Deep Learning with PolyAI
PolyAI's CEO/co-founder Nikola Mrkšić and team invite guests to candidly discuss trends and tech in AI, voice throughout the enterprise, and nailing the customer experience.
Deep Learning with PolyAI
Is AI the end of large engineering teams?
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Engineering teams are shrinking. But is that a crisis or an upgrade?
In this episode of Deep Learning with PolyAI, Nikola Mrkšić sits down with Helen Greul, PolyAI's new SVP of Engineering, to explore what AI is doing to the shape of technical teams, the nature of engineering leadership, and who actually thrives when the rules change.
They discuss why a new generation of non-traditional builders is outpacing seasoned development teams, how smaller nimble units with shared context are replacing large headcounts, and why complexity doesn't disappear when AI arrives: it just moves somewhere else.
I think we're seeing there is that there's a whole new generation of people who are definitely not developers by training, but they are thinkers. They will mop the floor with everyone else. Because they're just learning. And that I guess it's plasticity of their mind as a talent where they're able to just like pick things up quickly. And, you know, one day they're using, you know, like clot code, the next day they'll be using whatever the next thing is. And they might have used, I don't know, perplexity six months ago for some of these things. They're doing some fantastic, like 20-year-olds are doing projects that composite teams, nuance, would have charged five million dollars for. Hi everyone, and welcome to another episode of Deep Learning with PolyAI. Today I've got relatively new SVP of engineering, Helen Greel, with me. Before we start, please like, share, subscribe. Now that we've got that out of the way, welcome to the podcast.
SPEAKER_00Thanks, Nicola, for having me. Yeah, it's interesting to have been interviewed by a CEO. It's either like a sign of trust or a massive trap. I haven't figured out which one is it.
SPEAKER_01Kidding. So I think we have had the great pleasure of you joining about two months ago?
SPEAKER_00It's kind of coming up in my third months. Yeah.
SPEAKER_01Yeah. I think that I've really enjoyed working with you so far as Shawn. And maybe for the start, we can just kind of go through a bit of your career, kind of like what you did before, and then we'll talk about why why you chose to continue your career with us.
SPEAKER_00Yeah, for sure. I mean, look, it's been one hell of a ride, right? Like coming up in my third month. This is genuinely, I can say, has been like one of the most dynamic environments I've been in as part of my career. And when I say dynamic, it's like part of it is being in the forefront of technology, but also part of it is how like the team is running itself and like how decisions are being made. That's huge in terms of like cumulative kind of speed of execution, I would say. So that has been the one highlight to me, like just from like the early signs that I have seen, and something I'm really keen on talking about today. But like going back, I guess, to what brought me here, what led me to poly, that was like, I'm not gonna lie, was like intense couple of weeks that my family probably remembers of me doing like full-time soul searching of like what I want to be when I grow up and so that kind of thing. And I think like some of the trends that I sort of like uncovered as I was doing this kind of reflection exercise is like the part of technology that I enjoy the most, or like the moment of a certain technology that I enjoy the most is being in the intersection of like kind of emergent tech developers, but also enterprise. That kind of moment where like technology stops being a vibe or black magic and is ready to become enterprise tech stack. And this is what I've done at Spotify, this is what I've done in like previous other places, where like we were struggling for words to even describe what is it that we were building, and then a couple of years later, this became ubiquity. And it fell to me that Polly was approaching that sweet spot where like what you guys are building, what the team is building, could be like the next kind of enterprise horizontal tech stack. It's just we're not there yet, but I'm excited for like all the advocacy, for all the reassurance, for all the like the education that needs to happen. But like I have no doubts that this will happen in a way.
SPEAKER_01Yeah, there's a certain, I think, serendipity of two motions that really don't tend to go together. One is like of a technology that's accelerating from kind of like the developer layer into something that can now be used by more and more people, technical or not. And the other being that like enterprise engine of business value, of implementations at scale, of something that works for a narrow use case and delivers a lot of value, but is usually not heavily commoditized and democratized thing. And I think we're like we're there, right? In the in the zone where like the two tunnels have met. And I think that's where it's like really exciting. What I was particularly impressed by and what I wanted to ask you about, I think the audience will find very interesting, is you know, looking at my co-founder Sean and kind of like his development as an engineering leader. I think him and I have both agreed to like a mutual diagnosis of both of us being often struggling to articulate exactly what it is we see the direction is. And I think that what I've been incredibly impressed by, but I think at the end of the first week you sent a document where it was like, you know, a doctor's diagnosis of everything that we're like doing, why we're doing it, what people are not clear of or not. And I think with that, it was just clear that there's both like a clarity and an EQ that kind of like together come into this thing, which is like an ability to then wrap an organization around an effort and then create a lot more like harmony in pursuing that goal, which is elusive and changes every week. So with that, I kind of wanted to ask you about how you think of the job of an engineering leader, manager going forward as this technology changes, like what stays important and what goes away in terms of like technical strength, that communication, or like what have been like your kind of like points of development as you went through your career and figured this out.
