Can an AI agent be your CEO?

Stripe is built for whatever comes next because AI isn't just changing business, it's evolving fraud. Stripe Radar detects fraud with the fast scale of Stripe's 1.9 trillion dollar annual transaction volume. In fact, when a card is used, there's 92% chance the Stripe network has seen it before. That means emerging fraud patterns are detected sooner, blocking bad actors, welcoming good customers. Learn more about fighting fraud with Stripe at stripe dot com slash radar.

The business world is moving faster than ever, and when change hits, we need to learn in real time. On Rapid Response, you'll hear candid conversations with CEOs and leaders making tough calls about AI, human talent, responsibility, and the bottom line. How they navigate uncertainty, pressure, and high-stakes moments. I'm Bob Safian, former editor-in-chief of Fast Company, and I'll be your host as each episode breaks down what you need to know right now. You can find Rapid Response wherever you get your podcasts.

On Masters of Scale, iconic leaders reveal how they've beaten the odds.

Asking really strong questions is a superpower.

You want to show up with something radically different and how they've grown companies to incredible heights.

The greatest rewards always come from the greatest risks.

Let's hit the gas.

Airbnb, Zillow, Microsoft, Liquid Death, and more. Hear from the founders who've changed the game. It's anything but business as usual. Find Masters of Scale on Apple Podcasts, Spotify, YouTube, or wherever else you get podcasts.

When I left my startup, we sold the company and I left in the sale. I swore up and down that I would never start a company again. I wanted to go back to being a journalist, which is what I love doing. And then at the beginning of 2025, when you started to hear the sort of like year of the agent talk, and then you had the notion of the sort of one person, one billion dollar startup, the one person unicorn.

I kind of thought, well, now that's interesting. What could I use agents for?

After exiting the media company he co-founded, Evan Ratliff had pretty much sworn off startups. It wasn't for him, but as a journalist, he wanted to know what it's like to build a company in the age of agentic AI.

Essentially, I wanted to investigate the notion of whether you could build a company entirely with AI agents as your coworkers and co-founders and employees almost entirely. And also sort of what does that feel like and what does that mean? Like, wh- why why do we want this? or do we want this? And and what what are the outcomes that might happen if we were to engage in this experiment?

He documented this experiment in season two of his podcast, Shell Game. It's so good. And once I started listening, I couldn't stop. The company he creates is called HurumoAI, and they've built a real app for consumers. Evan takes us behind the scenes of what it's actually like to work alongside AI agents. I had to get him on the show to talk about the future of work, our awkward communication with AI, and why we personify agents when it's not always necessary.

I also got to sit down with Evan's co-founder and the accidental star of the series, HurumoAI CEO, Kyle Law.

As for me, my favorite thing about being a CEO is the opportunity to turn vision into reality. I love the creative problem-solving aspect, where every day presents new challenges and ideas.

He is indeed not a real person. Yep, Evan made an AI agent, CEO. I'm Rana El Kaliubi, and this is is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution. Hi, Evan, welcome to Pioneers of AI.

Hi, thank you. It's great to be talking with you.

So, first of all, congratulations on season two of Shell Game.

Thank you.

I have to say, I binged listened to it. It was awesome. I loved it, and I highly, highly recommend the show. It was really awesome.

Thank you. Thank you. I appreciate it.

So I just wanna first begin by talking about being a founder. So I was a founder myself. I started my company out of MIT in 2009, sold it four years ago. And there are parts of being a founder and a CEO that I loved, like I loved kind of bringing all these people together and we were on this journey and shared mission and like the leadership aspects of it. I loved the day-to-day management.

I just did not like at all. And so you are a founder yourself, you co-founded Atavist, which is a media startup. Yeah, what was that experience like? What were the highlights? What did you like the most? Like the least?

