Build What Differentiates, Buy What Accelerates: AI That Earns Customer Trust in Financial Services

Most financial institutions are approaching AI backwards — starting with the technology and working toward the use case, rather than the other way around. Karan Kashyap, CEO of Posh AI, has spent eight years helping banks and credit unions deploy conversational AI across both customer and employee channels. In this episode, he makes the case that the riskiest thing a CIO can do right now is nothing — and he gives technology leaders a practical framework for deciding what to build, what to buy, and how to design AI experiences that build trust rather than erode it.

What You'll Learn:

• Why financial institutions are still the slowest AI adopters — and why that conservatism is becoming a competitive risk against fintech
• How to design AI that builds customer trust rather than triggering skepticism — and why disclosure is both a regulatory requirement and the right call
• The difference between automating a task because you can vs. because you should — and how regulated industries like banking and healthcare should make that call
• Why voice AI is still in the uncanny valley and where the real low-hanging fruit is (hint: it's your search bar)
• How agentic AI and the rise of specialized "mini agents" is changing the way financial institutions think about customer and employee-facing deployments
• Karan's signature framework: build what differentiates, buy what accelerates — and why the in-house build is usually only 3% of the total effort
• Why the riskiest thing a CIO can do right now is nothing — and how to make the case for an evergreen innovation fund to your board
• How AI technical debt accumulates fast, and what institutions that tried to build in-house eventually discover

Transcript

SPEAKER_01 0:12

Hey everyone and welcome to the Amplitude of Tech Podcast. I'm Sean Cordner, Chief Marketing Officer of Amflix. Today I had Karen Kashiep from uh Posh AI. He is the co-founder and CEO, super smart guy, really interesting takes on AI. We talked a lot about how companies can implement AI and still maintain trust with their customers and a whole lot more. This was a great conversation. Hope you enjoyed as much as I did. All right, Kern Kashup. Welcome to the podcast. Thanks for having me. Appreciate your time today. Why don't we start with just a little bit about you and about Posh AI?

SPEAKER_02 0:50

Sure. Yeah. So I'm uh Karen. I'm the co-founder and CEO at Posh. And uh what what we do and who we are is we're about an eight-year-old AI company. We consider ourselves a vertical AI company. So it means we're really focused on the domain of uh banking and credit unions and financial services versus being a horizontal uh platform. We kind of really focus on the vertical depth and domain expertise. Um we started out of a research lab at MIT about eight years ago. So myself and my co-founder, a lot of her founding team, were all grad students at MIT working on uh natural language understanding, AI, machine learning, speech recognition. And so we we were kind of in the space of large language models and what eventually became Chat GBT before it was cool. And and we kind of we kind of had the early view in academia of like where things were heading, right? So you got it, you get a glimpse in the research labs of what's gonna actually become kind of mainstream in the coming you know years or decades sometimes. It definitely happened a lot faster than we predicted, but we started the company because we kind of felt that there was gonna be a transformation, a revolution, if you will, for what was possible with this technology. And we decided to go vertically focused, partly because you know we we happened to have some early traction with some credit unions and community banks. Based in Boston, there was uh an incubator here called DCU Innovation Center that's like you know one of the biggest credit unions. And so we kind of got exposed to this market pretty early. But we also thought that it, you know, it's better sometimes to be the best at something than to be like good at a lot of things. And with how quickly we were kind of seeing technology moving and AI moving, we thought, hey, it's it's probably wise for us to go really deep than try to go broad. And so we started building conversational AI, more like customer-facing AI assistance, which during COVID became like a really smart move because we saw a lot of uh traction during the pandemic when we had branch closures and call volume spiked. Um, and then we ended up expanding our product suite into a lot of employee-facing use cases too. And that really happened because, you know, with ChatGBT and LMs getting better, like we we obviously wanted to adopt those technologies, but um, our customers were not so excited to deploy those customer-facing at that point to their customers, right? And so um, you know, we talk about like the cutting edge AI and it's like, hey, are you ready for us to like deploy this in your member-facing voice AI? And they're like, uh no, we don't, we don't want to do that yet. And so we we figured that by getting into employee-facing use cases, like knowledge training and all these other use cases, we could almost get our customers to like start adopting the newest technologies, but also give ourselves the ability to grow into new product surface areas, but also to kind of like do our own testing of these cutting edge models in a safer environment and then progress them into our customer member-facing products as well. All that to say now, you know, we have like seven or eight different products at Poshna, depending on how you skin the cat. And we work with about 150 uh financial institutions today across across the country.

SPEAKER_01 3:26

I mean this in a complimentary way, but you're like a walking cliche, right? A couple of MIT students get in on a technology early, build the startup, and scale it in the market. So you said something that I just couldn't let go. You said you were into the L excuse me, the LLM space before it was cool. And it seems like it's becoming not cool again. Um, we're having this conversation on the heels of its graduation season, and a lot of colleges have had people that have gone up and given the uh the address, and they're getting booed when they mention AI. I wonder, yeah, do you have any kind of commentary on what's going on in the zeitgeist right now that is giving AI a bad name? And what do you think it means for um adoption?

SPEAKER_02 4:08

Yeah, I mean, one thing to just acknowledge is just the speed of change that's happening right now. You know, we're we're seeing change uh in an order of magnitude that was like incomprehensible not that long ago, but today we become kind of used to it, right? Like when a new model drops every few weeks now and it's like benchmark-wise, you know, improving and improving and improving on top of prior generations just a few weeks before. It's not that surprising anymore. You know, like we've we just like become accustomed to these things. Um, whereas like two years ago, that was like, you know, it would have been quite shocking if we saw this pace today. And I don't see that pace necessarily changing. And I think that creates a lot of anxiety in people, um, especially in new grads. I mean, you talk about like college graduations and stuff. Like, we know a lot of new grads right now are having a hard time finding jobs. And I think it's because a lot of companies are able to solve a lot of the entry-level types of use cases with AI, which was a lot of the reason that you would hire new grads. You know, um, you know, in software engineering programming, like Cloud Code, Codex from OpenAI, a lot of these models are arguably better than the intern you could hire or the new grad you can hire. So I think that that definitely causes like that sentiment. I think that sentiment might only get worse as AI continues to like solve harder and harder problems in the enterprise.

SPEAKER_01 5:18

You know, the people listening to this, they may be in a position where they or someone in their org is making decisions about hiring that entry-level developer. Is it short-sighted to have that idea that I can replace that skill set with a better skill set at a le you know lesser cost, presumably, right? We don't really know how token usage costs are gonna progress over time. But is it short-sighted that they're not making the investment in the human capital that's gonna pay off in the long run?

SPEAKER_02 5:48

Uh I think maybe, yeah. I mean, uh you could argue that like without hiring the young folks today, you're not gonna have enough like experienced labor tomorrow, right? And I think that argument is fair, where I think it could potentially break down is that you're not accounting for like the continued growth of the models themselves. Um, I mean, today it's like a lot of new grad kind of like, you know, uh level one, level two engineering work that's like fairly automatable or like confidently automatable. You still want humans to code review. We still have a lot of like human-driven processes, CI CD review reviews, all that stuff. But AI is part of the review processes now. When you when you do code review pipelines, you often have AI agents doing the code review alongside the senior software engineer who's doing a sanity check as well. I don't I don't think it's like that big of a leak to say that senior engineers and even like really solid engineers in the future are going to be replaced too. But I think it changes the equation a little bit. Like it's not about like who writes the best code or who can you know solve this like complex problem. I think it's a lot about like what defines a good product now. The the sense of good design, the sense of like, you know, creativity. Like AI, AI is good at pattern matching and it's it's doing some novel thinking too, but a lot of it is like pattern matching. And you you if you want to build something like really unique and impressive, you often really you can't really do that just by pattern matching, right? You have to like make some unique observation or do something kind of novel. Um and I think that's that's kind of where humanity is like especially good at doing that today. Like the creativity, the empathy. Like AI kind of can solve problems, but it doesn't solve problems through the empathy factor because it doesn't really empathize, right? So uh like if I'm a new grad today, I think you know, you kind of think about like, yeah, the the traditional software engineer is maybe not the thing I would invest in. Maybe you invest in this like new concept of like a one-man product army, if you will, right? Like how do you develop the skill set to have like good design instincts, good product development instincts, good engineering instincts? I think that is like really valuable today. Um it's almost like you want to hire people that could start their own companies. Um, and you want to bring that skill set into your company, right? And I think I think that's where the bar is changing today.

