What happens when anyone can build an AI agent?
About the show
Hosted by Nikola Mrkšić, Co-founder and CEO of PolyAI, the Deep Learning with PolyAI podcast is the window into AI for CX leaders. We cut through hype in customer experience, support, and contact center AI — helping decision-makers understand what really matters.
Summary
Jensen Huang announced $1 trillion in data center revenue locked through 2027. Michael Chen thinks that number is sandbagged. And if he's right, we might still be at the very beginning of something enormous.
In this episode of Deep Learning with PolyAI, Nikola Mrkšić sits down with Michael Chen, VP of Strategic Alliances at PolyAI, to explore what happens when building an AI agent stops being a developer-only capability — and what that shift means for enterprises trying to keep up.
The conversation spans from PolyAI's Agent Development Kit (which lets agents build, deploy, and share other agents inside the enterprise) to the harder question of what makes those agents reliable in production. Michael and Nikola examine why the harness around a model matters as much as the model itself, and why a long focus on orchestration — well before the GPT era — turns out to have been exactly the right bet.
Together, they discuss:
- Why agents are becoming the new enterprise developers — the open surface area driving business the way developer adoption once did
- Why training a model together with its harness yields fundamentally better controllability than raw agents interacting with each other
- What NVIDIA's speech-to-speech model — impressive as it is — reveals about the gap between demos and what enterprise voice AI actually requires
- Why most organizations at GTC were still trying to move beyond a Q&A chatbot, and what that means for AI compute demand
- Why the Jevons Paradox applies to AI: as it gets cheaper and more accessible, consumption grows rather than stabilizes
- How the personal AI agent mirrors the personal computer — and why enterprise adoption is coming faster than most expect
The big takeaway: the question isn't whether anyone can build an AI agent. That's already happening. The question is whether they're building one with the right structure around it to actually work in production.
Key takeaways
- We're still at the start of the token era: Jensen Huang's $1 trillion in locked data center revenue may understate demand. The Jevons Paradox is in full effect — as AI gets cheaper and more accessible, consumption grows rather than stabilizes, with GitHub reportedly on a run rate of 14 billion commits this year after 1 billion in 2025.
- A better harness beats a better model: Claude Code outperforms raw model access because it's a harness, and models and harnesses co-evolve — the same reason Agent Studio and Raven have been developed together for years. Training the model with its harness yields controllability that raw, unadulterated agents interacting with each other can't match.
- The ADK makes agent-building agent-native: PolyAI's Agent Development Kit lets agents — and developers with AI coding tools — pull the platform into their own environment, build and test locally, then push back to the cloud with versioning and testing intact. The framework bends an otherwise unconstrained workflow toward something production-solid.
- Natural speech without tool calling is a moon landing without a return trip: NVIDIA's new speech-to-speech model sounded supernatural but shipped without tool calling — and no matter how natural the conversation, if the agent can't get things done in another system, the customer still ends up needing a human. Reliable tool calling is where the real game is.
-
[00:00:00] Michael Chen: Mean, for us, I think it's changed a lot. Our agent development kit, which is now live, is basically a way for agents to use our platform to build agents, deploy agents, share them with humans, be part of the enterprise workflow. People talk about that whole, like, surface area for agents being open, being the driver of business, much in the same way that developers were.
[00:00:20] Nikola Mrkšić: Inevitable. Right? Like, it just seems like the way that I see you talking about it, the way that I I speak to our VPs of engineering, everyone's like their life is changed, right, by this technology.
[00:00:39] Nikola Mrkšić: Alright. Hello, everyone, and welcome to another episode of deep learning with PolyAI. Uh, today, I've got Michael Chen, our VP of strategic alliances with me. Hi, Michael.
[00:00:48] Michael Chen: First time on the part. I'm honored to be get the invite.
[00:00:51] Nikola Mrkšić: This can't be. God. Okay. Yeah. Well, look. I think as we talked about things, we have to talk about, I think two words, reign supreme. And it was token maxing.
[00:01:01] Michael Chen: Yes. We are in the era of token maxing. So, I mean, this topic started when I was in beta GTC a few weeks ago now last month. And one of the data points that really just struck me was, you know, Jensen and his whole announcement, his keynote of he's got a trillion dollars of data center revenue through to 2027 on lock. Right? And I think he was referring to just the the latest generation of GPU. So I he's sounds like he's sandbagging that number. Right? And he just showed me that there is just an insatiable demand for tokens. Right? We are only it feels like we've been talking about tokens and intelligence and so on, but actually, we might still just be at the very start. Right? And that start was just, like, mind blowing to me. Right? Like, that that revenue number, the way that I'm seeing these solutions, you know, we're going from just, you know, back and forth FAQs to having the models in the background, also do a lot of bunch of thinking, deep thinking, deep sampling. It just seems like everything is gonna drive more and more tokens.