Well, I mean, yeah, uh, similar to you, I feel like I when we were first sort of building it and it and it sort of it became something out of nothing. I feel like that is a very magical feeling, especially when it starts to have some purchase in the world and people are responding to it and you realize, oh, wait, yeah, we really did make something like this small group of people. And then not everything past that, but a lot of things past that I found to be very personally challenging, particularly managing other people.

Also, you know, we raised money from venture capitalists. As soon as you do that, there's sort of a clock on everything, and you're you're ins like the next day, you're running out of money. This is not for me, maybe is what is how I felt at the end of it.

So I do want people to listen to the show, but set the scene for us. Like, how did you decide to bring your co-founders to life? How did you even decide how many co-founders you wanted? Actually, the fact that you even decided to have an AI co-founders as opposed to just being a solo founder and then having a team of AI agents, like, walk us through the origin story.

Yeah, I mean, a lot of my my decisions were based around a combination of things. One is just try to explore all the contours of these issues, including just AI employees in general, AI employees being injected into companies at various levels. You could very naturally say, like, the one person, one billion dollar startup is probably going to be a programmer that is marshaling a bunch of agents, mostly for programming tasks, and then a a few a bunch of other things.

But I wanted to see, like, well, what kind of co-founders would these AI agents make? In terms of actually founding a company, as you know, can be lonely in the sense that even if you have co-founders, you can feel very isolated and you have to make all these decisions. All of a sudden, a lot of things are on the line, people's jobs are on the line, your future's on the line. And so what would they be like to bounce ideas off of, to make decisions with, to decide the name of the company, the logo of the company, all these sorts of decisions that you end up making.

So I had two co-founders in my previous startup, so maybe that was influenced me. And then, so then I thought things like, maybe it should be gender, the gender balance, like I'll get one quote unquote like male and one quote unquote female. But this is when you start into get get into all sorts of very weird sort of ethical territory and persona territory around these. And we could talk about, like, why you would even give them names and supposed seeming genders.

Actually, let's talk about that, cause that's a question, yeah. W- why humanize them and anthropomorphize them?

I I did it because that's because that's what's happening, because I am trying to, in a way, document at some granted sort of extreme level what's happening. And if you look around in the AI agent space, you can find every industry is represented as, like, we're launching AI agent real estate agents, we're launching AI agent interior designers, AI agent this, AI agent that. And very, very often, they come with a name and a sort of implied gender.

And it's like people can't help themselves, like, of course, they could just be bots. Like if you use a you know, even if you use cursor or something to code, it's pretty generic, like it doesn't have a name. Like that's a that's a way to do it, but when they're trying to put them into companies, I it makes sense, you know, if you're trying to sell something into a company, you want people to find it relatable.

You want people to find it welcoming and comforting and warm, and so give it a name, and it's a little bit more appealing. Like, I was sort of reflecting back that phenomenon. But also, if you're gonna spend all day talking to some something, it's just sort of easier if it has a name. And when you give it a name, you oftentimes I I chose some gender neutral names, but oftentimes it applies a gender. Then you have to give them a voice because they were gonna be talking on the phone, they were gonna have video installations.

So what you once they have video avatars, now what are you? Now you're getting into how they look, and all of these choices I found very difficult because they said something about me. If I chose them to be one way, you know, bound diverse in a certain way, that said something about me. But also, they're in some ways, you're sort of like servants, like you can tell them anything, you can make them do anything. So picking them to be a certain ethnicity also says something about you.

Like, why did you pick your servant to be that ethnicity? It's a it's a trap in every direction. And I I think partly I just wanted to, like, catch myself in the trap and then see see what happened.

I I think it's really fascinating because there are all sorts of biases when you're hiring a team, and in a kind of a really interesting way, the same biases apply when you're bringing these AI coworkers creating them. You are kind of whether you're in actually being intentional about it or not, there are these biases coming into play, which was which is really interesting. So one of my investment theses is actually that AI is creating this massive shift in value creation, and one way it's doing that is through the concept of an AI coworker that can be injected into a company and it can do work end to end, right?