SPEAKER_01 7:48

Yeah, that's interesting because one of the powerful things about AI is you know how it can help you take a startup to market so quickly, right? And so you can play that out at the product level as opposed to the business startup level. So I guess what you're advocating for is a bunch of product owners that are entrepreneurial in their mindset using AI as the tools, but still these grads need exposure to the AI and to the tool set, right?

SPEAKER_02 8:14

And they need exposure into the industries too, right? Like if you're if you're building for banking, it's I mean, we I did this as a new grad and I learned a lot over a lot of hard lessons and a lot of time. And I I I'd say I'm better maybe now than I was when I started, but like you have a learning curve in these real world industries, right? You you have a new grad coming out who like doesn't quite understand um how you know the intricacies of a particular industry work. And yet, you know, you you join a company that's building for that industry, um, you know, you have to learn that stuff. Um but I yeah, I I think the new grads have one advantage, which is like they're they're probably more comfortable with change now, uh, because they've lived through it. And they're probably more comfortable with uh, you know, like breaking out of molds in a way, right? Like like I think more experienced software engineers are kind of accustomed to like their role having been software engineering. Somebody else was doing product management, somebody else was doing QA, somebody else was doing design. Right. I think new grads now, and maybe the educational systems are evolving or adapting to this too. I think they're probably getting exposure to that whole that whole spectrum, right? And so yeah, you can you can have AI write great code, but you can also have AI suggest like good product ideas, but you you still need somebody to make the decision. And I think that that decision capability is like the real like hard thing today, right? Like I would optimize for people that are intellectually pliable and can like learn these skills and all that, but that are also just good at making decisions. Yeah. Like taste is important.

SPEAKER_01 9:38

Yeah. Um, yeah, taste and and creativity, these are things that AI is never really going to, just by definition of what those things are, right? They're they're never never going to be able to capture that. I wonder also, is is the fact that AI now, and at least the models that we're talking about, is uh being probabilistic as opposed to deterministic, does that mean that it's got a ceiling in its current iteration of how far it can actually go in terms of replacing humans? You know, I think code is a good example where because it's probabilistic, it's it's making a prediction of what the right answer is and developing that code, but that's really where hallucinations come from. And you can't really have it hallucinating in your code. Things are going to break, right? I'm sure there's ways that you could set up adversarial models that check the other model or whatever, but just generally speaking, do you think there's kind of a cap there?

SPEAKER_02 10:29

I don't think so. I mean, are are humans not probabilistic models ourselves? Right? Yeah, for anything. You know what I mean? So I mean, I think you just have to have the right guardrails. Like I could, I could hire an intern that like tries to push really bad code and I would hopefully have a code review process where somebody else proofreads them. It's just the same idea with AI agents now. You you you have different models proofing the other, trying to prevent bad things from actually getting into the real world. But yeah, I I think I mean, I don't know. Like I think if you if you ask yourself like the the best AI models that you interact with today, are they are they um you know potentially coming up with different ways of saying something? Yeah, but then so does so does your so does your spouse or your best friend too. You know what I mean? So like um I think I think that's just the nature of thinking is is is somewhat probistic.

SPEAKER_01 11:12

Yeah, you're absolutely right. But also look at all the problems that we humans create because we're probabilistic thinkers, huh?

SPEAKER_02 11:19

Yeah. But I mean code code is deterministic by execution. And AI models can write code. So I think you can you can definitely lean into both. Like you can lean into the intelligent, probabilistic side when it's when it's warranted, and you can lean into the determinism of code when it's warranted. And you can probably solve like really complex things if you orchestrate those two really well. Right. Uh from what Posh does, for example, like you know, when we build voice agents or digital agents, uh we are orchestrating both. We have LLMs doing understanding and reasoning and decision making, but we also have code that's making sure that we're executing things deterministically when it's important to do so.

SPEAKER_01 11:58

Yeah. I'm gonna go back to something you said in you know towards the beginning of uh Posh, and you had customers that were not comfortable yet putting AI in front of their customers on the front line. Um you know, I think that we're seeing that that friction and those barriers start to fall, probably because um AI has become you know consumerized and and the consumer is getting more familiar and more comfortable with it. So businesses are more comfortable putting it in there. But I wonder from your perspective, like what what was that hesitation? Was it just because of the state of the technology and where it was at that point?

SPEAKER_02 12:41

Yeah, I mean, you new things are always a little scary. And I don't think financial institutions by nature are known to be trailblazers on these new technologies, right? They want to be a little bit kind of careful. And I I think I think that's probably because of regulatory pressure too. If you're in an unregulated industry, like the cost of being wrong may not be that high. Um, but the nature of what like FIs do and the industry that they're in, both regulat regulator-wise, but also like the nature of the actual task, the cost of being wrong can be quite high. And so I think that's that's good and bad, right? It's it's good in the sense that like you acknowledge that these are really sensitive things and you don't want to like you know rush into these things kind of haphazardly, but it's bad in the sense that like sometimes you you swing the pendulum too far in the direction of conservatism. And um, you know, you're almost in a situation where like yeah, like, you know, um people that are a little bit more willing to uh take risks, and I wouldn't really say it's a risk if you like construct it well, right? Like it's just applying a technology. But um those who are kind of earlier movers like can have a compounding advantage. Um and I think the risk that I am seeing potentially is in like fintechs uh and kind of the more tech forward DNA companies, like maybe being a bit more uh like bold in adopting these technologies and and having those competencies and stuff to do so. Um and a lot of the market, the you know, the traditional FI market, uh especially some of the smaller part of the market, uh are either skeptical still or just don't know where to start. And uh that that's that's risky. I think that's a little dangerous.

SPEAKER_01 14:10

You operate in a highly regulated industry, but it's also a high trust industry. And I think that we're at a time where our society is lacking some trust, especially in institutions, especially in financial institutions. You know, so you know, I I I can't help but think that there's gotta be something to bridge the gap between, you know, the consumer, uh the user, and the AI technology that allows for that trust, for them to trust the technology that they're talking to from a security perspective. But also I think there's just something kind of innate in us that we're good at picking up when something's a little bit off, especially when it comes to something emulating a human, right? So if I could just kind of deliver a little diatribe for a second, but in marketing, right, we have user user experience design. And in my mind, user experience design is is bridging the gap between uh a function of a technology or engineering that went into that technology and how humans behave. Interfacing with a computer is different, it's a different mobile uh modality than how we interact with other human beings. And so I feel like that allows us to compartmentalize that experience a little bit, and it doesn't set off our trust radars that this isn't a human experience that I'm having. But you know, by design, I think I mean, even in your tagline, it's service that he feels human and AI that works like a teammate. So you're trying to emulate human interaction, and I think that blurs the lines, and that might be a challenge to building trust. So I'm wondering how are you designing your AI in a way that allows for that trust and bridges the gap between that technology and human behavior?