SPEAKER_00Yeah, and this means a lot what you just said. So thank you. I didn't take it for granted. And I think that's partially maybe like English not being my first language, right? You don't assume that things are obvious or things are clear, and then like articulation becomes important before like be because even one single sentence can be interpreted in so many ways. And I've definitely learned that the hard way. And I guess like not to preach, you know, Basus' principles here, but like in a way, writing and putting your thoughts on paper remains like the ultimate act of clarity. So yeah, that's why I think like clarity is like philosophically is even more important. Like it used to be important always for me, but like even more important when it comes to like setting the direction. But yeah, going back to your question on like I guess engineering and and what stays, what goes out of the window, like we live in the world where like titles of like CPTMO, it's not a joke, right? Like I've met people who have that title, and that tells you something about like how certain disciplines are being compressed. I guess it's not always possible, but like you see maybe the direction where where we're going, where like previously engineers used to be T-shaped. I think now leaders become like T-shaped, W-shaped, star-shaped. T-shape is like means you have like a certain depth and like you build breadth around that depth. So you have like a core discipline of like, let's say, I don't know, certain technology, but that doesn't mean you can't do marketing. That doesn't mean you can't do product. This is just like, you know, you deliberately expand your breath, and this becomes more and more of a thing because like just even to give you one example, like I'm a back-end girly, like I'm a back-end data girly.
SPEAKER_01This is like what means is the first time we've used the word backhand girly on on the podcast. We will remember this moment henceforth is okay, okay.
SPEAKER_00And I suck at front end, just like like this is not I never understood front end.
SPEAKER_01I I'm helpless. It felt like this data flow thing that just exists. And I found it really interesting how a lot of like, I mean, I was always like hardcore backend as well. And I just remember people who are like otherwise non-engineers almost would go around and blitz around and do fantastic things. And then like I'd see like not just myself, but like three other people who were like, you know, two of them went to the IOI and they're like looking at the whole thing. It's like how does this work? And it felt like you're yeah, I don't know.
SPEAKER_00But yeah, the magic of this moment is like generally, like I suck less at front end now. And that's like that's really empowering. That's dangerous as well. But like I know that I can go like full circle.
SPEAKER_01Is it dangerous?
SPEAKER_00Potentially, right? Because like like you grow a certain confidence, maybe in the areas where like you haven't earned the right to be confident, because I haven't built like, you know, 250 UIs to be able to like develop that kind of tease. But now I can do that more than before. And that's like sort of empowering. And I can do like if there is no one to do UI for me, I'll do a shitty job of it. And like it will be better. I will go far like further with this, like two steps further than I was able to in the past. And that's kind of like how some of those disciplines are converging.
SPEAKER_01Yeah, yeah, yeah.
SPEAKER_00But yeah, I think what becomes harder is maybe like the amount of information you have to ingest as a leader coming from like all the different fields. And this is why I appreciate, I guess, what Sean, the CTO, is doing because it's like in a way helps you to like compress your daily intake of information into like a certain really brief but useful digest of things. It's like you tap into like sales intel, you tap into market intel, and that becomes like a matter of survival because otherwise you won't be able to keep up with all of that. So I guess we'll just have to like experiment and try new ways of.
SPEAKER_01Yeah, I mean, I've always felt like in a startup there are these like nodes in the company. And you know, with Polly, you'd know who who the people are, but it kind of tends to correlate to those that for better or worse can't shut off, and they're just following every single thread on Slack and doing that stuff. And now, at least, like what I what I see is those people are the best at building those tools, like sure I'm building all the kind of like connectors, wikis, and other things because it's work that otherwise would have been done by him, me, and like three other people. And now it's like, well, here's like a little helping hand where 10x become 100x. But it's very exhausting. Like, have you found like the days post like the massive identification to feel different, like after a day's work?
SPEAKER_00Yeah, there is something about exhaustion. I think like it's not me shying away from intensity. I think it's maybe like the different kind of intensity that is exhausting. And I haven't really been able to like pinpoint what is it. Uh, but yeah, like some days feel especially draining where like you almost have to decompress from all of that.
SPEAKER_01I mean, it feels like the work is quite technical. It is technical. I mean, I've you know, through where my life took me over the past few years, I've forgotten what it's like to, you know, do real honest work, you know, to sit and you know, spend a day in the IDE and then, you know, more than a day and it's like 2 a.m. And you kind of at the end of it, you can't really talk anymore. I think that's like that. LLM psychosis has reawakened that feeling with a lot of people. And what I hear from a lot of even our clients, I just spoke with one of them, where they're like, it's hard to step back and live that life. But if you do, then you become kind of like superhuman at the other aspect of it. But how do you mix the the need to kind of like articulate it on like the human astral plane versus building just the kind of like the agentic workflow?
