Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Deep Learning with PolyAI

Deep Learning with PolyAI artwork

071: Differentiation in a generative AI era

May 14, 202520 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Damien and Nikola discuss the brave initiatives taken by executives in the field of generative AI, their successes, and challenges. Nikola Mrkšić, CEO and co-founder of PolyAI, shares insights on differentiation in the generative AI space, the viability of build vs. buy decisions, and the importance of proof of concept in AI projects. The episode sheds light on the technological advancements and strategic decisions that help organizations capitalize on generative AI, with illustrative case studies and industry comparisons. Follow PolyAI on LinkedIn Watch this and other episodes of the Deep Learning pod on YouTube

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

NikolaA bunch of executives put a lot of money in a lot of different initiatives to hopefully capture, early mover advantage, right? Of using generative ai. Of course, most of it fails. That's only means that they like took on brave initiatives and that's good. And now the good news is a lot of those things are starting to pay off,

DamienHey everybody, and welcome to another episode of the Deep Learning with Poly AI podcast. I am joined as usual by Nicola Mesick, CEO, and co-founder of Poly ai. Hey, Nicola. What's up?

NikolaI'm good. I'm good. Just on the road and WeWork telephone booth. How? How are you doing?

DamienYeah. Where in the world is Nicolo Metric today?

NikolaI've been on the west coast for a month now yeah.

DamienOkay. Are you Pacific Northwest? Are you up around the mountains and everything or are you still staying in the bay? Are you in the Bay Area?

NikolaI'm mostly in the Bay Area, but I've been north, I've been southern everywhere.

DamienCool. Speaking of traveling, I just got back from enterprise Connect last week in Orlando, Florida. And we had a cool little booth there. It was a lot of fun, a lot of buzz. Everyone's still really super jazzed about generative AI and basically our market in conversational ai. One of the things that I heard a lot by people who came up and were chatting with us at the booth is, oh, I've heard of Poly ai, but what, why are you different? What makes you different? What makes you different? It was the same question over and over, and I know that's partially diligence by people who are just out on the road, like wanting to find out truly what would make someone unique.

DamienBut it got me thinking, Hey, in this new era of generative AI powered applications and solutions. A lot of those foundational models operate in a black box under of themselves, and so therefore the solutions above them are just as difficult to explain and overall explainability plummets. So I wanted to take a little time today to talk about how a vendor in this space, how a solution provider in this space has what they have to do to differentiate themselves in the market.

NikolaYeah. No, I think that's a really good question, right? I think that like in more software, there's like the whole like. Infra in this case. It's like the models is like philanthropic better than OpenAI or not. And for now, like there are meaningful differences between them. But long term, I think a lot of people expect that to become a commodity, an expensive one to trade, an expensive one to compete with. Then you have the whole layer of like open source on top trying to do whatever it always has, which is the commodity software. Stop any vendor from creating a monopoly or an oligopoly. But then as you move forward into applications, there's two schools of thought, one is this stuff is just so good and so expressive that we can create curated things for anyone just by saying, Hey.

NikolaDamian needs this kind of application to run PR and comms for a series C startup, right? Whereas Nicola needs, a workflow for sending emails and responding to a hundred people every day and he needs this kind of application. And one school of thought is that you'll just be able to prompt your way around like what you need and the model spits it out and there's no more like need to code. There's no differentiation in that. And I think that. The other school of thought is that this is enterprise software. Like anything else, it has workflow, it has it's certain like stickiness and like form factors that people learn to use themselves.

NikolaAnd because of that, it still has a certain stickiness where the network effects of people using it and the kind of the. The power of enough people using it together means that like vendor A gets to stay in, even though vendor B can replicate their capabilities, right? So thinking that applica application layer really depends on what kind of application it is, right? And whether there even be humans using it or not. But like there is just the standard boring enterprise stuff, which is like switching costs, change management, et cetera.