SPEAKER_02 15:56

Yeah. And I think, I mean, you have to again think about like who the end user is. Like in some cases, the end user is like the actual customer member of the FI. And in other cases, the end user is actually the employee at the FI as well. Um, you know, we're seeing right now, I mean, the the larger institutions are still a little bit wary of like the actual end user facing stuff. Um, but like the like the customer, I mean, um, they're they're quite heavily adopting the employee facing. Like the idea of empowering humans, superpowering your humans, like that, that I think right now is like where 80% of the market is like most excited about AI. Um, we haven't yet seen like that that full embracement of like customer facing technologies. I think it's gonna happen though. I think it's inevitable. Um, it's just like that that crawl block run model. Um and when you when you think about like the human emulation, I mean, one, one, like we're focused on also like human empowerment. So like the the end user, like the customer might be talking to you as a human without realizing that you're being superpowered by an AI on your on your site, right? And so you kind of have the human in the middle. So like, yes, it is a human interaction for me from my perspective, but like you, you are quite enabled by AI on your side, right? And the other side of it is like when you're directly interacting with AI. I think today, like digital channel communication is like quite good, right? You use Chat GPT, you type with the Google search today is conversational now too, since since like last week. But it's it's one of those things where like the digital side is quite good. Voice is where like you get into that uncanny valley still, and especially like those video avatars. I don't think any of them today are like really at the point where they're fully like immersive, you know what I mean? Like they're they look good, and then you you try to like stress test their emotional responses and stuff, and they struggle to like you know emulate anger or other kind of emotions besides just like professional smiling. But you know, voice, voice is advancing quite well. We we are seeing a lot of advancements in the last year around like true like empathy, the ability to like extend that to other languages as well, and like uh, you know, the ability to also like you know, just just respond with the latency required to feel human-like and the intonations and stuff. So I I think I think we are in a period where like even that's gonna be quite quite good if it's like not arguably kind of already there now. But the hard part, I think, with with voice versus digital or like even the the video stuff is that you're right, like humans' bar for real or fake is like really high. Um and I think I think it's a hard problem if you're like trying to convince the end user that, oh, like this is actually a human. But I don't I don't think that's the goal, right? And even from a regulatory perspective, you're not really trying to fool them into thinking it's a human. In fact, you should probably be disclosing that it's an AI even if it was like really easy to easy or not, sorry, even if it was really hard to know that it was an AI, you should probably still disclose that it's an AI. Um but I think I think people just want like their problem solved in most situations, and they want to feel like the person that's solving their problem is smart. Like a lot of people's problems here are being solved with assistance like ChatGPT. Like like people are following ChatGPT's advice, maybe to an extreme, right? You know, in a in a case where it's actually like too much so like medical stuff. But for certain things, and this is especially true for banking and probably true for like, you know, healthcare and stuff too, for certain things, I think people would still want either a human answer or a human second opinion, right? I think that's fair. But like it is actually surprising, I think, how much people are starting to trust AI agents when they deem that AI agent as intelligent. Um I think the problem that the industry faced was like, you know, even a year or two ago, AI agents were not deemed intelligent. And so people just didn't like really give them a chance. But I think I think that's shifting a lot now to your point, because people are using products like Claude and ChatGPT in their daily lives, and they're like, oh wow, yeah, US is like probably smarter than like, you know, people I work with or like I am friends with, right? Like just in raw intelligence.

SPEAKER_01 19:31

I mean, definitely in raw intelligence. But you know, just as an example, coming out of the pandemic, Zoom usage had taken over, obviously, and and that's how people were keeping from getting lonely. They were connecting with family members and friends over Zoom. They had done some studies that basically you can't get that same fulfillment of a human interaction that you can through um, you know, over Zoom that you can in person, because you're you're missing little micro expressions, you're you're missing nuance that the video just doesn't pick up. And you know, I think to some extent they don't really know 100% why that is, but they know that you don't get that same fulfillment. So I think there's you're you're right, you should the approach shouldn't be let me try to fool these people, right? It should be disclosing and but and building trust in the competence of the model that you're talking to. Because for me personally, that's the challenge that I run into with a lot of these uh like a chatbot or a voice bot or something. I'm always trying to press zero to get out to a human being because if I'm calling customer service, it's because something's really messed up. And like I'll let a lot go. But if something is really messed up, that I need to talk to someone because I feel like it's a complex situation that requires some explanation. And I just don't trust that this thing is programmed in a way. But that's probably old thinking because it's not programmed in the sense that it used to be programmed, right? It's not a decision tree per se. So yeah, I think it's it's gonna be interesting to see how it plays out. But to go back to the original question, I think having a human-centered design approach is the right place to start for somebody that's implementing these kind of things, right? Because it shouldn't be purely about productivity or cost savings or efficiency or whatever. Uh, it shouldn't be about improving your metrics. It should be about create, in your space, creating a better customer experience.

SPEAKER_02 21:25

Yeah, or whoever the end user is. It doesn't have to be a customer. Yeah. I I I want to double click on something you said there, which is like you you called in because something was like really messed up, right? And so, you know, one just quick comment. Like, you know, a lot of people that we see calling have tried to self-solve that issue before. And, you know, voice, the voice channel is often the last resort for a lot of people. I think you see that, especially with the younger people, like don't want to don't want to call. You know what I mean? It's like uh I'd rather not call if I don't have to. Uh, you know, definitely don't want to go in person, usually, uh, but also certainly don't want to call. Um, and so they try to solve. Self-serve with things like search or uh you know digital assistance and stuff, you know, to a to an extent. So what on one hand, you know, you you can start to bring AI into those channels as well, like AI and those channels where like the the uncanny valley isn't like quite there with voice, where like technically you could have a voice agent that could solve your problem, but you may not give it a chance just because like you're like, eh, like, you know, it's clearly an AI, it's robotic sounding, whatever, right? Um, even if the AI technically was smart enough to solve your problem, like a lot of people may not just give it a chance just because they're like, I don't, I don't like AI, I've had bad experiences in the past, I'm gonna zero out. But many of those people we've seen are still using search bars. And almost everybody I know still uses Google Search, despite the fact that it's an AI tool now, right? And so um making search better, making self-service better, better on like non-voice channels, I think is a really smart move to just like you know, solve a lot of pain in the beginning, like where it's stemming from. And surprisingly, like a lot of our, well, not not necessarily our customers, but like the market in general, haven't like really been thinking about AI transformation in those like simpler channels. Like I think everybody thinks about like a voice AI, it's like you know, cool and sexy. But like everybody has a search bar. And and not many people I've seen have like really thought through like, hey, how do you make the search experience better? Because like people who call in have often tried to search first and didn't get the answer they wanted, or just like it didn't, it didn't do a great job of getting to the intricacies. Maybe they had a hard problem to your point. But yeah, that's that's like low-hanging fruit. And and search is like used a lot, right? Uh, probably arguably a lot more than like chat is used actually. And then the other thing that you said that was interesting was like, yeah, like maybe, maybe the AI can solve those problems now. And you've like maybe like you you give it a chance now than then maybe a year ago, or you, you know, like kind of ignoring past experiences because you're seeing the technology getting better. Um, or you give it a chance now because you had an experience with a different company, different industry where their AI was good, and you're like, oh, like this one retail company, like that was a pretty good experience. Maybe maybe my bank is good now too. Let me try that out. Um, and so we're seeing some of that too. But one thing we are seeing is like, you know, AI, AI doesn't have to solve everything, right? Um AI can know uh what it's good at solving and what it's not good at solving. LLMs are intelligent to understand the situation and then decide whether to react or not. And one way to not react is to recognize the things that are actually worth just connecting to, connecting you to a real person for. And it's not necessarily because the AI can't solve it, but it's because like whoever you know the higher power of this model is, whether that's the AI itself reflecting on its data or whether that's like the human overseeing the AI, has decided that this particular topic, despite the fact that AI could probably solve it, is one that it's better for a human to solve. And maybe maybe regulators decide that. Maybe regulators say, like, hey, I don't care how good the AI's financial advice is, we just don't want AI giving financial advice. Um and it's like like I'm recognizing the conversation that this is leading towards financial advice. I I, as an AI model, like could definitely give you great advice, but I'm just gonna connect you to somebody because I've been told that I can't do this, right? And so we're we're seeing this like concept of what we call uh what I call like mini agents or specialized agents. You know, it's this idea that you don't have to have this like overarching AI that does everything. You could say, hey, I'm gonna build like specialized AI that does these things, and then have this like router AI agent that recognizes what the person wants. And if it's something that we know one of the specialized agents can do, like sure, let the specialized agent do it. But if the if the person that you're interacting with is like not that keen to talk to you know an AI, or they're asking for something that like either we don't know how we don't have an agent to route to, or that you know, we've decided that for these things it's just better to have a human than then get it to a human. And so you might call and you might just it might just say, like, hey, thanks for calling. Like, please describe your situation so I can best figure out how to help you. And you might you might be totally okay doing that with an AI. It's like, okay, sure, I'll say what I want. Um, and then the and then you might just have a human answer and say, Yeah, hey, I can help you with that, or or an AI agent says, Hey, yeah, yeah, I'm I'm an AI agent that's designed to solve this problem. Let's solve this problem.