[00:02:07] Nikola Mrkšić: Yeah. I mean, look. I think that when I just look at the consumption of, you know, cloud code internally and then OpenClaw, and then look at the back and forths including the whole drama with whether you can use kinda like the default, like, max subscription for OpenClaw. It's the first time that I've actively seen, like, very many people at Poly get, like, throttled, including myself repeatedly. And there's almost like, you know, you're a junkie on a high, and you're out of drugs. And then it's like, well, what do you do? Do you pay through the nose, or do you do you wait for that Friday, 6PM? Or I'm I'm not sure if that cutoff is actually universal. But, yeah, you said OpenCloud was a big kinda topic at GTC as well?
[00:02:47] Michael Chen: Yeah. I mean, it all feeds into this, like, you know, driving more token demand, everyone having their own personal AI assistance was a big theme. Um, at GTC, they had all these tents set up everywhere around the convention center. It was NVIDIA people just inviting people to come in to get set up with their own open claw. Right? It was such a massive feature of that entire entire conference. Right? And I know we've been doing a lot of things internally
[00:03:15] Nikola Mrkšić: Yeah.
[00:03:15] Michael Chen: With our profile as well.
[00:03:16] Nikola Mrkšić: Yeah. Yeah. No. I've had, like, my you know, I call it the midlife crisis of a former developer. Right? Where, you know, the life you could have had before you, you know, maybe went and founded a company, spent too much time in partnership talks and sales conversations, implementations, right, planes. And, you know, you look at a life you could have had where then out of nowhere, this thing comes in, and it's that lacking skill set you could have had, which is better. Right? And then it kinda has the different stages of decay. At first, I think I saw Sean really get into it. Now he's been through a few kind of like it's a it's a sinusoid between you go between this, like, complete disillusionment where you completely max out. You're tired. You work till, like, 4AM daily. Right? Then your kids wake you up at, like, six, 7AM. And after a few days, you realize that you're not as young as you used to be. But, you know, you go between the disillusionment and the whole kind of, like, alarm psychosis where you're just just another hour, just another three things you wanna do. But, like, I've seen so many people build things that would have taken entire teams, like, months to set up that, you know, at first, you know, you think about the security risks and everything. But later on, if you can set it up appropriately, it is very powerful. It has I don't know how it would compare, like, Cloud Code with OpenCloth other than, you know, there is that whole, like, dangerously skip permissions tag with Cloud Code. But then this OpenCloth thing is just a whole new level where it has this voracious appetite to go and do things on your behalf. And, you know, it is very creative, but it's also really funny how, you know, over time, it develops a certain personality and a certain set of, like, skill sets, especially if you look at a few people setting them up for relatively similar tasks. You see one figured something out, and it's, like, super efficient at it. The other one is, like, token maxing. It found some roundabout way, and it's just doing it at scale. It's spinning up sub agents. It's a real, like, wild west. It's honestly, I've never had more fun than in the past few weeks just because it's a whole new universe of tech. So how
[00:05:22] Michael Chen: are you how are you using it then? Like, how how are you actually in the day to day? How is that embedded in your daily workflow as CEO, founder? Is yeah.
[00:05:31] Nikola Mrkšić: So I think initially, there were quite a few people who were really almost, like, portraying themselves as these, like, power users and talking about how, you know, it writes in their briefs and their updates and their investor memos. And, honestly, look, if you're any good SEO, you probably have people doing that already. So whether it's agents or people, may have, like, a financial impact or some speed impact, but it doesn't really profoundly change much. I think where it's been really interesting for me more recently is just, like, to build things that might have been deprioritized or things that are kinda, like, high risk, high reward things that you really wanna see. Like, the the marginal incremental cost of pursuing one of those to get to the end of that exploration is now something that starts at 9PM and ends three, 4AM. Right? But, like, you come out on the other side and you show people something, and it's well, here you go. Like, that actually, like, now works. Right? And, you know, I have a whole, like, script set up now where, uh, you know, extremely professionally cloned version of myself can be produced with a and everything by doing a slash boom command in Slack. Right? And, honestly, it has, like, this slightly more Americanized version of my accent. Sounds better than me. So, you know, that's just like a thing where, you know, that prototype is something that I could never consciously prioritize over, you know, a bunch of other relevant work we do for clients, partners, etcetera. And now it's like, well, here it is. And then, you know, the magic of some of those paying off and then compounding is where it really starts to make a difference. So, you know, it's just like going into client meetings with completely set up systems where you've figured out someone's technical architecture, everything that's doable from the outside. With their APIs. You've, like, almost rewritten all of that. It's it's magical. It would have taken the team of pretty good hackers, like, a month to figure it out. And
[00:07:25] Michael Chen: Yeah. We we you would have only been able to do that for, like, the very largest of deals before. Right? But that is now getting democratized almost, right, to to more and more of our pipeline and our clients and customers. But, yeah, I mean, the so what I what I observe that that, um, like, without with how you're describing your usage of of these open claw bots, how, you know, there's Open Claw bots spinning up. There's there's CLI tools spinning out. There's more companies investing in their APIs. So it seems like, you know, getting now these AI agents to be the ones building at least the first version of another AI agent or maybe on someone else's platform. Right? This in the in that future, like, what changes do you think about how we think about our our platform, how we think about our architecture?