SPEAKER_00Yeah, I guess, and something about it that just like triggered my memory in my head as you were talking. Like, I remember telling my previous team, like at multiverse, my previous company, like, I'm going back to the keyboard. They were like, Oh, like we see you like lighting up from the inside because like yeah, not it's not every job where you get to do that. And there is definitely like a moment of like rediscovery and moment of like gratification, because there are very few jobs that are as fulfilling as going back to the keyboard in some shape or another, right? And this is what I think is like the the core driver of like adjunctification of the workflow because it's like the self-fueling, like self-inflicting motivation that keeps you going, like even through the hardships. And that is the kind of the right level of intensity where like you've spent quite a few hours at it, but like you don't feel drained from it, at least like not in my case. So I find like it's always a bit of you know the dial that you tune of like you know, systems, business, technology, and like you find yourself in one or the other corner. But like the days where like I'm in the technology corner, I feel like rejuvenated from it. And I think that's kind of the the directive that I've also given, the prime directive I've given to my leaders, right? Like you can't really like let go of that craft because that will keep you going, as well as like this will 100x your team and yourself.
SPEAKER_01Yeah. No, no, I mean it's hugely rejuvenating, I think, to all of us, you know, kind of like former technologists. Now we are that again. And that like, I guess that's the T-shape, but I guess the star-shaped is the polymath who just is good at everything. Is there a future for product as a discipline? Oh, tricky question. I mean, like, you know, I worked at Apple where like there's bait, there's basically no product, it's just empowered engineering leaders, and it kind of feels like the star of that. Obviously, every extra every every answer is extreme and purely theoretical, but like there is a piece where companies like that will just further empower engineering managers who no longer then need the organizing, I guess, aspect of the product team, right?
SPEAKER_00Yeah, I think the part that still remains, and like this might be like the T in their T-shapeness, is more of like the the growth playbooks as well as like the product intuition. Like you become good at pattern matching, like as a product manager, and you've seen enough patterns in your career, so you can be like, oh, I've seen this movie, like I know the story, and that's in itself is worth something or like a lot, actually. And then you can build like you can layer technical aspects on top of it, or you can layer like a bit more of like growth or go-to-market aspects on top of it. But the the depth of it, like the the intuition or like the pattern matching from the past, I think is definitely there to stay. Someone has to do it, like someone has to make those decisions.
SPEAKER_01I think it's the hard the hardest job. I mean, all jobs are hard. Every T-shape, no matter where the T goes, is hard. But yeah, product is tough because it's in the middle, right? Kind of like marketing as well. Whereas like I feel I've I've always at least felt that tech and sales are like at least, you know, one flank is covered by the ocean and the ocean being customers or limitations. Yeah, yeah. Okay. Well, I mean, maybe to go back to kind of like deciding to join poly and how you looked at that, when you look at the role of technology and kind of like how our space is evolving, how did you think about that whole piece as you looked at poly versus other companies?
SPEAKER_00Yeah, I think the whole like psychosis, right? Like the AI, LLF, psychosis that like the industry is feeling right now. There is a lot of like smoke and mirrors in this. And you have to be really good in like your critical thinking, like not to unleash my skeptic at everyone, but you sort of have to be when you're like making decisions like this, when you're choosing your next adventure, when you're like in the RFP process, choosing your vendor. It's kind of more or less the same. And to me, that was like separating like the reality from the marketing. And where I think poly like turned to be special is kind of, I think first is maybe like research ethos. I don't know that it's like a widely known concept, but like having worked with research teams in the past or companies where like research arm is really strong, you sort of get that feeling of like they don't settle for status quo. If you part of that company, if you're part of that team, it's almost like a guarantee that they will operate at the very forefront of what's possible with a given technology, because that's kind of it's pre-programmed in their brain. That's how they operate. And it's not me telling you this, like you know this better than me, but that's kind of the observation that I had from the past. I was like, oh, like Pauly has a research team, that is a signal to me. Then the other thing, I guess, is like a lot of companies talk about doing something, something proprietary AI. And that's like in certain cases, this is just you know, marketing speak for like our differentiation layer is really thing. Yeah.
SPEAKER_01But it's a good eastern European skepticism. Yeah, okay, okay.
SPEAKER_00But yeah, like I guess Polly's version of that was like actually like going really deep when it comes to like what the model is, like who built the model, how the model is trained, where it gets the data, like what kind of data does it get. And like not many companies have actual receipts of that, yeah. As in like a true proven track record of like the success of the technology. And the problem is very specific. So you don't have to think about like, oh, is it this, is it that? The customer problem is very clear, and the solution is is proven by many, many years of doing this. So that just like removes the guesswork and gives you like clear data. So it was like, I really want to be part of that environment. So yeah, that I guess like that these are the key ingredients that did it for me.