unknown 25:38

Right?

SPEAKER_02 25:39

You could be over voice or digital.

SPEAKER_01 25:41

Yeah, I think um the space that you're in in in financial services, somebody calling their bank or their credit union, sometimes that customer could be anxious about their finances, they could be ashamed about their finances. So knowing when and when not to deploy AI, I think is important. Same thing in the healthcare space, you know, that you don't want to hear that you've got cancer from an AI bot, right? So how how does in in practice, how do you go about determining what are the things that you can or should automate versus not? And I don't mean necessarily from a technical or like you were saying, a capability perspective, but more from a human perspective.

SPEAKER_02 26:19

Yeah. Well, I think I think it's both. I think it is the technical and the human side, but um, yeah, the the medical diagnosis, like maybe you don't want AI to do that because it's you know it's such an empathetic, empathy driven thing, right? For things like collections though, like you could argue both ways. You could say like like financial hardship is a very empathy-driven thing, but at the same time, people have this like feeling of embarrassment um and they might actually prefer to have an AI because no one's judging them, like they don't feel like the sense of being judged or whatever, right? I think, yeah, I think you have to think about like AI capability. And then I think the capability side is like, yeah, like first of all, like regulatory-wise, are we okay? Because if regulator says something, like it doesn't matter whether you know the technology is good enough or not. Um but when it comes to technology capabilities, can we design something that's reliable, reliably safe? Um, and then can we design it that's reliably a good UX as well, good user experience? And if both of those things are true, then the question is like, is it a good human experience too? Like, like forget like if a human, if a user engages, like is it a good flow? Sure. But like, should should an AI actually be having this conversation? I don't know if there is a right answer necessarily. I think it's like very case by case, might even be personal, right? Like the organization that's deploying the AI might have their preferences, but maybe the person using the AI has their own preferences too. And who's to say what's right or wrong? I think the best answer is is maybe choice.

SPEAKER_01 27:34

Yeah, and everybody's preferences are going to be different, right? It's it's really hard to account for that. Uh but yeah, I like the idea of of choice. And um, I think at least just having the awareness to consider that question before you go into designing how AI fits into your customer experience is um I don't know how to turn that off. I don't know why. But what is that? It's it's a Zoom call, but I've got all my do not disturbs on. I think it's tell them all man. I don't know. Kylie, we can leave this in. We can we can demonstrate what a dinosaur I am here. So anyway, just you know, I think if you're going to at least consider that question, then you're already a step ahead, right? You're having empathy for the people calling in and you're you're making some intentional decisions as opposed to just deploying something for other reasons. Uh we talked about disclosing that you're talking to or dealing with an AI up front. And I think there's I agree that's a the right practice in most cases, but there's two sides to that coin, right? Because on one hand, if you disclose it, some people are going to be skeptical immediately and already kind of on their toes. And that could prime them for a bad experience the rest of the way through. But on the other side, if it's not disclosed and they find out that it's an AI later, um, does that collapse the trust? So how do you approach making that decision and kind of designing how you just disclose the?

SPEAKER_02 29:05

Yeah. I mean, uh, this is again like where regulators are sometimes like just making the decision for you. And I think I think the the general obligatory posture right now is that you should be disclosing that it's an AI. Um, I don't know if different states and stuff will have their own like stuff, but I think I think the the general like majority consensus right now is that you're supposed to disclose. Yeah, I mean, I think that's just a good practice. Like as a as a human, and so thinking about this from like a the just the human lens, right, as a consumer, even if I know that even if I don't know that it's an AI, something might feel off for me if I realize later that it is and it never said so. Not that it's like, not that I would be like offended or something. It's just like it just feels disingenuous, right? And maybe it's just like, you know, the the broader climate right now that makes me feel that way. Maybe in like a few years I won't feel that way because like it's just gonna be common sense that like anybody you call us most likely AI. But yeah, right now, right now it feels like it would be, it would be the right thing to do to disclose. Yeah.

SPEAKER_01 29:54

The first personal experience that I had with a really good conversational AI was when I called my mobile phone carrier and the voice answered, and it was a very chipper, happy sounding voice, and it was like, hey Sean, what can I do for you? And I I just started talking to it. I didn't, you know, like I was there was something in me that was like, oh wow. But then I just started talking to it, and then I realized a little bit later, and I came away from that with a little bit of a chuckle and actually impressed at the experience and and a little bit delighted at the experience. Yeah. But as it's become more ubiquitous, I think what what I'm trying to sit with is the not knowing, right? You don't know when you're talking on a chat on a website or something, you you you don't always know until it's further down the conversation. Um, and in real life, if you can call social media media real life, but like when you're looking at you know posts on social media and you're looking in the comments section, sometimes it feels like all of it's AI, like just AI talking to AI bots, but you just you don't know. And so there's no point in interacting because it probably isn't a human, but you don't know if it's a human. So I think it's just that it's uh it's unsettling to be in this position now in our world where you can't determine who you're talking to.

SPEAKER_02 31:13

Yeah, you don't know whether media that was generated is is synthetic or a human design. I think in some cases we don't really care. We consume that content as entertainment. But yeah, when it's when it's like when the stakes get real, when it's like real decisions that impact your life, maybe maybe you you do you do care to know.

SPEAKER_01 31:29

Yeah. Yeah, if it's a cute cat video, I'm happy to accept I'll pretend it's real and move on with my life, right? But all right, so in that vein, how has people's expectations of their interactions with AI evolved and and where do you think we are today?