NikolaWe're like, sure, maybe someone else has built a comparable product, right? I don't think many people, even back in its glory days, thought that Facebook was something to behold. Technically. Sure. It was interesting and look, it used a lot of frontend technology in innovative ways, but it was clear that you could replicate it and China, Russia with their kind of close to form Sure. Internet did exactly that. The contact, and I forget the Chinese one. They're pretty much exact replicas of Facebook.

DamienSo then fundamentally, does it become an early branding exercise? What do you mean? Like you said, it's a Facebook. More or less created that concept of a social platform, maybe beyond what was happening in IRQ chat or a OL, whatever. And so in essence, they ushered in that era often duplicated, very regularly duplicated. Yeah. But because they were there first and established a rooted brand, they carried that through, or was it a pioneering victory for them?

NikolaI mean I think it's both. It was pioneering in that unlike, I dunno, what was it, like MySpace and a few others that never really hit that escape velocity. They built something that was good enough and once it was really hard for anyone to disrupt it, right? Sure. Then Instagram came in and they bought it, which is like great m and a, right? It's not, but that was like a form factor built for the phone rather than for the web. And it took like a change of the environment for Instagram to challenge Facebook in that way. So I think that's a good point to a lot of gen technologies. Again, you can flip it to ways like is it consumer or enterprise?

NikolaAnd I think that when it comes to like enterprise applications, it really is a matter of there is differentiation, how the product is put together, the requirements that it's built for. And the other part is you are also like when you pay. A software web vendor, seven digits, eight digits a year. Like you are buying more than just that piece of technology. You're buying a throat to choke, right? Whether it's a contractual or not. No, but really no, seriously. It's Hey, I'm paying you a lot of money to solve that problem.

NikolaWe had a conversation. You gave me this thing, and maybe one key thing changed and it no longer works. Like I know that if you're gonna get that payment from me in the following year, you're gonna go and get that fixed for. Yeah, that's a good point. Have built that thing up from scratch and have committed to undertake, to continue to maintain it in the future. Because if you don't, then I should just go and build the whole thing myself. And that's the. The eternal build versus buy equation,

Damienright? So you have a vendor in, you have that throat to choke. Obviously change is dictated by the mighty dollar there, and you're in the driver's seat, right? But what about on that initial sell? What about what brings a vendor into an organization to begin with? One could argue that in the past they would say, show us your track record. But it seems like in an era of generative ai, it's more about show us how you can make LLMs more predictable. Show us how you can. Ensure that the outputs that we're gonna receive from this are a little bit more consistent.

NikolaWe're confusing questions from different levels here. I think that good point you're bringing up are absolutely valid. It is what people are asking, but that's because they're being sold applications and functionality that is new. So to see how it works, they're like, okay, if I use this alternating current instead of direct current, can I, can it kill me? Will it electrocute me? And then you have Thomas Edison Electrocuting sheep, or in New York to prove that Tesla is wrong. Which I think he went up

Damiento an elephant, if I'm not mistaken.

NikolaDid he? We're always pro. Yeah. I'm sure he did bit a jerk, but, Oh, for sure. True businessmen, unlike Tesla. But if you just look at that, it's okay, like we are building, a certain technology, we wanna make sure that the whole thing is safe. Now, whether you cross that bridge of believing that it is or that it isn't, ideally you, you now believe that you can do that whole thing. The application form factor is something else entirely, right? Whether it's good and should you buy it or build it? I don't think people are jumping to buy this. They first need to answer that question, convince themselves that this new functionality, features, et cetera, that are being promised by these new products are actually gonna work in ways that don't destroy their business.

NikolaAnd bring the promise. And if they do. Then I still think that they're facing the same decisions like before. How do you roll it out? How do you ensure ROI, how do you ensure that you don't make a mistake? Who in your company needs to underwrite the decision? The patterns of like sales and adoption are very similar. Used to be software, it's not really all that different, right? I think the propensity to build by people is higher because it's so easy to build it now. Yeah. And like the whole kind of

Damienlet's not conflate easy with tempting. There's a lot of options out there, to go and build yourself. And it may appeal to someone's inner ego to say, okay, this looks pretty intuitive. Maybe I can give this a shot. But like you said, then the only choke that there, there is to, or the only throat that there is to choke at the end of the day is that individual.

