we're focusing on specialty clinics, think orthopedics, pain management, things like that. We go back to, they're very workflow heavy, gather information, book an appointment kind of thing. I hesitate to use language like this, but those are orders of magnitude easier to pull off. Anytime we're faced with having to look up knowledge, it always goes back to the same thing, if you don't have good knowledge management already, there's no hope of getting that product up and running. But if it's, we need to design some workflows for you...
David Cady: That's something that, in reality, Netfor's been doing for thirty-some-odd years, designing workflows for clients based on what the knowledge tells us we're supposed to be doing. The only thing we're doing now is inserting AI for capacity. Anytime we're strongly following the advice of whatever we expect our humans to do, we expect the AI to do as well. Things like making sure it uses the proper greeting, the proper closing, doing QA on it the exact same way we do QA on human agents, which allows us to iterate on it and make it better over time.
David Cady: And that's probably the main thing we've learned, that you have to iterate on it and make it better over time, exactly like you would with a human.
Ian Medley: Yeah, train them, upskill them, promote them, continuous learning.
David Cady: Yes, when you first hire a call center agent, they may be a little nervous, or they don't have the recall yet to instantly remember things, there are lots of things that get better over time, just naturally. This is the exact same situation, and I don't think the market understands that yet. I think a lot of CX leaders expect us to turn it on and have it be perfect from the get-go, and it just isn't, it doesn't really work that way. We need real interactions with real callers to see where we need to improve it over time.
Ian Medley: And you think the ideal model, in a perfect world, would be voice AI solving really low-tier issues and then escalating to a human if anything persists, or if the caller wants to speak to a human right away, without giving them the runaround saying "we don't know where a human is, you have to deal with me," and having them keep pressing 0 or shouting.
David Cady: We had an interesting idea exactly about that, I'll talk about the rest in a minute. Most AI products actually have more than one AI operating at the same time, it's almost never one model doing everything. We had the idea of having a second model review the conversation in real time, and after either the caller or the model is done talking, that second model had only one job: is it necessary to get a live human on the phone, based on the context of what's happening in this call? And we turned some of those settings I was describing all the way down.
David Cady: We turned them all the way down to where this second model only had two words in the English language: yes or no, that's it, that's the only response it could give. There was zero chance of it saying anything else, as far as that model knew there were only two words in the English language. So after every turn we'd say, review the context and decide, should we be transferring to a human? That created a neat little way of monitoring the situation to make sure all the normal escalation paths we set up were happening correctly.
David Cady: Anytime a caller says they want to talk to a human, we pretty much exit immediately.
Ian Medley: Yeah, that's what everybody wants.
David Cady: Or if it asks you the same question more than twice and can't understand what you're saying, we exit to a human, or at least offer it. Anytime we detect a bad customer experience, we're trying to exit to a human.
Ian Medley: Customer sentiment's huge.
David Cady: Yeah.
Ian Medley: And it measures that based on tone of voice and words used?
David Cady: Yes, absolutely. Words used is always the easiest one. But monitoring sentiment, a lot of people treat it as post-processing, after the call is over you run some process and get the sentiment of the call. We do it during the call, we want to know that from the beginning of the call to the end, customer sentiment at worst stayed the same, but ideally got better over time. That's what we're monitoring.
Ian Medley: Is there like a meter on the agent's screen showing their happiness level, like at the hospital, a scale of 1 to 10, somebody smiling or frowning, a "warmth level" that measures it after each sentence or interaction?
David Cady: Now you're talking about something called real-time interaction guidance, what's being displayed to the actual agent. It depends on what you mean. If we're back on the LLM and it's just the LLM interacting with the client, there's really no need to display something in real time, because the LLM can just read the numbers off the text. But if it's real-time interaction guidance, the LLM isn't responding to the caller, the agent is, and the LLM is monitoring the call for things like sentiment. Some systems are good enough to make it real time, telling you after every turn of every call, a turn being either the caller or I have finished talking, that ends a turn.
David Cady: And then it updates the sentiment based on that.
Ian Medley: Awesome, and we're already there. Okay, cool. Back to what we were just talking about, what's your BI team doing right now in development, specifically with specialty medicine, managed patient access, stuff like that? What does success look like from your seat if someone implemented that into one of those new practices?
David Cady: Success looks different depending on which hat I'm wearing. If I'm wearing a technical or engineering hat, success looks like when I ask you for your name, I can accurately record the response. When I'm executing code to call an API, that code is written correctly and executes correctly every time. That's from a technical engineering standpoint, and the measures of success are really the same as they've been for any technical product released in the last 40 years, it's just that the product functions a little differently now. From a marketing, from a sales standpoint, success in the market is trending toward interaction. So instead of paying per minute, a lot of people are paying for successful interactions.
David Cady: So for instance, if I'm building something that books you car rentals, you want to pay based on the number of car rentals booked instead.
Ian Medley: Instead of how many minutes.
David Cady: Correct, or how many interactions it took, that kind of thing. Because the scale is so different, if it's not a human answering those calls, I can answer 10,000 calls in an hour, and in a lot of situations when a client had an influx of phone calls, they wouldn't necessarily see that in their bill, because the way those calls are answered kind of fixes that problem. With an AI, it just answers all those calls at the same time, within a couple of seconds, and bills you for all of that time. So instead, the market seems to be trending toward paying for a successful interaction.