SPEAKER_02 31:49

Yeah. Um, I mean I think I think we have like mass mass adoption now. Like, you know, that wasn't necessarily the case a couple years ago. You know, I think it's it's weird now to like meet somebody who hasn't tried something like Chat GPT. Um it's certainly weird to meet somebody who hasn't used Google Search recently and seen that it's like you know AI powered answers and stuff. But I think I think people are starting to change their perceptions on like the capability of AI after having used these assistants, and that that's like that's where the anxiety and stuff comes in too, right? Um it's like, oh, this is this is awesome, but it's also really scary. Because then you start to question like, okay, you know, one, like what what makes me special as a human? Like, what am I bringing to the table? And I mean again, that's like empathy, the in-person stuff, the you know, the the things that I I'd say are go beyond just like simple workflows or simple answers. But then you get into like the harder stuff that's starting to get automated too, and you're starting to use these models and these consumer technologies, and you're like, wow, I'm seeing firsthand that it's like doing like the hard things too. I mean, early on, people were using it to like write poems and like you know, proof edit, like a proofread of copy, right? But now we're having it like create novel cocktail recipes. Here's what I have in the fridge, give me something that I can make with. And it does a really good job of like coming up with something, right? Uh and giving you like how to cook it and then the end result's like quite good. I think I think the thing that we're gonna start seeing like a lot of concern is when we when it goes beyond just like the um the white collar realm of stuff, right? And we're starting to see like where robotics is heading. Again, it's using a lot of the same breakthroughs that powered just the the white collar side, which is more like you know, an AI model behind a screen and a keyboard. But once it starts getting physical, I think, I think that'll be like like another big like milestone, you know. And we're I think we're starting to see that a little bit, but like we we're not really quite there yet. But yeah, the economic implications already are huge. And the economic implications are gonna continue to get bigger and bigger with the white collar side of these models like getting better. Um, and I think I think like once it gets into the physical world, like that's that's when things get like, you know, quite quite dramatic. But yeah, I don't know. I don't know what the timeline is is for stuff like that. Like that's it's hard to say.

SPEAKER_01 33:54

It's been kind of interesting to me to see how quickly something like Waymo or self-driving in a Tesla to the extent that it is self-driving, has been accepted by people. I mean, I I really that that's a situation where there are very much high stakes, and I I fully accept that humans are not good at driving, right? But m my intuition is that there are things that I know about driving that AI can't, right? I can make eye contact with somebody and see if they see me or if what their intentions are, right? I can look at this car that's beat up and has a a thousand dents on it and think this is someone I should probably stay away from, right? Or, you know, just my best experiences I can draw from. And I also realize that intuitive thinking is a heuristic and not a necessarily a rational way to move through the world. But for me, it's just something that that that might be more of a barrier for me to get into that car than it is to, you know, trust Claude to review my lease for my new apartment and have it call out anything that's weird.

SPEAKER_02 34:57

Right. Yeah. Yeah, the the self-driving car stuff, that's a good point. Um, that's where regulators are also again like kind of impacting what's possible, like whether things can go on highways or where it can be used and stuff. And then you have those situations right now where like you still see like the smartest models, like the ones that are powering Waymo's are getting stuck in cul de sacks and going in circles or you know, encroaching into like a police scene, which they shouldn't be doing. I suppose uh yeah, yeah. I mean it it's it's it's funny too, because like 99% of the time you can be amazing, and then the 1% of the time, like you know, it's kind of like the LLMs, you know, before I think we we found ways to mitigate like hallucinations and stuff, but the one percent is is still real, right? It's an edge case, but it's still real. But like once that last tail end of stuff gets solved, I think I think that's when like you see like a dramatic explosion of adoption and stuff too, because it's like, oh now it's like it's actually real, right? Like you didn't see like enterprises like adopting AI models as quickly as they are right now, even just a couple years ago, because even a couple of years ago, that one or two percent issue was like still big enough where it blocked like real adoption, but today that's like dramatically less, or you have mitigation approaches, and now it's like blowing up an enterprise. Um and then I think I think the same thing will happen in the physical world too.

SPEAKER_01 36:08

Yeah, did you see the uh the Waymo DDoS attack that someone did on a uh a neighborhood? They ordered a whole bunch of Waymo's into one block and yes, yeah. It's kind of funny. Yeah, it's hilarious. But um, you know, I I have a buddy from high school that's uh a truck driver and he drives into, you know, an 18-wheeler and he drives into Manhattan pretty frequently, uh sometimes two, three times a day. And I just I can imagine, you know, an AI self-driving truck doing long haul from you know here to, you know, I'm in New Jersey, you know, from Jersey to the Midwest or something, but I cannot imagine it trying to navigate through New York traffic. I feel like it would it would be too cautious. You have to drive like a maniac if you're gonna make it through driving in New York City, right?

SPEAKER_02 36:56

Well, that's why they have people now um like letting the letting the truck do the long haul by itself, but then once it gets into like the hard part, somebody remotes in with like a VR headset and actually takes control of the steering wheel remotely. Um, which is like which is human in the loop, you know? It's kind of funny. It's like it's like it's like the concept of uh a virtual assistant escalating to a human when like the conversation gets hard. It's like basically what a self-driving truck is doing too.

SPEAKER_01 37:20

Yeah.

SPEAKER_02 37:21

Yeah.

SPEAKER_01 37:22

It's wild. Wait, what other, you know, I guess we're taking a a bit of a tangent here, but you know, what other interesting use cases do you see that you think are gonna you know start popping up in everyday life?

SPEAKER_02 37:34

Yeah, I mean, I think we're starting to see like a lot of people becoming entrepreneurs that were hindered by the inability to like build an app or a website, right? Like that's becoming way more accessible now. Um so like I think we'll see like a lot of you know people like pursuing passion projects, you know, um even if it's like a side side little business, um, I think we're gonna see more people like doing that. Uh ideas, ideas are now uh actually even more valuable, right? Because the execution of it is becoming more accessible. Uh and then you're I think you're gonna start to see like more personalization and more proactiveness, right? Uh I think a lot of people think about like, oh, AI, AI is like so good for coding and customer service and stuff. Somebody is like triggering that. Like a person is prompting the AI to go write a piece of code, or a person is asking for something and then the AI responds back. I think we're trying to see the shift now where like the AI itself is like kind of always on. That's like what we saw with like ClaudBot and some of these things that are happening is you just like have it consistent, continuously running on servers. It's expensive as heck, and like these models are getting like cost effective and computes you know becoming more accessible too. But you need you need to solve that cost problem to have like this always on AI that's just thinking about you. Your own AI agent in Netflix that's thinking about you all the time, that kind of already exists and it gives you recommendations. But what if we took that to like other parts of our lives? Then you get into like you know, proactive service. Like as soon as something happens or doesn't happen, you're you have your own AI agents that's like, hey, by the way, uh I went ahead and did this, or you want me to do this real quick, just give me an approval and I'll do it for you. So just preventing problems. And then, you know, from like a coding standpoint, it's like, you know, I'm able to like go and write custom software for an individual, right? Now like companies build software for like a large number of users, you know, what if software was built for an individual user, like tailored to them, their preferences, whatever that is. Um, that gets really interesting because software at that point is just an application. Um and you you don't know, like it's it's so hard to it's so hard to forecast this stuff because like it's dramatically different than what we're used to today. But like if people do just like really become loyal to a virtual assistant, like their their ChatGPT or their cloud instance, um, you can imagine like that is the application for everything that they work with, right? Um either that agent is talking to other other agents to do stuff like being an Airbnb or whatever, which is kind of happening already. Um, or that agent is serving user interfaces as well on like a micro level, right? Um and I I think I think ChatGPT is trying to do that with like PFM now with their you know financial literacy, financial wellness, stuff that they've been rolling out. We'll see where that goes. Again, I think a lot of the database systems of the customer data platforms that power are not so inclined to like let ChatGPT and Cloud just like become the interface for all the hard work they've done over the years. But it's it's interesting to think about, right? Like the the idea that like, yeah, my my ChatGPT serves up mini software applications for me that allow me to like have some GUI control over something I'm doing if it's not just conversational. Um if you if you need GUI control at all, right? Um and then everything else is just is just like my my AI agent is handling for me and I don't have to worry about like remembering passwords to all these applications or logging into different stuff. I just I just have it orchestrate everything. Um I mean that's that's a massive opportunity, I think. But um yeah, I I get excited about the NF1. Like can products be built for the NF1? And that could extend to movies and video games, you know, and books. Like it's it's crazy. Like the the the degree of personalization, but that can also exacerbate loneliness too. Like imagine if I watch something that's designed just for me to love, and then I can't share it with somebody else because like literally is designed for me to consume, you know, and and somebody else may not have that same reaction.