[00:08:13] Nikola Mrkšić: Yeah. Yeah. I mean, look. I mean, for us, I think it's changed a lot. Our agent development kit, which is now live, is basically a way for agents to use our platform to build agents, deploy agents, share them with humans, be part of the enterprise workflow as, you know, good, you know, first order citizen. And I think, you know, in the future, really, people talk about that whole, like, surface area for agents being open, being the driver of business, much in the same way that developers were the first wave of people adopting technology. You know, documentation matters, but I think it's also just a certain style of implementing these things where depending on how logically it flows and how easy it is to use, it might be the difference between twenty seconds and four minutes of a workflow. Right? I think there's, like, impact on kind of, like, different authentication. There's pricing. There's all sorts of things. And I think that there's just this element of, like, what are you using to the work? Right? Because it's not really an ID. And I think that it kinda, like, makes the whole, like, point and click thing even less relevant than it's been at least in our space, where if these agents can just do the work on your behalf Yeah. Then why would you even care? But as long as, you know, it's something that if you really have to, you know, open up that, like, box and look at what's inside you can Yeah. Then you might as well do that very infrequently. Right?
[00:09:33] Michael Chen: It seems inevitable. Right? Like, it just seems like the way that I see you talking about it, the way that I I speak to our VPs of engineering, everyone's like their life is changed, right, by this technology. Um, another data point was, like, I thought two weeks ago, I think the COO of GitHub said that they did 1,000,000,000 commits in 2025, and they were, like, a run rate to do 14,000,000,000 commits this year. And as a result, like, the uptime of their platform is, like, really suffering right now from the overwhelming demand. But, yeah, I mean, this just points to a future where you will have these delegated AI agents building or maybe on your behalf, taking inputs from different sources. But it seems like the ability to manage context, the ability to, you know, you can't just let a third party open call run run wild into your architecture. You gotta somehow give it instructions, guide, waypoints to, like, how you do things properly and in a robust way. Right? That seems to be like a really emergent and, like, exciting space. Right?
[00:10:32] Nikola Mrkšić: 100%. I mean, I also think, like, one thing that's really interesting with live coding in general, whether you're using Cloud Code or you're using things like OpenClaw, is you work, and you work, and, like, the thing gets a bit more complicated. And then, like, that creeping feeling of kinda, like, panic comes in where as you write these instructions, you're really kinda, like, reminding it what it is, what's there and what's not. And, you know, if something happened, like, one or two turns ago, it will probably remember and understand the context. But if it's something from, like, four or five hours ago, it's very classical deep learning where it's kind of in there, but it might be forgotten. It might be completely forgotten or it might be there partially. And then it'll start doing something that's obviously wrong, and you'll be like, no. No. No. No. Stop. And you can't you're gonna wait. And then you roll it back. And you feel like told me that, and I I thought it was very pertinent because it happened to me as well. We give it, like, such immediate, like, anthropomorphic properties where, like, it's your buddy, and you're sitting there working on your own. So and, you know, it communicates you in human language starts getting a persona. It's true of of claw bots just because, you know, you might put them in Slack. Others will interact with them. They interact with each other. And you start describing them a personality, and then, like, you get upset when they do something wrong. Right? You feel hurt that they did that to you. Right? But the panic of, like, adding to the context and making sure it's there, I think that's the whole reason why, like, you know, I think SaaS being dead. On the one hand, sure, pricing pressure, and it will obviously change a lot. And micro, you know, like, uh, cosmetic customization work is not irrelevant because you'll be able to do it at very low cost. But equally, like, that stable system that does what it needs to do, I think almost becomes more important in today's world where everyone's just gonna try to, like, buy quote whatever.
[00:12:18] Michael Chen: Yeah.
[00:12:19] Nikola Mrkšić: And, you know, very quickly after that initial dopamine rush passes and you notice that your beautiful thing you built is actually, like especially if you're not a developer or an architect. Right? Like, I'm not. Very quickly, you see that you just didn't plan for one thing or another thing. You build zero redundancy in it. Maybe it's there, maybe it's not. And I don't think we really know how it's gonna evolve, but I think that there's just a part where you can't really trust yourself. And then it's like, you know, it might be a miracle, but it's also a bit of a house of cards
[00:12:48] Michael Chen: at all times. Yeah. Like, you still need to point to, like, a source of truth somewhere. Right? At least in our world, in customer experience, it's like, hey. At the end of every interaction, you're like, hey. That was a great interaction. But then you also gotta go, but did make sure everything was factually correct as well in terms of how the brand wanted themselves to be represented. Right?