NikolaYeah, that's right. That's right. And I think that, most people know that they shouldn't undertake, shouldn't make those decisions to do it that way because it's their jobs on the line. And

DamienWhat about when those people are pioneers of innovation or like heads of innovation within organizations? That's who we talk to more often than not. So why aren't they taking that initiative internally? They are sometimes right? Sure.

NikolaAnd then like they try and then they succeed or they don't, right. It depends on the space and, you can build a lot of things. I think that especially, with things like, cursor and other things that like make coding easier and reduce that barrier to building something from scratch. More and more people will try to do something'cause it's really fun. But then when the clock starts ticking and if you imagine the whole thing is like you're sitting in an exam, you've gone through eight pages of paper and you're still no closer to approving that thing, then you're like, shit, what do I do now? Maybe I start from scratch. Maybe I go back to using someone else's product and making,'cause at the end of the day, it's a bit of a, like an architectural, but also a financial decision.

NikolaHow long do you have to play around? How confident are you if you're 30% of the way through with 90% of the time elapsed, like then you're gonna start panicking and you're gonna end up paying more to buy the solution externally.

DamienSure. As the pain point grows, obviously the more your leg is severed, the more quickly you need to get to the hospital.

NikolaAnd we all know how this goes, right? Someone else will have bought, had success, right? And then it's starting to look pretty non-sophisticated for choosing to build it on your own. And then the cost of that decision for you and your organization is just starting to grow and grow. And tell you backtrack,

Damienall right, so we have an anxious buyer. We have someone with a problem that needs a solution. What we, they have four vendors sitting at the table. What? And it a, let's just say it's a generative AI based solution, right? Let's wipe all the other factors off. What is going to have one stand above the other?

NikolaAre we talking the application layer or are we talking about Yeah, let's,

Damienlet's keep it in the application layer.

NikolaI think it's software like anything else, right? I think you have your set of requirements and case studies will still matter, right? Sure. Like the word of people similar to you telling you that they've successfully solved that problem will matter. Whether it's their reference call with you or a case study online that you read and that is, prominently featured you're gonna care about that because. It's not just whether technology is good or not, it's is it built in a way that naturally like congeals around your organization to provide that outcome?

NikolaAnd you care, one of the CEOs of another company and one of our investors' portfolios said this to me. He's selling to banks is all about people like you. Trusting that you buy things like this from people like me, right? And that's there are three factors there. Is, has there been someone like you who sold it with that particular product? And are you buying the same product, right? So you can't really, place enough importance on the need to find.

NikolaThat comparable case study, right? And whether it's industry, similar, organizational shape, similar kind of problem usage of other things in the technical stack, right? And that's what like all those things add up. So when it comes to one gen product versus another, I think the real question would be like, how have they changed the outcome? What are they providing to you? With generative, not everything has to be generative, right? Sure. We use plenty of software that, a lot of, it'll probably get better as AI integrated further, it'll be more personalized.

NikolaIt will surface the right things for you. But the question will be like product A versus product B. Which one has delivered a better feature for me, whether it's generative or not. And I think generative AI and these models allow companies to build. Features that they couldn't even imagine a year ago. So there the question is just have you built that feature? And it's a credible to people believe that it works in practice. Either it's simple enough for them to try it out and see. And I think, there, like with anything, this is serious software, often highly customized for a particular client.

NikolaWhether the modeling layer or the application layer across, it's actually more of the latter, right? And they're like, you're gonna do pilots, right? You're gonna have some kind of trial period and ways to like off ramp the whole thing if it doesn't let it out. Yeah.

DamienExpand on that a little bit because how has the concept of a proof of concept or a pilot period changed in specifically?

NikolaGartner has done a really good job telling people like actually it goes both ways. On the one hand, people are like, wait, Garner's telling us that most of these projects are failing? Oh no. But then you go alright. More than ever before. A bunch of executives put a lot of money in a lot of different initiatives to hopefully capture, early mover advantage, right? Of using generative ai. Of course, most of it fails. That's only means that they like took on brave initiatives and that's good. And now the good news is a lot of those things are starting to pay off, right?