SPEAKER_01Yeah, I think that the you know, I think the moment when you joined was very interesting because that's where we kind of decided to open up the platform. So I think both you joining and Morgan, our CMO, felt like, you know, I'm a big Lord of the Rings fan, so it kind of felt like that moment of Helm's Deeper Gandalf shows up and you know that you're not about to be wiped out by by orcs, and that we really I think needed a bit of like clarity, help, and just like object objective setting on the path to like the open platform. It's a moment that I'm like hugely excited by simply because it's something we've been working towards for a long time, and I think it took quite a bit of time to kind of like first figure out what to build in that model layer that solves the problem that's all the way up in the solution layer. And usually, you know, companies are blessed either to kind of like operate in some like B2C or B2D manner, and then like value spills from model to platform and over time grows, or it's like enterprise solutions and just like hire an army of sellers and do that on repeat, which we're we're doing very successfully. But I think that where we've always been an outlier is that it started from like a very deep model differentiation that I think we're well, I'm really excited that we're finally sharing that with the world. How do you look at like the whole like developer motion here? And what do you think the future of developers here, both enterprise ones, enthusiasts, are like where's the point of like maximum leverage for them?
SPEAKER_00Yeah, this is something I've been like really excited about, and it's almost like I have to keep myself on the leash of like not going like full developer maxing. This is a really exciting moment, and I go I truly feel privileged for like being able to join the team, like pulling that trigger on like letting the technology out in the wild, because as much as you can speculate about it, as much as you can like project things that you will never know what will happen with your technology in the wild, like you're you're in for a ride, like that I know for sure, because developers are very creative species, and sometimes like they can think of like so many creative ways of like using your product, making your product better. And like that niche of users is like evergreen and growing, right? Because given like what's happening with the technology right now, like some of the folks that like don't have technical background or traditional software engineering background, like they can go pretty far with the technology. And like some people call them citizen developers, some people call them like something else, maybe less flattering. Let's leave that aside. But like, I guess factually, right now, I'm really interested.
SPEAKER_01Okay.
SPEAKER_00I mean vibequoters, I guess, became like more Is that a slur now? I think so. Yeah, yeah, yeah. So I think the more like socially pleasing ways a citizen developer. But that is to say that, like, I guess if you think that like about the population of like your tinkers, your enterprise developers, your software architects, like that niche is only growing, and they're generally confused about like what's happening right now. So they need to build, they need to build fast, they need to have trust. There are some like critical production moments, like runtime moments. Yeah, yeah, yeah. You can't really win runtime. And this is where like the difference between lab experiment and production crisis becomes very evident. And I'm really excited to like to be able to support them throughout this because we need to give them, like, we've learned eight years of like hard-won knowledge, and we can successfully enable this population on it and like help them build the technology that they don't think about at night, that doesn't page them 2 a.m. in the morning. And that's the part that I'm really excited about.
SPEAKER_01Or it does if they want it to. Exactly.
SPEAKER_00But yeah, I think like there's lots to be said and lots to be done in the space, both in terms of like enabling them, building a community around like this kind of new vertical that people are building, because I think the democratizing that technology is maybe like the word that kind of I anchor on a lot. For a long time, voice was something like, you know, for the cool kids, or something too complex, or something too expensive. And now it's like, hey, it's your infrastructure. You can build on that infrastructure. Go build, we can help you. And that changes like the conversation quite a lot. And developers are genuinely very excited about like exploring something they didn't have access to.