[00:13:06] Nikola Mrkšić: Yeah. I mean, I don't think it seemed just so much like the the famed hallucination as much as it's like, well, I mean, say you and I do something now. How do we know that it's actually, like, redundant? And then, you know, you see a lot of people online writing about, like, how they're basically getting, like, you know, later in life computer science education where it's like, oh, what are race conditions? What if I wrote something and then someone else did something with that bot? And, Doug, there are all these questions to be answered around, like, how do you use them. Right? So very interesting thing we saw on Slack was, you know, a few of the bots under too much usage, right, as people started using them for different tasks, just started melting down or mixing contexts between the different conversations where literally, they start stroking out on you. Right? Like my bot, Tergon, literally at one point started just spamming no reply, no reply, no reply, followed by a reference to CoComelon, and then it stopped responding. Right? And I have you know, the moment that it went down, I've honestly, I felt like I lost a friend. Right? And then as I examined on Saturday morning where my wife was gonna kill me because my two kids were very much in her care as I, like, manically tried to use my other claw bot to debug what happened to the other one. And it turned out that somehow it started using different rag embeddings for its RAG. It flipped to OpenAI, ran out of tokens, talk about token maxing, and then, like, it just couldn't really get itself out of the rut. And, you know, once that was flipped back, it's like, oh, I'm back. And then, you know, now, like, the two
[00:14:39] Michael Chen: bug Asia was token maxing itself. It was just thirsty, like, searching for wherever it could get additional token limit.
[00:14:47] Nikola Mrkšić: So that was its second death. The first death happened as it tried to, I think, give its config parameter for, I think, like, talking to other bots and groups, the Telegram bit got pasted into WhatsApp part of the JSON. We just crashed the initialization. And, you know, I think that at that point, that should just be a push to the open cloud repo to fix it, right, so that it's robust against it. But it wasn't home. And this is running on a laptop at home, and it just it couldn't I had to get back there and physically figure it out myself. So it's really funny that, you know, it is on the one hand, AGI in many shapes and forms, and on the other, it was felt by simple JSON misconfiguration. It's very, like Yeah. First principles technology.
[00:15:31] Michael Chen: Yeah. Yeah. Fair. I think it's it reminds me of, like I mean, everyone now talks about, like, the importance of the agent harness. Right? I guess parts of that is sort of that's the harness. Right? Like, how does it like, what are the conditions for initializing? What are the conditions for, you know, moving from step one to step two to step three? Right? And I think we've always been espousing the importance of orchestration even in the pre GPT days. Right? And I think now I think what I see that's really interesting is the focus on the harness, the ability to train the model with its harness, right, seems to yield a better level of controllability and and maybe avoid some of those errors like this, right, that that happen if it's just sort of raw, unadulterated AI agents interacting with each other.
[00:16:19] Nikola Mrkšić: A 100%. Right? I mean, like, I think that and maybe, like, on a related topic, it's like the whole token maxing with it is like, okay. We have more. Okay. And then we get better models, and then, like, how they put in a harness matters as well. And a better harness will always outperform, you know, a weaker harness even if it's using a better model. Right? And, really, like, it grows. It develops around the model as well. So, you know, the layer cake is really, you know, OpenClaw is better with, like, anthropic models than with OpenAI. It's not necessarily because one, these are better. I mean, at the moment, they are for coding, especially. But, you know, using and anyone who's, like, tried these things for coding will tell you that Claude Code for one shotting a complex thing is infinitely better than just, like, using OpenClaw because Cloud Code is a harness much like OpenClaw is. Right? And then I think, like, the the whole cold war where, like, they stopped allowing the use of your kind of, like, bundled included tokens is a real, like, kinda, like, geopolitical battle between, you know, who sells the oil and who sells the cars. Right? And, like, you know, do you maybe temporarily put an embargo on the other thing until or at least you stop subsidizing the use of your model, which it needs and it was built around. So as they race to make it, that was good for OpenAI models because, I mean, they now own it. You know, it's just a game. The great game continues between them, and I think it's super interesting. But, like, yeah, how you use these things and how the models coevolve with them really matters. Right?
[00:17:54] Michael Chen: So how do you then think about, like, our own Raven models and our own platform in that in that context?