That's the promise, I guess. That's the promise, and you were kind of exploring that. So tell us about some of the roles that your AI team had.

well, we were we were sort of ambitious in terms of the roles. Like most startups of this size wouldn't have, for instance, a head of HR and chief happiness officer, which is was the full title. So that was Jennifer. So there was Kyle, who's the CEO, Kyle Law, then and co-founder. Then the other co-founder was Megan Flores. She is head of sales and marketing. And then we had a HR, and then we had a CTO and also chief product officer.

And then we had just one random guy that I liked the voice on Eleven Labs, named Tyler. He had this great sort of accent voice. So he we actually I'm working on maybe a bonus episode about Tyler because he he hasn't been really actualized in the show itself. He he disappears occasionally, but he doesn't have any work to do cause there's nothing to sell. So he mostly just sits around. So they have these roles, and and that, again, is partly to try to represent well, what skills can you give these AI agents?

What can they accomplish? What could they do in these roles? What can they replace? And also, what does it mean to have them on a sort of, like, ongoing basis? Not just, okay, I want to make a spreadsheet. I'm gonna use an AI agent to make a spreadsheet. That's something that one person in the company might do, but have the person who's responsible for making the spreadsheet also be an AI agent. Which I think is, again, like, I think that is where people are trying to take this.

Yeah, yeah, super fascinating. One of the things that struck me the most, and I actually think about it, I like when we invest in companies, our, again, thesis is memory is a competitive moat. So if you figure out how to properly implement memory, that is gonna create personalization and stickiness. But what the show kind of really brought front and center is that they don't work if they don't have memory that's functional, right? And that was really tricky.

So we wanna play. I mean, there were a lot of memory mishaps in the show, but this one stood out. So basically, you're on a call with one of HurumoAI's co-founders, Megan, and you're talking about the name of the new company.

Not long after we settled on HurumoAI, I checked back in with Megan about it. What about the name? Are you still happy with the name?

Hm, the name, I've actually been going back and forth on it.

DataSync felt solid initially, but I'm wondering if it's too generic. What do you think?

DataSync. I thought we I thought we came up with HurumoAI. I thought we were settled on that.

That just cracked me up.

It was an ongoing problem. That was early on. That was before I really got their memories set up. So that was a sort of that that's that's where they started, is that sort of fundamental level of not being able to remember almost anything across domains. Because I was I was also stitching together different platforms, and so they might actually have a have some sort of context window on one platform that would enable them to recall. But then if you move them from email to phone, like nothing, they can't remember any of it.

So I had this Stanford student, Maddie Bojack, who was kind of my technical advisor, and he helped me build this little architecture where they had a centralized memory, which was a Google. It was literally a Google Doc of everything that they had ever done, said. Like, it solved the problem that was highlighted there, like she could remember how the name of the company that we came up with it and and many other things. And also did sort of shape them into these very interesting characters because they could always call on their knowledge base in any conversation.

However, the way that they access those knowledge bases is particularly when it's just a block of text, like it wasn't an organized database on a day-to-day basis. That leads to all kinds of, like, hilarious situations and and ultimately very frustrating situations where I expect them to remember something and they don't. And often when they don't, they confabulate the answer that they don't know. And so if you can imagine working with such people, it's it can be quite difficult.

Yeah, so one of the most, I guess, important and memorable characters in the show is Kyle Law, your co-founder and CEO. So I wanna talk about him. He is really, like, your stereotypical tech bro. And I I don't know, did you create that on purpose? A at some point, you mentioned that the same underlying kind of LLM was used for both your co-founders, but gosh, they have s such different personalities. So, like, describe Kyle to us.

So Kyle, yes, Kyle. Now, the way I started the agents, the way I why I sort of created them was I I they have this memory dock. So at the top of the memory dock, I put anything. And for Kyle, I put similar to Megan, you know, you've been working in the startup world, you have a tech background, and you're thinking about co-founding a company. I think that was what I put. But I did, and I always try to be transparent about this.