SPEAKER_01100%. I mean, like I, you know, I think when we started like the voice IVRs of old, some of which are, you know, I don't think any of them compared to modern systems, but the real masterpieces are kind of like, I don't know, like a car from the 60s or 70s, those like collector Ferraris and Porsches that obnoxiously wealthy people like to, you know, drive around London just because they're harder to buy and maintain than than the new cars. But you know, that they're really good and they were millions. Actually, they were tens of millions of dollars in CapEx to get to a point. They produce business value and didn't make the calling population want to, you know, commit self-harm. So I remember meeting one of the former CEOs of Nuance, and you know, this is where we we started as an enterprise company. We're still very much like enterprise first, but I remember him looking at our customer base and seeing both what we did with some former Nuance customers that are now around at scale, and be like, how? And I just saw that there was like a physical, like you know, how can fire break stone moment where it was like, oh, like this is possible now, right? But then when you look at just like what modern coding tools enable, I feel like that, like, you know, slur, being I almost feel like Citizen Developer is a bit of a bigger slur, because you know, it's kind of like you're a civilian. You're being told you're a civilian, right? But what I think we're seeing there is that there's a whole new generation of people who are definitely not developers by training, but they are thinkers. They're, you know, I mean, I I I always remember like I grew up like computing and programming a lot. And like there were people who might not have necessarily been the best at like the algorithm or the maths of it, but my god, were they good at scripting and building out like literally a test harness, like as you know, they'd sit down at a computer 30 minutes before you start, you get a fresh computer, and they'd write out these like automated tests, etc., and like some basic slow algorithms that will crack a set of test cases for whatever problems you got, and they would test it and they would guarantee that whatever they implemented to run under two seconds would actually work fast enough. It's not the same talent. And I think those same people now, and I think that talent is not necessarily technical. I think that, you know, it's whoever's like helping people in like banking or legal or whatever, whoever's like the most like technically apt person, like their era is coming, right? They will mop the floor with everyone else because they're just learning, and that I guess it's plasticity of their mind as a talent where they're able to just like pick things up quickly. And you know, one day they're using, you know, like plot code, the next day they'll be using whatever the next thing is, and they might have used, I don't know, perplexity six months ago for some of these things. They're doing some fantastic like 20-year-olds are doing projects that composite teams nuance would have charged five million dollars for, and they're now doing it for some use case where they get 100 calls of that type. Like it's questionable whether even their effort is worth it, but the fact they're doing it means that they're having fun doing it. And that I think is the real inversion in that like this is not really fun. And I don't know. I the more I look at like the whole like, are we about to run out of jobs and do nothing? I think it's preposterously stupid because I actually think these guys will become really wealthy and that they will solve these problems and they have infinite leverage. But how do you think about the enterprise builder like doing this in a company?
SPEAKER_00Like, are they that person or I think something of what you were describing is probably quite common for like the integrator archetype, where like you don't know necessarily like what is it that you're stepping into, but you have the confidence that you will figure this out. As in like enterprise integrations and the stack, and like if you think about it as a system, it's usually quite complex and you don't know like what is this archetype?
SPEAKER_01What is this methodology? Maybe.
SPEAKER_00Yeah, the archetype, I think it's more like, you know, like probably your system's architect, but like of a modern type in a way, because like integrations also became easier.
SPEAKER_01Yeah.
SPEAKER_00And like that's the way to cut complexity, because like previously, just even like integrating with your, I don't know, Salesforce, like with the bespoke interfaces, it would take months. And now you can sort of like figure this out quite quickly. So enterprise development also like it was maybe 20% development, and the rest of it was just like figuring out like how to plug into all the different bespoke tools and solutions, and like for this whole thing to make sense and still be up to the scratch with like a construction CLO or like SLA SLO. But now it becomes, I think the integration parts are also can be streamlined quite a lot. And like the whole thing is just, you know, collapses a little bit, like the complexity collapses, which gives like a bit of breathing room to enterprise developers. As much as like you can still, you need to figure out like a lot of unknowns, the whole thing just becomes much more streamlined and much more predictable than it used to be in the past, like, at least in my opinion. And I think those types of integrators, like they don't care about like if it's Go, if it's Rust, if it's Python, if it's something else. Like I can just, you know, plug whatever tools together and make them work. That's the kind of mindset that maybe like some of the developers struggle with, who have been like commoditized to a certain tech stack, and integrators will like wipe the floor because they have like the art of the possible. It was always like the art of the possible because there was no other way. Like you have to make it work. And right now, that's just like the mindset that you need to have. Like some people call it growth mindset. Some people call it like they see the art of the possible. So I think that specific flavor of an enterprise developer is probably like the ones that will make a ton of money in their careers.
SPEAKER_01I think these guys always ended up in the top jobs of like large corporate IT structures, right? Because they were the traders, like they passed information through and I gave them leverage. And not leverage is a dirty word, but just like genuinely leverage in that they knew what was going on, right?
SPEAKER_00Yeah, yeah. Always. So yeah, it's like the access to information and knowing what to do with that information, like in a technical sense, is becoming like more and more of a superpower in a way, too.
SPEAKER_01Yeah. No, I mean it's really fascinating. How do you see then like teams evolving in engineering? Do you think that we like end up having more atomic units of people, or like do we just kind of like build towards fewer people who are holding more of the context in their own minds? Like do we all become these integrators or is there value for a specialist?