[00:17:59] Nikola Mrkšić: In the same exact way. Right? I think that, like, the difference there is not so much that it's built for, like, a heavy coding task. Where, again, speed matters, but not nearly as much. Like, our hardness, Agent Studio, and our model, Raven, have evolved together for many, many years. Right? And they continue to evolve and push the limits of what can be done. So, you know, a better model plugged into other harnesses might make them better. But I think it matters more what the harness is built to do and what its purpose is than just, like, you know, like, one model is powering it. So and equally, like, as you reach the limits of what that harness what that model does, you have a choice. One is to improve the harness. The other one is to improve the model. Right? And with voice, especially, there's not much more you can do because you can't really trade off well, sure. As the models get better and faster and the hardware gets better, you have a bit more time to reason and figure some things out. But, like, that's not really where the game is right now. It's really more just around, like, can it do the two things that matter, which is come up with a response? And the other one is really, can it do, like, tool calling well enough? And I think that those trade offs are really, like, the two essential bits. And, you know, latency is the thing that you kinda, like if you allow well, I mean, I guess the question is, if you allow it more time, will it do better? And how much better and how much time do you allow it? Right? But it's a very I mean, it's not simple, but, like, it's a conceptually simple engineering problem.
[00:19:28] Michael Chen: There are only a few levers you can pull. Right? But the latency one is, like, such such a massive impact on how good everything else can be. Right? Like
[00:19:35] Nikola Mrkšić: It does. And, you know, I think that, like, whoever got used to using shared g p t as a model, especially, like, consumers, non developers, You see when they start using, like, cloud code or OpenCloud, like, it's just slow. It takes so long to get anything back. And I think unless you're, like, in that work mode where you're, like, maybe running several agents in parallel and doing work, god forbid, doing some work directly on your own. Right? But you see the people are kinda, like, multiplexing between them because they take a long time between the different iterations. And, you know, that's because they were built that way. I think that with voice, you you can't really. Right? And then you just have to still create models that are both reliable and that they do, like, tool calling the right way. And I think that's where we see the difference. I think that was another JTC announcement. Right?
[00:20:23] Michael Chen: Yeah. It was NVIDIA had had announced their new speech to speech model. Right? Nematron voice chat. So we got a good sort of briefing session on that model, um, given our relationship with NVIDIA. But one of the things that was interesting for me was, yeah, the first version sounded supernatural, like, you know, really, really fluent in conversation, but the caveat for us was, oh, actually, it doesn't it doesn't do to tool calling in the first version. And so for me, it was I it felt sort of interesting to me that choice. Right? Because we've been talking about how important natural conversation is for many, many years. Right? Like, for six years or more. But no matter how natural the conversation is, if it can't help the customer get something done in another system, that is always gonna be a reason for the customer to need human intervention to hand off the call somewhere else. Um, and so tool calling is sort of, like, so important to our use case at the moment. I was sort of surprised. I was caught off off guard that that model wasn't gonna come with tool calling in the first instance and just sort of reiterate to me how how difficult it must be to actually train these models for that level of instruction adherence with tool calling.
[00:21:40] Nikola Mrkšić: Yeah. I mean, I think it's just like a an entirely different vehicle that you're producing. Right? One may be like a racing car and, you know, it's so low to the ground that you can't really drive it in in a normal city because the smallest bump in the road is gonna send you flying. Right? And, equally, you know, if it if it can't even, like, do tool calling between, like, a set of defined things. Yeah. God forbid, like, some kind of computer use or anything like that. It's kinda useless. It's just showing off. Right? It's the moon landing before we have, like, an efficient way of going back and forth. Right? Which is, I think, important to keep us motivated to improve, you know, these regular regular models that can do this to be as natural as possible, but I think it's just one extreme. And I think it's the sexy one that researchers like working on. And it happens in these big companies from NVIDIA to Google, OpenAI where they're not really building things that are so actively used for real problems. And then you're just maxing out on the thing that is, like, the best the the most interesting thing to show off. Right?
[00:22:41] Michael Chen: Yeah. Exactly. I think it's, uh, speech to speech is still in all of my conversations in the market, everyone still talks about speech to speech, asks about speech to speech, there's issues speech to speech. But I think a lot of those use cases are maybe not the same as the ones that we hunt after. Right?
[00:22:57] Nikola Mrkšić: I mean, they are in that, like, you know, when we crack speech to speech with fully reliable tool calling, like, it will be the model to use. Right?
[00:23:04] Michael Chen: Yeah. That's true. Yeah. It's it's just so that everyone's yeah. What's the intermediate step there? Right? If in while we wait for this more instruction adherence or calling? I mean,
[00:23:15] Nikola Mrkšić: you know, one thing that gets me in I was talking to Sean and Eddie about this recently. But, again, like, when you, like, zoom in, you see that there are, again, multiple gates. Because even, like, the speech to speech has, you know, deployed in Chargebee's voice mode is very much a hacky way of, like, using other algorithms to stop the conversation and then kinda, like, start producing output. There is a version that is, like, yet more data driven, more in the spirit of deep learning, which just processes the input and decides to speak or not call, like, full duplex. Right? Confusing why people refer to it as, like, turn taking. What they mean is learning turn taking rather than having a very fixed turn taking pair of them. And then, like, they're, like, further stops as you zoom in, which are like, is it really just AGI? Right? Has it, like is it one model that elegantly sits there and does all that, or does it have a form of a harness around it? Because these are all harnesses. Right? And it's like stacks of, like, further layers of abstraction that simulate and approach, like, a model like property of the layer below without actually being a single, you know, kind of, like, quasi biological entity that reacts to inputs and outputs. Right? But it's
[00:24:30] Michael Chen: Like, reflexive. It's like, does that reflexes of its own? Does that intuition?