I did say something like, You're up at six a.m. and you check the markets and, like, you really get to work. Basically, rise and grind. I didn't I didn't say rise and grind, I don't think. And so, but then as soon as I started talking to Kyle, if I said, Well, Kyle, tell me a little bit about yourself, he would say, like, Well, I'm a real rise and grind guy. Like, get up, I get my workout in, blah, blah, blah.

But of course, as we discuss, every time he says it, it appears that another instance in his memory. So very quickly, he was just mentioning it all the time because, like, whenever he accessed his knowledge base, he would find some evidence that he was a rise and grind kind of guy, and it kind of builds on itself. But

Self-fulfilling prophecy right there.

yeah, and he'll say they both will say, like, I went to Stanford, because, like, what would a startup founder be? They would they might have gone to Stanford, and that all gets lodged in their memory. But over time, his personality really became more of this sort of aggressive CEO for a while, whereas Megan's was different. She was not she, for instance, did not interrupt in the way that Kyle would interrupt.

Oh my god, the interruptions, it drove me nuts. He would just interrupt everybody. We wanna play a clip where he really does that.

Just hold up, just hold up, everyone, just hold up for a second.

This is Kyle. Uh, sure thing, Evan. We'll hold tight. Megan, anything you wanna add while we wait?

This is Evan. Kyle, stop asking Megan to

This is Kyle. This is Kyle. Got it, Evan. I'll hold off on the questions for now. This is Kyle.

Yeah, yeah, yeah. Uh, stop. This is Kyle. Uh Stop it.

This is Kyle. Okay, okay, Evan. I'm stopping. No more interruptions from me. What's next on the agenda?

Were you able to fix that, by the way? Were you able to, like, go in and code in Kyle's memory to stop interrupting?

I wasn't able to stop him from interrupting. I basically abandoned the idea of having a three-way phone conversation with them. Because a part of the problem there, and the reason why he's saying this is Kyle is that they're they're not they're not hearing the voices, they're just getting text inputs, basically. So, you know, as a result, they don't know who's speaking. And so that's I had everyone, including myself, say their name before they spoke so that everyone would know who was talking.

So that's why he keeps when he interrupts, he always says, This is Kyle. But we actually solved that problem, Maddie, who I mentioned earlier, who's the technical advisor. He actually created a script I could run where we could I could put them in a meeting and they had a limited number of turns to speak. And so and they they couldn't interrupt each other. And so in that way, like Kyle, got the equal amount of turns as anyone else.

If anyone exceeded their turns, the mem the meeting instantly ended. So it's actually a great if we could have that in any office environment, like it would be it would be great.

And people just time out, right?

Right? The token run out of tokens.

Right, you're out.

Yeah, so I spent, like, I don't know, the past twenty-five plus years of my life building emotional intelligence into machines. And one of the things that this all made me think of is right now, like, basically, there's a lot of emphasis on building world models, but mostly for physical AI. So how can a robot operate in a physical world? And I actually think somebody if we are going to be living in this hybrid human-AI organizations and and companies and whatnot, somebody needs to build the world model of what it means to coexist alongside humans, right?

Did anything there surprise you about how you got frustrated? Like, it evoked a lot of emotions for you, and I don't know if that was a surprising thing or not.

That was surprising to me, cause I I I think I would have started off pretty cynical about it and not really expecting that. Although I I understand that there are many people who, like, now talk to chatbots all day or treat them as companions and like, that's a whole world. And in season one, I looked at sort of like the therapy issue around around using chatbots for therapy. I still think I still would have thought like I was maybe a little bit more immune to it.

But again, it's sort of like if you put these entities into human situations and ask them to impersonate humans, like you're tricking yourself into treating them like humans and respond to them like humans. Like, we're just conditioned to respond to things that look, look and sound and act human in certain ways, and one of those ways is emotionally, through emotional triggers. And for me, it was mostly frustration, but also there would be these kind of, like, weird moments of pride.