NikolaSo as you separate, the winners from like losers in terms of those pilots and things, you hopefully have realized some major ROI among your winners. And really companies should think of it much more as they're not buying bonds here, they're buying a VC portfolio, right? One company will return 30 x one pilot will return 30 x of the investment and others will all fail. And that probably, like Sequoia is well known as a fund for having most companies go to zero, far more than the expected average of of a run rate seed or series A company.

NikolaAnd that's because they're actually picking for variants and they're picking those that will like really go far. And that means that many of them are taking ludicrous risk and we'll go to zero. That's okay. I think that's been one of the great successes of American venture capital compared to other parts of the world. And similarly with AI was something this new. You're gonna have a lot of that. And only now with Gartner, I think we got to the point where we're looking at the gen AI landscape for applications and, I think they're starting to see whi which vendors have been able to realize very serious gains with generative ai in terms of the application layer took quite a bit of time until they formed a view on that.

DamienAnd not to do a self-congratulatory back pat. But speaking of Gartner and generative AI announcements, they just in their innovation survey or innovation report that is an iterative take on the generative AI landscape is subcategorized in four different categories in ai. Infrastructure and engineering and foundational model providers. And in a very broad gen AI applications what they defined as knowledge management apps and general productivity we just announced, just appeared today on their emerging market quadrant for those generative AI apps.

DamienSo that was

Nikolait. It's a big moment, right? As a, it's huge. Yeah. But it's

Damienfunny to see vendors like us next to Grammarly or a WSI

Nikolathink the, I think the two closest to us are Bio and Palantir. I think they're like, were a,

Damienit's an eclectic vendor salad, that's for sure.

NikolaNo. Absolutely. If you are about journalist, you'd be like, Paul Ai Beat Maidu. Yeah. Really Compare that way. But, it's a one,'cause I think there. Through their work trying to separate the case studies where vendors have made a big difference to their customers by delivering a gene application and not many have, and most players in our space are not really doing generative properly at all.

DamienA lot of'em. So speaking of risk takers, you're talking about Sequoia. I think there could have been a time in a very not too distant past where people would say generative AI is very edge tech. Is that the case anymore?

NikolaNot really. I think it's,

Damienno. Okay. So with that, in, with that in mind, are we seeing more interest in established pioneers? And again, let's keep this to the application layer more, more like pioneers like I, I don't wanna call'em legacy vendors, but folks who have been around and have built a portfolio versus the flashy young upstarts that someone is willing to hitch their saddle to in, in hopes that they are a breakout vendor.

NikolaDamien, if you see me driving a Porsche around this,'cause you just crisis here, right? Yeah, look, I think that there's golden like balance there, which is are you credible enough to have served large enterprises and credibly, delivered an outcome that they're willing to underwrite and tell the world about but also seeming young enough and having enough life energy in you. To grapple with all these new technologies, it's getting harder and harder to stay on top of it. And one of the things that I always tell people is like, Hey, we're just young enough to have never known anything but deep learning.

DamienRight.

NikolaAnd I started working this three years earlier, like I might've been saying like, Hey, we're not gonna switch to like word vectors away from like gram models and stream matching. So we were fortunate in that we started just the right time to still have that was as an energy while having had enough time. To build that advantage that allows us to say that we surface that we serve some of the largest customers in the world when it comes to doing the work of a thousand a of a thousand agents. As well as humans might hope to do. Yeah. Yeah, like I think it's a good time to be a vendor like us.

NikolaI don't know that an emerging challenger has much of either like workflows, integrations, go to market separators, or just establish relationships for it be easy, right? So a lot of people are now talking about the first like billion dollar like one man, billion dollar company. It will no doubt happen, right? And it's easier than ever, but it is mostly in consumer, right? Any kinda, sure.

DamienYep.