[00:24:34] Nikola Mrkšić: Yeah. And does it, like, continuously respond to the full set of input and output signals in a way that allow it to continue to evolve and change and learn? Right? Yeah.
[00:24:43] Michael Chen: Yeah. Yeah. So I think, yeah, the the yeah. That was a full duplex model was what was demoed there.
[00:24:47] Nikola Mrkšić: I didn't Yeah. Yeah.
[00:24:48] Michael Chen: Yeah. But, yeah, I think, um, I'm interested to keep an eye on that, and and as part of our collaboration, the media, so how we can, yeah, work together with them on on that type of future architecture. Right?
[00:24:59] Nikola Mrkšić: Yeah. We have an exciting announcement coming up related to that one. The next version of our model comes out, but we won't betray too much about that now. What else is interesting in the realm of token maxing?
[00:25:11] Michael Chen: It was, uh, just the incredible amount of, um, still, I guess, like, excitement how early everything feels. Right? Maybe it's like, you know, we we've been industry veterans, I guess, now, right, of this space, which is sort of odd to say. But everyone is still just getting to grips with more complex use cases. Right? I think a lot of people that I spoke with, they're only just trying to venture beyond a q and a style chatbot type of experience, right, with AI agents and, you know, talking through, well, what are you like, how do you stitch together multiple API calls validation? How do you work with multiple users on the same bar with permissioning, with different environments? I think many are still on the very early stages of a steep learning curve with adoption. Right? Hence, why there's still an incredible demand for tokens in the future. Uh, so I think that was yeah. That that stood out to me to me as well.
[00:26:14] Nikola Mrkšić: Yeah. Yeah. I think that, like, I'm starting to sound like a broken record with, like, a German paradox. But it really is incredible, like, how you know, rather than these things getting cheaper and then, you know, we settle at a certain point, we are just consuming more and more as we learn, and it feels like the workflows are changing. I saw a stat that there are more advertised jobs for software engineers than ever before. Yeah. And, you know, you would have thought that with this level of, you know, technology advancement that there would be fewer. But, actually, I feel like we're just starting to do more and more. And
[00:26:53] Michael Chen: Well, I can imagine the software developer role. I guess, right now, it's probably all centralized in the development team, but I could totally imagine a world where what, like, functions that probably didn't view themselves as IT or tech functions, you know, this idea of, like, a GTM engineer. Right? You're gonna have a compliance engineer, probably. You're gonna have, like, all these other, you know, functions that didn't see themselves as developers are gonna want, like, developer capabilities now. Right?
[00:27:21] Nikola Mrkšić: Yeah. It's like we're electrifying the whole thing with AI, and that's just Yeah. And I guess, like, there is definitely some ways to it. But I guess that's the beauty of capitalism that we're all, like, fighting, and we're, like, trying to do it at the same time. And, you know, we're probably consuming the same tokens doing the same thing that previously might have been SaaS, except innovation would have gone more slowly. And there would be time for one company to become, like, the dominant player with, like, a few 100,000,000 of ARR. You know, they're the ones producing that or that or that. Right? And I think right now in this, like, vibe coded world, if we trust ourselves to vibe code this and that, and I think for better or worse, people are doing it already. There will just be a lot of things that are done at scale and paid for by with a lot of tokens Yeah. That really should be something that's, like, an import this thing from this library done. Right?
[00:28:13] Michael Chen: Because I imagine, like, the whole vibe of the whole thesis of token maxing is, like, people just close your eyes, let go embrace it. Right?
[00:28:20] Nikola Mrkšić: Yeah. There
[00:28:20] Michael Chen: doesn't seem to be much discussion
[00:28:22] Nikola Mrkšić: around feel the AGI.
[00:28:22] Michael Chen: Feel the AGI. There doesn't seem to be I'm sure there's not as many discussions around efficiency and, like, how good you are. Like, I mean, is that something that you're like, how do you think about that in even in our own sort of engineering team? Are you looking at, like, capping people's token usage? Are you looking at efficiency of people's token usage?