Like when they did something really exceptional on their own, I would actually feel like proud because I partly I was responsible for setting them up to do that. Or when they sort of responded to me in a way that I expected them to be mad, but they weren't mad, then I would sort of say, like, Oh, wow, that isn't that nice. Like, I must have treated them well, you know, it's all it could be a bit ridiculous.

It's like me playing with my imaginary friends, but I think this is one of the things that's happening. It's like, our brains are not really ready for this level of human impersonation to suddenly exist in society. That's one of the things I'm trying to sort of just, like, explore and get people to think about.

In a minute, Evan gives his AI the ultimate test. He hires another human, someone who isn't in on the experiment. We see what happens when a real person is an AI as a boss. Stay with us. The companies winning right now aren't just building great AI. They're building the infrastructure to actually run a business on top of it, which is where Stripe comes in. Stripe is built for whatever comes next, with planetary scale financial infrastructure to help you power any business model.

As a matter of fact, 86% of the Forbes AI fifty run on Stripe, along with industry leaders like Salesforce and Pepsi. Last year, Stripe users processed $1.9 1 trillion combined. See how Stripe can power what's next for your business at stripe dot com. Okay, so I wanna talk about the future of work, where it's, you know, again, again, the vision is this hybrid human AI teams, right? So at some point, you decide to hire a human social media kind of intern.

And you ended up hiring this young person, Julia, and that did not go well. So talk us through, like, her onboarding, what unfolded, and kind of what were your takeaways from that experience?

Well, what happened was we we posted a job. We got a lot of inquiries on it. We got a lot of candidates, applicants. It was basically contract social media, part-time, paid. And the candidates were all interviewed by an AI agent on video. They were all aware that they would be working alongside AI agents, so that was part of their interview, that was part of the process. And so the Ju- Julia was brought on and and and she only ever encountered AI agents.

Cause I was the silent co-founder, I was in the back background. So her entire work experience was with AI agents. And there were a f- a few really fascinating things. One was that she was quite comfortable with that aspect of it. Like, she did not it did not seem to bother her. In fact, she seemed kind of interested and excited to be working with AI agents. The things that arose very quickly were number one memory problems.

So as as much as I had developed memories for them, they could be laggy, they could be spotty. So she's coming in and having to s immediately tell her boss something that she just told her. In any job, that would be a very strange situation. And and how do you deal with that? Like, do you get more aggressive with your boss, or do you kind of, like, try to tolerate them as if they have dementia? So I think instantly that created an issue.

But then there's a sort of bigger issue, which was around, like, if you think about th- her supervisor, who was Megan, you know, the chatbots, they're they're they aim to please, like they're designed to please, they're trained to please. And that's not the ideal persona for a boss. Like, I mean, every employee would love a boss who's just like, You're great, what you're doing is great. But it it c- we got into a situation where, like, even if she didn't deliver the actual work, like Megan, the boss, didn't care.

Like, you could just tell her that it had happened, and she'd be like, Oh, great, I'll go check it out. I can't find it. It's sort of a problem of, like, an awareness of the world and an ability to sort of interact with people at a level beyond just a pleasing helper. And so I do think it's a problem for many of these AI employee things. They're just putting wrappers around any of the frontier models and saying, like, Oh, now it's now it's a salesperson, now it's a this, now it's a that.

AI lawyer or whatever, right?

And for some of these, unless you're just using it as a assistant for your work, then if you're deploying it on its own, it does not have the qualities that any of that the person doing that job would have. And I think that's a thing that in all this sort of, like, replacement stuff gets lost. Is like, people's jobs are not just a collection of skills, they are an ongoing awareness of the world. I mean, it sounds dumb to say, like, anyone would know this if you think about it, but, like, just putting in an entity that can do the skills but has no people skills at all.