Ian Medley: Yeah, that makes sense, a per-interaction pricing model. And is that being well received by clients looking for these kinds of services?
David Cady: From what I've seen so far, yes.
Ian Medley: But is it also moving toward a permanent basis?
David Cady: No, it's also kind of a newer development, I'd say 2026 is only just now when there've been enough products on the market and enough experience with them that the people buying these products are starting to understand what's possible and how they want to pay for it.
Ian Medley: Okay, perfect. I think we're going to close out pretty soon, but I want to circle back to one more fact I wanted to discuss: 85% of AI support failures trace back to poor knowledge management and no clear protocol for handing off to a human. That reiterates what we were discussing about knowledge and guardrails being the most important things to evaluate for the next step in customer experience evolution, whether that's voice, chat, email, text, or social. So there's a lot more to discuss, but I'm glad we hammered out what voice AI looks like, and you explained your stars-and-galaxies analogy to me, I thought that was fun. Something you and I were discussing before we hopped on here, I don't know if it's really off topic...
Ian Medley: I guess it's kind of on topic, but the generational gap, you're Gen X, I'm a millennial, and there's a wide array of different generations. How do you think each one is welcoming this change, and which generation is being the most resistant? I think we all know it's probably the older generation, maybe being a little...
David Cady: Bit more resistant, but no, exact opposite. Right now, Gen X, unfortunately, is becoming the older generation, and the baby boom generation, they are more well received to this than I anticipated, because some of these products are good enough that you almost cannot tell you're talking to an AI. When it goes off the rails, they're obviously quick to point that out, but generally I've found the baby boom generation to be perfectly amenable to most of this stuff.
Ian Medley: Interesting.
David Cady: Gen X again, for the most part from what I've seen, we grew up with no internet, then the internet, then no AI, then AI, so we're kind of used to this kind of stuff, we see it as almost inevitable. Once you get down to the millennial and definitely Gen Z generation, the demeanor changes pretty dramatically. I think the way it's being handled, with a lot of news around how big these data centers are, what it's taken to get them up off the ground, a lot of news about layoffs, unfortunately millennials and Gen Z have, economically speaking, run into hardships pretty much their whole lives, and it feels like maybe they feel like this is just the next one.
David Cady: It doesn't have to be that way, it doesn't have to be a bad thing, but there certainly are companies that aren't handling it in the best way. I mentioned one of my daughters lives in Lake Tahoe, there was a pretty famous story a couple of weeks back about how their electricity provider is just going to stop providing them electricity so they can provide it to a data center.
Ian Medley: Probably a lot more money, I mean it's not right by any means, but profitable.
David Cady: Yeah, and they're left wondering where that leaves them, in the dark. What's literally going to end up happening is they're going to have to buy it on the open market and pay a lot more than they were anticipating. I don't think it's a matter of them actually going dark and not having electricity, but I think it'll be an expensive thing for them.
Ian Medley: Yeah, and you mentioned data centers, and I think I saw this, I don't know if it's factual, data centers in space?
David Cady: There are just so many issues with that, I mean, it's cold.
Ian Medley: It's cold up there, I know.
David Cady: In theory, but it would be a lot easier to put a data center a thousand feet under the ocean or something, where it's also cold. The problem is, data centers have to use not just fresh water, it has to be extremely clean water, they can't use salt water, it would corrode everything, there's a lot of metal that water is touching, so it needs to be extremely clean and cold, continuous space. Sounds theoretically nice, but the logistics of getting that material into space and assembling it there, the International Space Station cost about $80 billion, I think, and is relatively simple compared to a data center, and it's kind of small.
David Cady: It's bigger than you think it is, but compared to a data center, it's very small. And then there's getting the data to and from space.
Ian Medley: That's kind of like satellites though.
David Cady: I mean, it is, we know the technology exists, I don't think the economics exist to do something like that.
Ian Medley: I know there's going to be a company in 10 years that has space data centers, and we're going to be here talking about it.
David Cady: We'd need to lower the cost of getting material into space dramatically. Not to mention, you'd need staff 24/7/365 in that data center in space.
Ian Medley: Data astronauts.
David Cady: Yeah, data astronauts.
Ian Medley: You're going to be the first one.
David Cady: I would do it, I would 100% do it. I would sell one of my children to be able to go into space.
Ian Medley: That'd be pretty cool, not selling your kid, but the first data astronaut, whatever you want to call it. There's got to be a space force term for it.
David Cady: You know, it's like mining asteroids, a fantastic idea, theoretically would work, but the logistics of it, we're not there yet.
Ian Medley: Elon will get us there, or somebody, Bezos. No, well, he just had a rocket explode.
David Cady: Yeah, I mean, that's how you learn.
Ian Medley: That is how you learn, for sure. Well, we're not exploding rockets, we're just talking about voice AI, but thanks for coming out, I appreciate it, it was fun as always.
David Cady: It's a blast.
Ian Medley: Try to get a better cadence, I know it's been a couple weeks since our last one, so...
David Cady: Yeah, we need to just set it up on a schedule.
Ian Medley: We will. But thanks again, I appreciate it, and look forward to talking to you guys next time.
David Cady: Absolutely, thank you, Ian.
Full Episodes Available on YouTube and our Podcasts page