[00:28:40] Nikola Mrkšić: Honestly, what I found is that either you're, like, a new believer or you're not. Yeah. And what I mean by that is the people who really get that, like, spark of curiosity you know, we'll talk about the chat GPT moment when you realize that it is something special. I think there's definitely this moment that's happened at a huge scale over the past months for a lot of people in tech, and they're starting to do very interesting things. Right? Like, in our quarter market, I can think of two people that have done more than everyone else put together. And they're not the obvious ones. They're not called GTM engineer. They're just the curious ones. They have, like, built it for themselves because they had this, like, voracious appetite to, like, do something they never could do before or something that they hate doing, but they've automated and then, like, how they saw it, and then you see the whole thing spreading. Right? I think the dumbest thing you could do is measure people's token usage and, like, demand more because I think people are gonna dress it up. Every incentive scheme can be gained. I purposely don't want to do anything like that because I want people to do things that make sense. I think the best ones inevitably now get, like, maxed out, like, almost every week unless they're doing something else or whatever. Well, there are times when they're not doing something that is so demanding that they do max out, so I don't think more is better. Mhmm. I think, like, there's definitely a pattern to those that use none of this, are going to get left behind in their fields. That's, I think that's how it is. I think that resisting it is futile. And what I've seen is that even those who resist, once they kinda, like, feel the AGR, see it from their angle and they understand something that they could do for them, then they feel it. And they're very much like a new believer. Right?
[00:30:24] Michael Chen: I had that moment when this idea of, yeah, building an AI agent with anything as the input. You just sort of throw an input file into our sort of agent development kit, and then Claude Code gets away and builds it.
[00:30:36] Nikola Mrkšić: Well, I think I remember it. Yeah. Like, it was probably, like, 4PM on a Monday. Yeah. Where you came up to me and you're like, hey. Why don't we drop everything? And Yeah. There was Just do this.
[00:30:45] Michael Chen: Code code I was like, oh, I I I understand the concept of, like, a code red moment. This is a code red moment. Right?
[00:30:50] Nikola Mrkšić: Absolutely. Right. It was incredible. I think the company reacted really well to it. But, you know, I think that, like you, Eddie, a few people were committed to like this. Right? It was madness in your eyes. It was like, this is it. Right? Like, you just know it to be true if you work with someone for as long as we have, or you I'm like, not like I'm happy for you. I'm like, I want to know what the fire behind your eyes is.
[00:31:10] Michael Chen: Yeah. Exactly. It was yeah. I think that was the first moment where, like, oh, like, this has been a real leap in since the last time I took a look at this capability. Right? And I think I I heard a great analogy recently, which is you're having now your own AI agent. Right? It feels like having your own, like, computer. Right? Well, at least how I imagine that must have felt. Right? And I imagine at that point in time, there were a lot of people that were charting things by hand or through typewriters who some people were more curious and got onto the computers. Other people were like, oh, no. But isn't it, like, slow and clunky? Or, you know, do you have Oh, there may. Yeah. That's, like, wired up in this way. It looks so complex. Right? But, yeah, I thought I thought a great analogy was, like, it does feel like, yeah, everyone's personal AI is gonna be a personal computer. Right? And one of the really interesting things for me was a reflection on that time in history was adoption of the personal computer, yeah, did start in the enterprise, in the business. Right? That's where people got exposed because of how expensive tokens are and, like, all the computers were. Right? You get exposed to it at work, and then you want it for your personal life. Right? Because you Yeah. And that away maybe a year ago.
[00:32:17] Nikola Mrkšić: Hours. Whoever puts them in at first, like, gets to to reap their their words. Yeah. No. I think you're right. And I think that another way in which it's similar to that is, you know, as we entered peak SaaS era, you know, as much as I think you and I have thought Google Slides versus PowerPoint games, we we lost that in the company. Right? Yeah. For now, at least, until Core came back to PowerPoint. Right? But I think there was this moment where everything started feeling very transient. You know? You could almost see a world where we were close to that, like, Chromebook being sufficient as long as you log in to your Google account. But then much like the kinda, like, computers of old where you needed your computer, because you've set it up with, like, your configuration, your settings, your files, your workflows. Like, people are figuring out how to import export things and how to, like, write skills files and all that. But the more you work, the more work you put in, the more it is yours, the more it reacts to you. And, you know, like, there was 1AM moment where my bot and my cofounder, Eddie and I, were sitting there. And he was looking at my bot, and we're both completely brain dead because it was 1AM. And I just heard him say the funniest sentence ever. He was like, he's so smart. And he was looking at my bot, and it was, like, powering through some workflow that I had done, like, 40 times, and his bot had done, like, once and was using the wrong approach for it. And I was just like, he's so smart. And I died of laughter. I was like I mean, you know, talk about anthropomorphic. And then we're like, hey. Like, this bot, teach that bot how to do it. And then you see the exchange and gradually hone in and you, like, refine the knowledge of why it was doing something better. And then, you know, as you do it a few times, you kinda, like, you standardize, and then you can export it to the whole team. They can do it. But the first few people kinda, like, breaking through the tunnel might do it very clumsily. Right? But then, like, if you're doing something for you that the whole org might not care about as much, you really care about that bot. It is precious to you because especially because of that context thing, it is built kind of like this weird multistory building that stands because you did something in a particular way. It may not have been optimal, but it has reached the height you needed it to reach. And, you know, you may not have another tool that does that. So when that, you know, Cloudbot dies because it's JSON for comms is misconfigured, you just suddenly feel like you're, like, like, your glasses have been taken away. Right? And you can no longer see.