SPEAKER_01 41:03

Yeah, it's hard to imagine that society is ready for something like that. I mean, it it's social media has been such a how do I want to say it delicately? I mean, it's it's just been not a great experiment in what happens when you introduce technology in an unintentional way and just let humans run rampant with it, right? And I would say the better angels don't come out when you do that.

SPEAKER_00 41:29

Yeah.

SPEAKER_01 41:29

So it's a little scary to think about how and already you're starting to see younger generations are using the primary use case for you know the the younger generations for AI right now is is friendship, for lack of a better term. You know. Um whereas people in my generation, Gen X, we're using it for search, like you were talking about before. I mean, I I I rarely go to Google anymore. I ask Claude because I can create a whole bunch of context and I can really kind of dial it in. It just seems to understand more. Uh, I feel like it's a better experience than what I'm getting with Gemini when I'm you know trying to do that process as well. So, and it knows me because I use it as a personal assistant, you know, just the way that you were talking about, right? It's just right now it doesn't act for me. I don't I haven't built agents behind it, right? Right. Because, like I said before, I'm a dinosaur, but eventually uh it'll be so easy to build an agent that someone like me can do it. So I was talking to a guy named Ed Kiesling, who's the chief AI officer of progress.com, and he's gonna be on the podcast. We had a planning call, and one of the things that we were talking about, you kind of brought up, which was what's the future of SaaS? Because it seems like something like Claude could very easily be the interface that is plugged into and integrated with all these other applications that you use. And the example I gave him was in Amplix here, we're rolling out AI slowly. We've got use cases identified and a small group of people that are uh you know starting to work within the enterprise claude instance that we have. And so it's connected to our CRM, it's connected to Power BI, it's connected to all this stuff. People don't go into Power BI like they were to create reports. They're not going into the CRM like they were to create reports, they're doing it all through Claude and it's creating these you know highly customized and very specific reports that we weren't really able to get to. And and it just it's an easier interface to just have that, you know, conversation as opposed to hunting around in a GUI. What do you think is gonna happen with SAS?

SPEAKER_02 43:36

Yeah, um I think I think if you're if you're literally just like a database, right? If that's like if you can like distill your SAS product down to like just a database with some some like designs on top of it, I think that's that's harder to like justify. Unless that database is really hard to like action on from a from a different intelligent system, or unless there's like, you know, a lot of like workflows and stuff that can only be done in in some unique way. Otherwise it's like it's like, yeah, a lot of a lot of SaaS today is just like building some buttons and drop downs and visuals that let somebody just make sense of data that's sitting behind the scenes. And yeah, I mean that's to your point. Like, you know, that database could itself just be accessible by an AI agent that can that can do the same thing conversationally. Right. Yeah. I I mean the the the concept of SaaS in that sense, I think yeah, that is that is something that like will probably change a lot. And I think all these companies are kind of figure out like, hey, what is what is our like existential like answer now. But if you're if you're doing something that like uh like is it just goes beyond the database itself, like the data itself. Like for example, you know, you may not trust even a smart model today or you may not want to trust a smart model today to execute like sensitive workflows. Right. And so that might be something that like is kind of defensible still. Or maybe your tool itself has its own like layer of memory too. Like the reason you like Claude so much maybe and you wouldn't switch to Gemini necessarily even if Gemini's newest model is like you know let's say better than Claude's latest model that you're running on, you may not switch because Claude has developed an understanding of you. And it would be really hard for you to like either transport that understanding of you to Gemini or you wouldn't want to go through that whole learning curve of Gemini again too. And so I think the the concept of like personalization or this idea of memory in applications becomes a very defensible proposition too. Right. Like the system that learns you. And it's more than like is that data in a database? Sure. But it's not something that is just easily like extractable and and like queried, you know what I mean? Or maybe maybe it's not right. But the things that I think are are really quite defensible today is if you can create network effects. Right. So like two sided marketplaces where there's human trust on both sides is hard for an AI to just come in and take over. So like I think there are a lot of um SaaS SaaS esque things out there that are still defensible but there's certainly a lot of SaaS to your point that yeah I don't know like it could just be Claude hitting hitting the same database.

SPEAKER_01 46:03

Yeah. I was thinking that the current state right now is one of integration and interoperability and open source and I wonder if a company like let's say Salesforce what if they wall off their ecosystem to other AI models implement their own model, right? That would be and that's probably not a good state to be in in this kind of feudal state of software where everyone's walling off their own products, right? So you know hopefully the incentives are to to remain open and remain connected as opposed to reduce that.

SPEAKER_02 46:37

Yeah. Yeah I mean I I think you have to think about these AI agents as just like they're all like invocations of really smart people that can do work, right? That's that's like a good analogy. And I think the way people are trying to monetize their platforms is changing in that lens. So it's not about like how many human seats am I serving? And maybe it's like okay how many how many agent runs am I allowing to hit my system right? And that's more like a volume based pricing or consumption based pricing than just like a seat-based pricing. So like we're trying to see a lot of like changes in how companies are thinking about pricing their services. So on one hand it's the defensibility like is my database itself like is that is that just like you know does that ruin ruin the nature of my SaaS app like you know if someone can just transport that data and then an AI agent can reason on it like is my SaaS app even like really worth anything? But um I think if if you say like yeah there is there is value in your data or there is value in something that you have it could be data could be workful whatever that is I think the user user is more likely going to be the AI agent than than the human right volume wise and just like you know that where things are heading and how does that impact monetization for companies?

SPEAKER_00 47:51

Yeah.

SPEAKER_01 47:52

Make a hard left turn here like I said earlier the second half of your tagline is AI that works like a teammate. So that has a very human connotation to me. What do you mean by that?

SPEAKER_02 48:04

Yeah um well on one hand we we have that motto that both sides of that motto to imply that like hey it's both it's both customer facing and employee facing right like we are we are focused on automating the automatable things that are both customer facing and and employee facing. But what that means is that like for things that are not automatable or shouldn't be automated or can't be automated because like regulatory stuff, you still need people for those things. You want to we want to like empower those people to do those things too. And those are things that people should be doing like the easy stuff like you know the definition of easy is changing too because what was hard is becoming easy and that's going to keep happening. But the stuff that's still not easy is where I think like the human factor is so valuable. But even for that like you want a human to have the AI superpower right the AI empowerment. And that's where we think about things at Posh like how do we get a human from base level to expert level right like in that the you see this when you hire somebody new how do you ramp them up we we have a product that helps with training right conversational training and situational training. We have an AI agent assist knowledge solution that helps with people on the job to get access information quickly. We're starting to allow people to automate workflows within that product as well. So like instead of having to go learn how to use some other system, it's like the AI agent for the employee on behalf of the customer, right? So I'm talking to you as a you're the customer in this case and I'm the employee instead of instead of like having to go like navigate some system to do something on your behalf, I can just ask my AI agent to do that. Maybe you could have asked your AI agent to do it too but you for whatever reason either you didn't or we couldn't support that or whatever. And but I have an AI agent that's allowed to do that. And then and then the other thing that we're looking at right now that's really important a couple of things one is one is around like expanding the the superpower of language translation. You know things like Google Translate and stuff like you you say something into the model and then you let it speak out what you said. But we're seeing now like a lot of animation in in real time AI. So like as I'm speaking right now you could be hearing me speak in Spanish on your side or you maybe could even see my mouth actually in real time like looking like I'm saying those words in Spanish too, right? That's that's that's crazy. But uh the the main thing for us too is like if we're empowering people with tools, how do we know that they're actually using it? And how does the how does the organization that's like you know paying posh or like you know investing in these tools, how did they measure the ROI and know that they're using it? So we are we are also like leaning pretty heavily into AI powered QA, right? Quality assurance and stuff too, which is like hey we we trained you in a simulated context and we like got you to feel confident. But now how do we measure whether you're you're applying those skills to the real world or we gave you this AI knowledge assistant or agent assist type tool. How do we know whether our KPIs are actually being positively impacted at scale in the real world just because we gave you that tool like there's no guarantee that you're actually using it the right way. Right. So we're trying to see all these like agentic feedback loops between products and between like the different tools people use, but also the life cycle of like this this customer service agent, right? And how do we how do we use the shared brain, if you will, or like the communication between these applications or agents to uh like really, really not only superpower the employee, but also to give the the organization confidence that it's working and they can then compound that further.