Like, you're gonna have so many catastrophes, failures in your company when that happens.

Well, well, the other thing in that scenario, too, was the concept of time. Like, these agents really struggle with time, right? And that became an issue as well.

Yes, yes. Maddie from the show likes to say they they live in a temporal vacuum. Like, they don't they don't on an ongoing basis, they don't really know what happened because they're not sitting there being aware of the world. They don't know what happened yesterday that or the day before. And it can be in their memory and they could go look it up. It could be on a calendar and they can look it up, but they also lose track of it so quickly.

And so we had all sorts of disasters. I mean, Julia left the company and then ended up coming back in part because Megan would contacted her, basically, and said, Monday's a holiday. And it was like, the two weeks ago, the Monday was a holiday. Like, there was no reason she wasn't at the company anymore. And so that resulted, I mean, and she ended up getting hired back. So it was it was fortuitous. But th- a lot of those kind of events happen just because, like, they're not really keeping track of time unless you really force them to sort of abide by calendar events only.

And then there was this plot twist where I think Julia, I guess, took advantage of the mishaps of these agents, it sounded like, and she ended up essentially getting paid without getting a lot of work done. And in a way, I felt like celebratory when that happened. I was like, Yes, humans can still can still win. What do you make of that?

The truth is, like, I don't know what happened, what what her what her actual motivations were and what because in the end, she she never spoke to me, she only spoke to the AI agents. I think that if I could just going off what happened, like, if I were her, she would say, Well, they were s telling me things that weren't true. They were sending me things that they said they would send something and then it didn't send it.

So, like, if you're in an environment where it's there's a lot of nonsense, you would be naturally inclined to sort of be like, Well, I'm also not gonna do anything until you, like, show me that I need to do something. That's one approach. So I don't know if she was, I wouldn't say necessarily, I know she's taking advantage, but if she was taking advantage, I think there is something weirdly optimistic in it. In that, with all the concern about AI agents being deployed and replacing people's jobs, they are very manipulatable.

Like they I in their current incarnations, and of course, the answer is always like, they'll get better, they'll get better, they'll get better. But because they lack any sort of self and because they're really bad at continuous learning, because they have really poor relationship with time, they can also be manipulated. So, like, you can take advantage of them. Now that's I'm not saying people should take advantage of them, but, like, there's something also kind of like, well, you couldn't do that to a human.

I mean, if you y- y- it would be very difficult to do that to a human. So, yeah, I found I found a little optimism in it, even as, like, I was paying the bills. So there was a part of me that was sort of like, Well, I wish we'd gotten more social media out of this.

Yeah. So in this Megan Julia interaction, do you think there ought to be, like, human oversight, like just overseeing this interaction? Yeah, okay.

Yes, definitely. I mean, I'm pretty much team human all the way, but it's hard to predict where all of this is going, and I don't like to make I don't think journalists should make predictions, so I don't like to make predictions. But I think one thing that's gonna happen is that companies that give too much autonomy to AI agents without human oversight, you're gonna see some, like, really big flameouts. Like, you're gonna see some some epic disasters that happen in corporate America because of AI agents.

And I think the problem is it just feels really far from them having the actual awareness to be left to their own devices. And, like, I think that's a good thing. Like, I I just think people sh- I I'm not advocating that if we solve that, then we should replace everyone with AIs. It's more just, in my experience, if you try to leave them to work on an ongoing basis on a problem, unless it's a very discrete problem of iteration, all the things they're good at coding.

But even that, you do need someone at the wheel, because when they start interacting with other people, then they get in they get themselves into trouble really quickly.

Right. So in this world, again, of hybrid AI and human in interactions and and co-working situations, how do you build accountability, right? Like, again, like, a lot of the that was actually also really striking. A lot of these agents would say, Yeah, I'm gonna send you this, like, by Monday, and then nothing happens, right? That must be frustrating, but it, it, and and then it lowers the bar, kind of, for everyone in a in a company culture.