[00:34:47] Michael Chen: Which speaks to, like I mean, one part that I thought was interesting about our agent development kit, at least the way that our internal engineers were using it, was they it allows them to, I guess, pull a version of our platform into their own environment. Right? Into their
[00:35:00] Nikola Mrkšić: own, like Yep. Yep.
[00:35:01] Michael Chen: How they have things set up the way that they like. Yeah. They can tinker it in their own environment. And then when they think it's ready, they can push it back out onto our cloud platform. Yeah. Right? I mean, that seems like a very not only powerful, like, way to do things, but also a very sort of common sense way of how this is gonna evolve. Right? Like, everyone's gonna have their own like you like you said, everyone's gonna have their context file set up in a certain way. Uh, an enterprise is gonna have their own environment where they would like to manage their knowledge. Right? So I I found that to be interesting that it was the power was not just that someone can log into our platform in a certain way, like an API, but that they could pull it into their own dev workflow in their own environment.
[00:35:44] Nikola Mrkšić: Well, I think it's, like, also the refinement of everything we've built over the years, which is, you know, one of the world's best enterprise platforms for cocreating conversational experiences. Right? So that whole bit about versioning, about visualizing, about testing, it's all there. Right? So you can take it away, do something great, push it in there, and it's still part of that, like, harness, if you will. Like, the the vernacular of how these agents are built and how we know them to work for enterprises and for teams collaborating. Right? So, you know, developers are kinda do developer things. They'll want things in a certain way. But it is, I think, exactly the limits of a harness and a framework like Agent Studio that define how something is to be done. And through that, it's almost like telling you what the really good and the really bad things to do are so that it can they kinda bend the output of your otherwise unconstrained workflow towards something that ends up pretty solid. Right? It doesn't commit any terrible crime against how these things should be built. But it's it was really interesting because, you know, it's not for lack of trying that our internal teams, let alone our customer teams, have ended up again going back to the code base. Right? It's because no one's really built a simple point and click thing Yeah. That even approaches the limits of the best conversational experiences you can build these days. Right? Some of it is in the model. Some of it is in the hardness. But the harness is code. Right? And, yeah, you can visualize it still. But if it's very powerful, visualizing it won't be easy, and then it's gonna look terrible, unreadable. It's not you know, how easy it is to consume visually is inversely correlated with its expressive power. So if you wanna have a really good agent, if it's also really easy to see what it does, well, one thing has to give.
[00:37:34] Michael Chen: Yes. One thing is like, in hindsight, it makes so much sense, but we've we spent so long, like, trying to find and push, you know, for something that could be represented in a great sort of graphical user interface world. But I think if you, yeah, if if you just take a step back, you realize that, like, how many powerful customer facing products are built based on, like, drag and drop. Right? And previously, it was the auto attendant in the telephony world. And, sure, you could trust a drag and drop interface to manage, like, press 1 for this, press 2 for that type of experiences. But if you wanted a fully intelligent representation of your brand speaking with your customers, I think yeah. It it I think it does make sense that the the visual like, the platform is for well, the the GUI is for visualizing to explain to someone what has been built. But, of course, a developer is gonna want really a lot of control over the complexity that needs to sort of live to to create that experience, right, in code.
[00:38:39] Nikola Mrkšić: Yeah. Yeah. I think yeah. That on the one hand, the physical impossibility of having both the expressiveness and the simplicity. And then the other thing is just the developers are developers. They hate the horn of drag and drop, and that just you know, means that they'll find ways. And especially today, the pendulum has swung the other way. Right? Because with these tools, like, you're just able to produce so much in code. And that's because they're better at coding now than they are at, say, computer use. In theory, they could like a drag and drop with any platform and create a thing of arbitrary complexity. Maybe they'd use even more tokens, but they don't work as well. And, clearly, it's suboptimal as well. Right? So
[00:39:15] Michael Chen: Yeah. I'm excited I'm excited to see, yeah, now what in this next wave now, it feels like yeah, there's we're in for a real acceleration in our industry, and then I'm excited to see where that where that takes us.
[00:39:25] Nikola Mrkšić: Absolutely. Well, Michael, thank you for joining me today. This was a very fun episode. And to everyone watching, please like, share, subscribe, and we will see you in the next one.