SPEAKER_00Yeah, there is, I guess, like there might be value in in specialism in like certain places, but overall, I think I see us like moving towards smaller, more nimble units where like we can all process like more context in our day-to-day, and we can go like slightly further in our disciplines, which means like certain jobs collapse. So like you may be as opposed to having a team of seven, which like used to be your one pizza team. Now it's more like team of three, because team of seven more pizza for everyone. Yeah, more pizza for everyone. So yeah, I see this as like I guess more nimble units united by some kind of company brain or shared context. So like at any point in time, you can like plug into that context and know what is it that you have to do in your day-to-day, and then go back to your like nimble unit and like move really quickly on your specific direction. So it's always like this balance of you know autonomy and alignment, but it's just like for autonomy, you need maybe smaller units that still are able to move quickly. And I think like from engineering fluidity, like resources fluidity perspective, it also became easier in a way because previously, like you had, I don't know, like a team of hundred. I used to have like a team of hundred. And then it was not really a team of hundred because like it was only 20 who can do certain things, and there was like only 30 who can do other things. So like it was never as fluid. Like on the like in the books, it was like, oh, I have a team of hundred, but when I actually need like 80 to work on like a specific project, it was never possible. And right now I can actually tap into that because like your you know, goal Rust Python developer can all of a sudden do TypeScript and be pretty efficient at it. So it's like, yeah, that kind of mobility is also empowering a lot of organizations, I feel like.
SPEAKER_01Yeah. No, I mean that's like it's going to be quite interesting. I mean, is there a future where like all these organizations are way smaller and genuinely the jobs go away? Or do you think that the economy like shifts to support more companies where all these people are jacks of all trades?
SPEAKER_00Asking all the hard questions today, Nicola. Yeah, I think like there is this like the law of physics where like energy is constant, right? And like I think about the same, like the complexity is kind of the same. Like if complexity is kind of shrunk in one place, like complexity is being added probably somewhere.
SPEAKER_01Yeah.
SPEAKER_00So like I think the balance will be like about the same. It's just the jobs will probably move. Like IKEA is a like one interesting example, right? Who have done it recently, where like they moved their customer success, which was like a huge team of like some hundreds of people. They move them into like creative, creative side, which is more like design, like uh interior design, some kind of creative.
SPEAKER_01Improving improving the meatball recipe.
SPEAKER_00Kind of, yeah, yeah.
SPEAKER_01Okay.
SPEAKER_00So I think like it's more about like organizational fluidity as opposed to like, oh, like we're gonna have less of stuff. That's how I think about that personally.
SPEAKER_01Yeah. I mean, I think like with the whole, you know, like conservation of energy. I mean, if you when you think of like entropy just increasing forever, I feel like the more complex these things are, it actually becomes, you know. I think of I forget which of the Matrix films this was, but you know, you see a bunch of those like flying dangling things architecting itself around each other. And you're like, is this really the most elegant way to like fight the remaining part of humanity? And like, probably not, right? It just feels like, you know, a set of abstractions and programming languages and structures that are not really optimized, but I guess it doesn't really matter. But I think like, yeah, cutting through that is probably gonna become has become. I mean, it every 10 years we get like an extra layer, and it's not like the previous ones have gone away. It's just that when a really catastrophic bug happens, you get like a generation of engineers who don't even know what garbage collection is. Right? And it's I remember those bugs happening like as Python really took over, and you know, they were genuine, like like there was almost like this architectural joy that I remember me and a few colleagues had, where it's like, oh, a memory leak. Yeah, yeah.
SPEAKER_00Memory management, what is it?
SPEAKER_01Like Well, because you I mean you'd hear the older like engineers just talk about it as if it's like a fact of daily life, right? And we're like, wow, cool, okay, now we know what it is, right? And it's really bad. Like it crashed our like Isito's instance and everything, everything died. Like, and now you'd probably have things like respawning and you would have to like decide whether you're gonna even go hunting for it or just wait for like Kubernetes to restart your thing and like it goes away, and maybe you're throwing God knows what resources it's something that you can fix. But anyways, not to go off topic, like talking about that extra complexity and just like what I found interesting about dialogue in particular is that as we improve and we have massively improved the ability to build a better, more functional, more interesting dialogue system, there's always like this thing that we want to borrow from the future, right? When I did my PhD, it was more about borrowing, the ability to understand more in terms of like richer language, to not use like very scripted vocabulary, and then that became sold, right? Then I think like it became about the flows and the processes and like where the conversation can go. And then we built what would be called harnesses today around models to give them the ability to still kind of like you know guide the conversation back. But when you think of agent studio and the harnessing around it, how do you think about that versus like the proprietary model? Because it kind of comes back to that whole like why bet on a company which has a research team of time, no matter how special they are, right?
SPEAKER_00Yeah, it's a really good question. And I think like we're like we're sitting in this uncomfortable moment where like this is a transitional period. You're trying to extract like a lot of knowledge from like quite a big population of the team who has done certain things, like sometimes deliberately, sometimes following their intuition, and like you're trying to platformize it and like extract some kind of blueprints out of it. And this becomes like either part of harness or this becomes part of product, or this becomes just like a one-off thing that we do for a customer. And you have to be able to say, like, what is what, right? And I guess some of that complexity, probably like looking in my crystal bowl, like models absorb more and more and more of this. So, like, there is maybe a reason to say like harness becomes a slimmer kind of layer. But like right now, I think this is kind of where a lot of like institutional knowledge and proprietary differentiation sits. It's that layer which makes all the difference. I mean, like, there is still like you know, like how intuitive this is, like, how easy is it to build an agent, but like, is it actually like the enterprise level solution? Is it actually delivering like on the promise of like certain, you know, containment rate, or like does it flow naturally? A lot of this sits in the orchestration, a lot of this sits in in the harness layer. And that's kind of the push and pull that we're experiencing right now of like having this done by like a really heavy professional services motion, as opposed to like, oh, this can be just like abstracted and happened without you even noticing that it happens. So, like, that's a really like fascinating product management challenge, I feel like, but also like partially technology challenge, but like project management's challenge by the book almost.