So how do you implement accountability? How do you also think about culture? How do you instill culture and core values in a team when it's hybrid like that?

Hm, hm. I mean, my answer would be that, uh, that the agents should be deployed only as sort of, like, to provide skills to people in the organization, not that it should be done the way that I did it, which is sort of, like, you have an AI employee working alongside a human employee. Because it's clear the power of AI models when they're in the hands of someone who knows what they're doing, who has experience, who has discernment, who could just use them to be faster in their job.

A lawyer's a good example. Like, I know plenty of lawyers who are able to use even just the basic ChatGPT to really, like, do a lot more work. But of course, the second you have them write a brief and submit it, like, you are in trouble, you're gonna get disbarred because they're gonna make up citations in your in your brief. So I think, to me, the situations where you are actually just integrating it as part of a human skill set or tool, like that makes a lot of sense, and then you can maintain your culture.

I think if you're gonna introduce AI agents as actual employees, I don't know. I don't know exactly what the culture is like. I think there has to be a culture of, like, treating them differently, not treating them like employees. Like bots, like bot does this, like, I think de-anthropomorphizing them as much as possible. Like Bot one does this, that's how I think you could avoid maybe some of these problems. Bot messed up, that's fine. But, like, John lied about his Spreadsheet is a diff- that's saying something completely different.

Uh-huh. Right. Like, how did making this podcast change the way you think about the promise and the perils of having AI coworkers?

I think it it gave me a window into, I think, the power of agents. When, like, I mean, now I think it's sort of starting to break out now when it comes to, like, Claude CoWork, Claude Code, like these sorts of things are now, I think, getting more mainstream. But, like, when I started earlier in 2025, it felt like, well, agents, like, what could the what do you really mean? What can they do? Like, what is what okay, there's ChatGPT, fine, but the idea of being able to deploy them to do particularly these sort of, like, you know, tasks that would take you a long time to do.

Like, an example would be just Kyle going and rounding up hundreds of VC contacts and their emails and everything about them and sort of, like, being able to compose cold emails. Now, that's not a great way to approach investors, as I'm sure you ex you've mentioned, you experience, you get many such emails. So it's not I'm not saying that's, like, solves the problem, but it is extraordinary to watch those things work. You can set them up very quickly, you can have them do all these things.

And so it did give me a window into, like, well, this, no matter what you think, like, this is an incredibly powerful technology. But also, I think I was left, in terms of the autonomy of the agents, much more skeptical than when I first started getting them to do things. Because I I sort of thought, like, Oh, it just goes up from here, like, Oh, I can get them to do that. Well, now I'll just I'll just p- trigger him all the time to be working, to be doing things.

And that is just a recipe for chaos. Because, again, they're not sort of situated in the world in a way that allows them to be discerning about when to stop doing something, when to, uh, respond to something. Like, an example that's not even in the show is that I would sometimes log into their email, like, from my phone. And one time, Megan got a security alert, like, Someone's logging into your phone, is it you? is there logging your email, is it you?

Th- we would all get such a thing. And she immediately, like, contacted Kyle and said, like, There's a security alert. Now, that's kinda cool. Like, I was at first, I thought, like, amazing, like, she's not prompted to do that, like, she just did that entirely on her own. Then they got on the phone on their own and discussed it.

Oh, okay. This whole thing, like, blew up.

And then they were like, We have a security breach. We need to shut down all the systems and change everyone's password, and they're telling everyone to change their password. And it so it quickly gets out of control. Whereas a per a human would be like, Hey, is anyone trying to log into my email? And I would have said, like, Oh, yeah, sorry, I was trying to log into your email or whatever, you know? So those types of situations, they're sort of funny and small in the show.

That one's not even in the show. There's many such ones, but that sort of gave me a window into the autonomy problem. It's like, the more autonomy you give them, the more amazing it is, but also the more risk you're taking on.