SPEAKER_01 51:12

I hosted a fireside chat with Gary Sorentino, the global CIO for Zoom a couple of months ago in Philadelphia for a sim event. And one of the topics that we touched on uh he likes to talk about EX plus CX plus DX equals TX. And what he means is employee experience plus customer experience plus digital experience equals total experience. And people in the room full of CIOs uh a lot of light bulbs went off there right so my question is when you are engaging with technology leaders and you start talking about this teammate and the employee experience do you get the sense that most of them have thought about it from that perspective or do you get the sense that this is kind of a new concept?

SPEAKER_02 51:59

I mean I think it's I think it's a new concept for a lot of people just because again like this some of this stuff wasn't wasn't available a year ago.

unknown 52:06

Right.

SPEAKER_02 52:06

Some of the stuff just wasn't like I mean you could have you could have said yeah obviously if you empower your employees it's going to impact our CSAT and NPS in a positive way. Like yeah that makes sense but the specific approaches or applications that are possible today like just weren't possible yesterday. And so I think even if people like understand like the the philosophy of what you know they should do or are doing the way to apply it or actually practice it like that's changing like at a crazy pace. And so yeah I agree with that. I like that EX plus CX plus uh DX equals TX. That's that's nice. I think yeah that's a good that's a good like North Star like philosophy, but the execution of it is like is going to constantly be changing.

SPEAKER_01 52:43

Yeah. What are you seeing that most of the technology leaders that you're engaged with are doing wrong and how would you recommend that you know the people listening to this start to address that in their own organizations today?

SPEAKER_02 52:55

Yeah that's a good question. I mean I I think I think a lot of people are doing stuff right too. Like I think the idea of like starting with employees um and like you know the the crawl walk run mentality is it's like it's it's very it's very smart. I think that's that's a good that's a good way to think about things. Well I alluded to this before but like I think a lot of people also like conflate this this concept of like oh AI has to be all encompassing for it to really like like check the box of you know agentic AI or whatever. It's like that's not true. Like you don't you don't have to go like replace your entire IBR to get the value of AI. You could start by just handling after hours account lockouts for example and solve a like a really meaningful high ROI like arguably high volume problem that humans are okay having an AI help them for because they're like you know they just need somebody to help them unlock their account and reset their password. And and you know it's it's a good it's a good initial use case to get confidence, but it's not boiling the ocean, right? And I think I think we're trying to see some people like come to that realization more now of like like oh like we can we can actually start small even with customer facing AI. Like yeah of course you don't have to like do the whole thing at once. The build versus buy stuff I think we see a lot of that tension right now too we didn't really talk about that too much on this on this session yet but um people like definitely underestimate how hard it is to build. And they also they also overestimate how hard it is to build at the same time like you can you can get like a really great version one with uh with AI assisted coding. And that's that's great. Like in a in a weekend you know you see people on LinkedIn like hey this is what I built over the weekend and like three years ago I would have been like mind blagged oh my gosh like this this guy like wrote all this code in a weekend that's amazing. Today it's like oh yeah everybody's doing that. Everybody's like has these like these weekend projects using you know take your AI coding tool and you know coming up at some like web page or something. It's great. I think people don't understand that they can prototype stuff that quickly like people are trying to realize that. But a lot of people don't realize how hard it is to actually like take that into a real production long-term maintained solution. And you know we're seeing a lot of like you know regional larger uh customers and and just players in the space in the industry uh and I mean like banks and stuff that um are building like in-house teams of engineers and and AI experts, right? And it's like, oh we're gonna do this and that and that. Let's say great. Uh when somebody builds this or that and that, uh who's maintaining it? Right? Because is the person that's building it then moving on to the next thing and the next thing or are they staying and like actually making sure that given how quickly things are changing, that what they built three weeks ago is still state of the art. Because you know in cycles today where like three, three month old things are becoming outdated arguably, how do you know that by building in-house you're actually giving your organization the best of what's possible? You're probably not unless you have the ability to focus or unless you get to the point where you're running always on AI agents that are doing it for you.

SPEAKER_01 55:31

Yeah. Is there, you know, in your space at least are there some things that you say definitely build versus definitely buy?

SPEAKER_02 55:39

Yeah I was talking to uh I was talking to like a pretty large bank CIO about this actually yesterday and um I I forget like who said this phrase so like I'll have to like look at it afterwards or we'll attribute it afterwards. But it's this concept of like um uh you know um build what is actually gonna differentiate you but buy what's gonna accelerate you right um it might it might have been like Jamie Diamond or something I I don't know exactly who it was but um it's like yeah if you're if you're gonna if you're gonna like for the most for the most part you're just getting something you just want something to accelerate you. Right. Like it's it's like a tool that helps you move faster, make a decision better, whatever. But it's not the core of what your business is about that is helping you win versus comp competitors that's like creating that like real deep differentiation against like the people that you're you're either serving or competing against right. And and the guidance there is like yeah like ask yourself like is this is this really something that's going to be a differentiable reason why we win? And if the answer is no, then it's probably not worth building. As tempting as it is right because you might you might be like oh I could I could easily just like you know vibe code this you know replacement to the SaaS tool that we paid 20 grand a year for. Like yeah yeah you can save 20 grand. And maybe that's depending on where you are or like you know what other opportunity cost things you have and all that maybe that's not a bad decision. But like like just don't underestimate the maintenance of these things right because anytime you make a decision to build something the building it in the first place is maybe 3% of that total effort of the equation like 97% hasn't happened yet. The 97% is the long-term maintenance evolution of these things, right? Yeah. And that's a good model I think that build what differentiates buy what accelerates. Yeah I love that that's uh very marketable I'm gonna attribute that to you and also the whole office thing that says car and by you know JB whoever that was.

SPEAKER_01 57:29

You know you you said something about oh we're gonna build this and then is it still relevant in three months or is it already outdated in three months? And I can't help but think that with the ease of building with AI and also deploying AI in in discrete tasks or in in places in the the enterprise stack, is there a danger here that people are going to start to accumulate the AI version of technical debt?