NikolaBuilt and their masterpieces and they're phenomenal. And I've always admired companies like say WhatsApp, who I think when they were acquired for a few billion were 22 people. Right now that's like a talent density that is just unbelievable. And it's like a company that has a very clear, narrow vision of what they do, and the only thing they need to do is maintain an unbelievable technical, sophistication, elegance, and edge. That's great. I think for a lot of other applications, you're really setting out with a company to solve a complicated challenge.

NikolaFor on behalf of your customers, of which there will be many, right? There'll be quite heterogeneous. And how you pick that cohort is often the difference between being successful, wildly successful, failing miserably, right? It's are you defining product market fit? 'Cause that elusive firm is really what the whole thing is about. So I think, like we will continue to see companies emerge and stuff. I think they're systemic. Like barriers and advantages that large cloud companies have that continue to like compound, but equally, I think we'll see disruption in ways that we can't really expect.

NikolaThat's what's called disruption.

DamienYeah. Yeah. To more disruption didn't, wasn't words with front. No. Maybe that was a two person billion dollar company. Wasn't that acquisition by Zynga, like a billion dollar buy or something insane like that? I don't know. I dunno where so the answer is either, get good at gen AI or go into gaming. That's the

Nikolaone. Gaming's always been a really difficult industry, so Yeah, no doubt. Especially to crack

DamienN two. All and just to wrap up, we have developed those case studies obviously have some tenure and some successful deployments that you can talk about because gen AI is beyond the edge case and. Finally embrace the proof of concept. Like you have to go and you have to be able to demonstrate probably more so than in the past long time. And I think the really

Nikolaimportant thing is also like POC, the right thing, right? Yeah. I think people and organizations in general are now really afraid of missing the ai, bandwagon, but where it like flips on its head is. For a lot of companies, what that means is, oh, I'm gonna try to make my agents a bit more productive. And that'll be a natural segue into end-to-end like agent automation. Sure. That's not true, right? That's running in order to become the best tennis player, like you need to be able to run, but you need a whole lot of other things and you better just grab that racket and start playing. Otherwise, you're never really gonna improve.

NikolaYeah. So I think you have to start solving the real challenge. You can't just. Defensively do a little bit of something else like sure it'll get you in into better shape, but you still won't know how to play the right game.

DamienI just wanna go play tennis. You that, or words with friends. All right, Nicola, that was very cool. I appreciate the conversation. As always, enjoy your exploration or your exploration of the West Coast and we will see you next week, everyone else. Thanks and take care. Bye-bye.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. What if your contact center became a product team?

    Nov 17, 202635 minAudio

    Three years ago, Simplyhealth's PolyAI deployment was only resolving simple FAQs over the phone. Today, they run voice, in-app, email, chat, and WhatsApp conversations on the platform. Nikola Mrkšić talks with Paul...

    Credit the people on this episode

  2. What happens if your AI agents get 1% better every day?

    Sep 10, 202639 minAudio

    An agent that gets better every day compounds into a very different agent within months. That's the idea behind Wren, PolyAI's new release: an agent whose job is to make your other agents better. Some of what it finds...

    Credit the people on this episode

  3. Who's coordinating your army of AI agents?

    Sep 3, 202644 minAudio

    Companies are pouring money into AI, but not everyone is getting the returns. Ian Jacobs, VP and Lead Analyst at Opus Research, thinks he knows why. In this episode, your host Nikola Mrkšić sits down with Ian to work...

    Credit the people on this episode

  4. Why should CX leaders care about MCP?

    Aug 20, 202636 minAudio

    When PolyAI built its Builder MCP, one client helped shape it: Vixxo Facility Solutions, one of the largest facilities providers in the US. In this episode, guest host Damian Sasso talks with Vixxo's CTO Derek Neighbors...

    Credit the people on this episode

  5. Can AI really hear a call the way a person does?

    Aug 6, 202622 minAudio

    PolyAI just launched Dialog-RSN-1, its first audio-native model, and your host Nikola Mrkšić sat down with its builder, Matt Henderson, to unpack why it’s a game-changer for building voice agents. Most voice AI either...

    Credit the people on this episode