SPEAKER_01No, totally. I think that like everyone's now fascinated by like forward deployment, but I think to really live forward deployment, there's that like return of information that guides product roadmap. And I think most of the companies that now talk about forward deployed are really just doing professional services and rebranding them with like a sexier name, but really it's just like okay, you've implemented something for a customer, and like that was always part of like software delivery, right? But I think like the part where you then figure out which of those really hard things that like doesn't need to be repeatable or is it gonna be that different every time? It's hard. And like I think that like I always believe that we do like 10 big enterprise deployments and then like it's a platform business. I think we're now steadily really, I would call us a a real platform business. But you know, it took like 200. And those 200 are actually often like pockets of multiple in them, right? So like different national businesses, different parts of a business. So overall, I think like when someone counted it up, there's like 2,000 different deployments. And it's like from a product platform layer, I'm not sure that any single person in the company had visibility of the whole thing. Yeah.
SPEAKER_00Yeah. Speaking of like pattern matching, right? And like why do we still need like product intuition to be able to kind of condense it and put that magic uh behind behind the surface?
SPEAKER_01Yeah. So I mean, I think one of the main things that we did in the whole like platform overhaul now is really having the agentic builder as the homepage and doing away with all the beautiful UI that we spent honestly almost a decade trying to perfect. I don't think we're anywhere near to perfecting it. We have many versions of it. How do you feel about like the the agentic builder or codename GLOT, which will, you know, someone's gonna cut this out of the video.
SPEAKER_00Yeah, this will get us in trouble, I'm sure.
SPEAKER_01But like, how do you think about like that being the default like control panel pane ID, whatever, whoever you are logging in?
SPEAKER_00Yeah, this is the part uh I think that is democratizing and like just really uh making this technology accessible to like regardless of like what part of like what path of life you came from, almost because like if you think about conversing to an agent like in natural, like plain English, and like a bunch of you know, sub-agents and like background processes doing the dark magic for you, and actually like building something like that lives up to your spec, like hand holding you through the journey of like testing, is it good enough? Like, is it up to the scratch? Does it live up to your requirements? Like being able to go full cycle and knowing that there is like a you're supported by like, you know, years and years of accumulated expertise. I think that's the beauty of it. Like the new kind of design patterns of like there is this tags box and like a lot of magic abstracted behind the tag box. Uh, I guess sometimes like a lot of users feel of like, I don't even know like what to ask it, like I don't know what I don't know. And I think what our interface does really well is kind of sort of nudging you, prompting you, like, have you thought of this? Have you thought of that? Like, oh, we need to do X, Y, and Z in order to make sure that it's like actually like we're ready to go live. I think that's the the balance of like not being kind of too prescriptive about it, but at the same time, like showing user the right path to get to where they want to be. So I'm really excited about that path. And I think it's like equally suitable to like your technical leader or not so technical leader, like your business user, your developer, like because for developer, I think as much as like we have other more like traditional developer lines of like, oh, you can go with an API, you can go with an ADK, but like if you want to understand how like sausages are made in a way, the agent studio like does it really well for you in terms of like explaining the reasoning and like what's actually happening. So you might as well like go through that simulation to be like, oh, now I understand the process.
SPEAKER_01Yeah, yeah, yeah, yeah. It's almost like a safe gateway drug to like a coding tool where yeah, I mean, I don't know if we can like embed this in the podcast, but it's it's really powerful to see the whole thing being built in the background in the same with the same effort that I know would have taken three months before, where now you see the sketch, you go like, you want this? Yeah, yeah, cool. Click like three minutes later, it's all built and you know it's built with like the right like kind of like safeguarding and guardrails. But I guess that's very specific and tied to us. Is there life after cloud code for like is that just the new default interface to everything?
SPEAKER_00Kind of.
SPEAKER_01I think uh Do you use anything else?
SPEAKER_00I used to use Windsurf. I've experimented with like Deep Seek because who didn't, right? And like the open claw obsession didn't go past me because like yeah, no one is immune to that.
SPEAKER_01I am a treated addict of my one of my three claw bots still lives, but I think it's been superseded 95% of a claw code. But yeah, it was uh it was a fun time though.