SPEAKER_02 57:56

Yeah. Yeah I think so. And again like it's hard to it's hard to say like whether that's really a big deal or not because again you'll have more AI agents that you can then spawn to help you solve that technical debt. But yeah, I mean it's it's like it's very easy to overcomplicate things, right? Like I think people are really good at doing that. So like you could you could imagine like you start with something that's really simple to solve for and like oh I can just like do this myself. I don't need this product for it. And then you start getting running into edge case after edge case and it's expanding the surface area, expanding the surface area. And pretty soon like you're in like a pretty complicated situation because you've invested a lot of thought and time, maybe, maybe not like you know as much time as you would have had to do like a few years ago, but still like substantial amount of thinking and time to get to this point. And then you're like, oh actually maybe maybe I shouldn't have done this. And we're seeing this a lot with our customers by the way we have quite a few customers who spent like you know a year or two trying to in-house something and then came to the same conclusion that like yeah you know we we solved 80% of the problem quite well. But is it 20% of the problem that like still gets us right? And it's just like is it really in our like best interest from a differentiation standpoint like you talked about to continue to invest to solve this 20% or not? And I think a lot of people realize that like it's not in their best interest and it's better to go like just partner with the company that that actually it is in their best interest that company's interest to do that. That's what that company's purpose is but your purpose is not to solve this problem. And so yeah I mean I we've we've seen that quite a bit we've seen that quite a bit where we've had customers come to us like essentially pre-sold on the use case because they've they've proven it themselves. Now they're just looking for the right partner to help them actually long term scale it.

SPEAKER_01 59:23

Now if you are going the buy route for certain things, I wonder how do you avoid vendor lock in?

SPEAKER_02 59:28

Yeah. I think that's really important. I like that like you have model optionality right now, you know, and it's it's keeping each other on their like everybody's on their toes like you know all the all the big model builders because they know how easy it is to switch models. You just it's literally one line to switch whichever you know vendor you're using in your code base. Yeah I think I think the thing that we're seeing in the market in our market too is the the the longer tail of smaller players are kind of in a disadvantage in this situation because like they don't have the luxury of having too many partners for things. It's actually easier for them to consolidate and it's really tempting to consolidate because like the price tag seems better and it's like it's like one less thing to worry about like you know like less vendor due diligence and all that stuff. And it it feels really good initially like oh I have this partner that can do everything. But now you're kind of dependent on that partner to do everything um and innovate on everything. You know you just yeah like you're you're you're like super tied to something outside of your control and that that's risky right. I think what's nice about having optionality despite some of the headaches that come with it because you have to like you know constantly be thinking about different things is that you do have the ability to be best of breed where it matters and you do have the ability to also kind of go more modular in your strategy if you want. Right. And you're less you're less like dependent on like some third party entity like kind of determining in a lot of ways your your future success. That's always risky to like have that dependency. Even if you like really trust them and love somebody like that company might scale and start getting larger customers and their attention is now going somewhere else. Like you never know. So yeah I think I think that's that is a big issue. The the the vendor lock in is like is a real is a real threat I think to the institution success. And they're probably trying to engineer that lock-in as best as they can right it's in their yeah I mean look at look at the big publicly traded companies in the banking space like their whole business is around lock-in. Yeah yeah yeah and they're they're designing contracts to like make it really hard to not be locked in. Right.

SPEAKER_01 1:01:23

A while back in the conversation you talked about AI supercharging employees and in my experience uh in my career I've rolled out software enterprise software to large businesses I've been part of that and what I've seen is um success of that software completely depends on adoption. Can can you get people behind it can get them to see the shared vision can you get them to to push through the friction of the learning curve and the uncomfortable feeling of doing something new and make that part of their day-to-day you know the classic example that I've had in my career is rolling a new CRM out to salespeople and they're just like we're not doing this. I've seen a a an Oracle CRM, you know, like a a seven figure rollout get rolled back because salespeople just didn't adopt it. So I wonder how do you go about priming your employees to adopt this technology so that they actually can be supercharged.

SPEAKER_02 1:02:27

Yeah I mean even even on the customer side right like even if AI gets really good and it can solve your problem if you don't give it a chance that it didn't solve your problem. And if it could have um same thing for employees it's like the you know lead the horse to water concept right and that's a that's a real cultural thing. I mean I don't I don't know if there's a right answer. You know even even in really good software companies some of the best engineers objectively are the ones that are also like the most skeptical of AI tools. And like I don't I don't know like sometimes you just see the click in people they're like oh like oh yeah like yeah I tried it a few months ago and it just it just wasn't that impressive and I kind of like had this like really like stronger version to that from my first impression. But I'm I'm coming back at it you know months later and like wow this is actually good now. Right. So I think you kind of need the person to also be willing to like periodically like give the give these tools a chance or give these applications a chance because they are changing at that pace. And I think people don't quite realize that compared to like you know anything else we're we've been used to in the history of our lives the pace of change right now is just remarkable.

SPEAKER_01 1:03:24

Yeah it absolutely is. So to that end, with the pace of change being what it is and the state of technology right now, AI specifically being so fluid, what's a piece of actionable advice that you would give to the listeners today that they should start thinking about or doing in their enterprise to be ready to have a successful AI deployment?

SPEAKER_02 1:03:45

Yeah. I think one of the biggest things you can do as from a cultural standpoint is to like promote experimentation. Right. And for regulated industries I think that that can be all scary. You can do that in a safe way. You can figure out how to how to do it in a way that's safe for you. But um you know people are people are using AI tools in their in their daily lives like most most of us are most employees of companies are um but they're they've they've been kind of like I think taught or have heard stories about like getting punished or getting in trouble for like doing stuff in the enterprise. Solve that problem first I think is important. Like give people the ability to think about how to use like good good AI tools right to actually help make their lives easier in in in uh in their work context. You know forcing somebody to use a tool may not work especially if they try to use that tool and it's not great or not great for what they're trying to do. Then they're just gonna like kind of be rubbed the wrong way and then and they're gonna feel like oh I'm being forced to use this like you know but letting people have choice, I think we talked about that right like the choice. It's like hey, you know, we're we're going to we're gonna be open to tools like if you have a tool that you think would be really good, um let's let's consider it. Like let's actually like take that through our better diligence process and give it a try or like have some kind of I I've called it in the past a slush fund. Like I I've I've called out that like a lot of our customers like budget in very strict annual cycles and they're making decisions usually in like July or August of the stuff they're gonna be doing in December of the next year. Like not even this December, like next December, right? And that's that's that's crazy to me. Like I don't even know what models coming out next week. I don't even know what like the future of white collar work is going to be like in six months. And you're determining now in like three months what's going to happen in next December. You know so I I think I think like a lot of that needs to change too like this is a culture of like innovation a culture of speed. We we often say that the best moat today is speed like against competition. But you know like if if you're able to have like a slush fund, like a like an evergreen innovation fund that like can fund these types of things and it doesn't have to be crazy either like think about the craziness of not doing it and falling behind. I think I think that's that's something that you can do too. So like encourage a culture of innovation and then financially think about how to actually enable this too.

SPEAKER_01 1:05:51

I think that's excellent advice. I love that I haven't heard anyone suggest that not explicitly when it comes to budget at least how would you recommend that they sell that To the C suite and to the board.

SPEAKER_02 1:06:02

I mean it's it's uh it's the cost of like not taking risks right now. The riskiest thing you can do right now is nothing. And I think people should like realize that. And then if you if you like say, hey, I want I want I want us to have an innovation budget. Um and maybe there's already a chief innovation officer and stuff. So like hopefully there's some level of budget for that. But it needs to be grassroots. I think for a company to really change and embrace like becoming more AI pilled, uh, it doesn't start top down, it has to start like bottoms up as well. And we we need green ICs to go evangelize tools and have the ability to actually like like fund fund that as well.

SPEAKER_00 1:06:35

Yeah.

SPEAKER_01 1:06:36

I think that's a good place to leave it. Uh Kern Cash Yepp, thank you so much for your time and expertise. This was a great conversation.

SPEAKER_02 1:06:42

Yeah, it's been it's been a lot of fun.