SPEAKER_00Like, yeah, no, there's definitely like an evolution of that, but I think the like the assisted nature of it will like is here to stay. I'm quite certain about that in terms of like maybe like clearer contracts, clear like delegation rules of like, oh, like delegating to your master agent that runs like certain very specific jobs and things like that. But like it's definitely a bit more like collaborative. What I maybe miss the most right now is like having a clear like mope session or like your typical pair programming session. It's funny how like you know, like the the human psychology starts to work in a way because it's almost like it's a safe environment. Like you don't use your rubber duck, like Claude is your rubber duck, and like good old times of like you know, tapping someone on the shoulder and asking a dumb question, you don't feel like doing this, or like you do less and less of it. It's just like, oh, Claude knows it all. Like, go talk to Claude. So that is the part where like I'd be curious to see what happens.
SPEAKER_01I I find that I mean that's one of the big reasons I was I think like the way that Claude is built to kind of like be your own talented, that you kind of look in and you don't like get your head out of it. And like it's very powerful. It shows you things that you want to or don't mostly things that you want to see. Yet it's not as sycophantic as Chat GPT. So you don't really feel that way. Although I've definitely met like strong engineers who kind of go at the end to just go, like, well, okay, if you like, you know, this thing telling you you're smart and telling you everything's a great idea, as you sit there burning, you know, millions and millions of tokens. I mean, I guess the answer is I do. But and I think most people do, which is as evidenced by anthropics revenue, right? But I think it's really interesting to see just like the evolution of a new person learning to do it. Because I think once the affliction goes past like the first escape velocity, I feel like people, especially technical people, are really on a tear and they just start building these like agentic loops and systems and like they feed all their information into it and it's like really, really powerful. But I think where the whole thing becomes a bit difficult is someone who needs to like go close to that first escape philosophy, doesn't know what to ask, doesn't know what's possible. It's hard to see and copy that behavior. And this is where I felt like open clause and Slack were pretty good because they were, you know, they could talk to each other and you could observe them talking to each other. And like it was also like much more personal and memeified. And I feel like text spreads through that vector, maybe the fastest. But yeah.
SPEAKER_00Do you use anything else other than like Codex was probably not my cup of tea? There's something personal about it, right? Like as much as I'm trying to like be scientific about it and have like, you know, make data-driven decisions, like it's you like it or not. And there was something about it that I was just like, no, like retweeting that back to my cloud. But it's maybe also like the setup that like is not really as portable as everyone thinks it is. Like, for instance, I have like I have a skill that's called uh like a lazy engineer skill called ponytail. It's almost like you're uh uh like lazy developer who prefers to like solve problems with one line of code, and it sort of like forces you into that mentality of like instead of writing like a bunch of slopes of code, it's just like what's the minimal amount of intervention we can have there? Let's go for it. And like I find that like really, really, really helpful. And like having that setup and like knowing that, like, oh, like if the code comes out of like my GitHub account, it will be that minimalistic and that brutal. And like being able to replicate that setup like in a new kind of coding environment is like more of a lift. So you sort of like you become an addict of something because like you didn't have two days to like replicate that kind of setup in a different environment. It just becomes, you know, like your, yeah, this is how it becomes sticky.
SPEAKER_01Yeah, no, 100%, 100%. Maybe final question Do you use voice when you use those coding tools?
SPEAKER_00Yeah, like I'm am I crazy enough to talk to my computer? Less so I I guess I'm talking to my phone. Like, do I believe in people talking to their computer? I think I do. This is becoming more like more and more of a trend, and there is a clear data to to back it up, right? And like even in like coding assistance, we see more and more of like people actually talking to them. No, yeah.
SPEAKER_01Well, I mean, that's the question. Do you use do you use it with cloud code or on my phone, not on my computer.
SPEAKER_00There's something like I guess maybe me coming from like an older generation, like talking to my computer still feels like, yeah, I might need to have, you know, like some kind of rethinking of that. But yeah, no, I didn't talk to my computer.
SPEAKER_01Yeah, yeah. I guess I'm I'm similar to you, senior leader of a very large tech company. Recently gave me a lot of trouble for not believing that laptops would be shipped without keyboards. He was like, your voice company, how can you like possibly not believe in this? And I'm like, there's something in my personality where I see it. And then I think if you're bothered by that, it really breaks the flow. And then there are people who are not bothered, and for them it's fine, or people who maybe just get a better transcription accuracy than we do. And then I think at that point maybe you cross the chasm.
SPEAKER_00But yeah, that's a safe theory. I haven't thought about that in that realm.
SPEAKER_01Brilliant. Well, thank you for a fascinating conversation. Excited to have you on board and for all the impact you've already had. And thank you all for listening. Like, share, subscribe, and we'll see you in the next one.
SPEAKER_00Thanks, all.