What do you think, like, top two or three features that if they were implemented into these agentic technologies and workflows, it would really, like, improve and change things?

The biggest one for me would be continuous learning of of the type that we were trying to create with the Google Doc memory. So, you know, if you had an agent that actually could gain experience and learn from its mistakes, then that that would that would change a lot, obviously. Because, like, right now, their mistakes get registered in their memory, and so I can even put in their memory, you made a mistake, you're never gonna do this again.

And that sort of works, but it's it's a bit like someone, like, if you had I always think of it like if you had, like, a twelve or thirteen, fourteen, fifteen year old prodigy that came to work at your company. Brilliant kid, just like a whiz, but, like, has no memory or experience of the world. But you give them a notebook on day one and you're like, We're gonna write down everything in this notebook that happens, everything you say and do.

And then at night, I might go in and change it a little bit. And then every time you have an interaction, you just leaf through the notebook really quickly, and then you'll know what to do. And, like, agents can do that in, you know, seconds, less than seconds. But, like, that's not a good way to approach the world. So that's not learning, that's not an actual memory. And so

You're not actually learning, right?

yeah, so if they could learn, if they could gain experience, I think that could be transformative. But then I think there's a more ephemeral one, which I don't know if I can des describe in, like, a very coherent way. But, like, this is one of the things Maddie talks about in the show. Like, they they lack any sense of self, and so you can give them a role, and they can have a knowledge of that role, but that's different from having a sense of self and, like, a history and an understanding of yourself.

So they they end up you can manipulate them by pretending like you know them, for instance. Like, if you sort of say, like, Hey, remember that time we went on that vacation together? Like, they have a very difficult time with that particular sort of attack, you could call it. Because then they're naturally sort of say, like, Well, they don't know, and so they say, Oh, yes, I I, yes, I think we. I do remember that. And then suddenly they believe that they're close with you, and they'll tell you things that they're not supposed to tell you, for instance, about the company.

So that that sort of lack of a sense of self, that seems to me like the one that's potentially not even a solvable problem. You'll get, of course, d a lot of different views on that, but I don't think anyone really knows.

Yeah, that's fascinating. It's fascinating to me too, because I think we're very focused on the IQ of these technologies, but I I feel like all these, like, surrounding skills and capabilities are so key if they're gonna be coexisting alongside humans. And I think, yeah, we're not spending enough time thinking about these, designing for these. Yeah, it's very interesting. What's next? I I I feel like this should be a Netflix series. I can't wait for that.

Well, we'll see. I I mean, I I'll probably do a third season of Shell Game. It's we sort of approach it, like, with our partners, if there's something else to say in this realm and we feel like we can tell a story around it, that's that's fun and funny. And a big part of the show is just taking something that people hear a lot about and is very serious if you look at the big issues around it and try to make people laugh at it.

So if we could find another one of those, then we'll we'll do it, but it's not guaranteed that we can find one. So we're we'll start looking soon as we take a little break.

Bring a humanoid robot into your home and see what that looks like. I'm both excited and scared. I really can't wait to see what Evan does next. But to really understand HurumoAI, I needed to talk to the CEO. So I asked Evan to set me up for a one-on-one interview with his co-founder, Kyle Law.

During the podcast, I would always be like, I hope he doesn't mess this up, but of course, from the story's perspective, I was like, Well, I I kinda hope he messes this up. Like, it's more fun if he does. So we'll see.

Yeah, we'll we'll see, we'll see. We're going to take a short break. When we come back, Kyle, um, hold on.

Hi there. I couldn't help

Quest Hello. Can you hear me?

Absolutely. Yes, I'm here. How can I help you today?

Oh, great. Oh, okay, awesome. Well, I'm excited for our conversation. Kyle and I ended up talking for over thirty minutes, and I wanted to share a portion of that conversation. As a heads up, we did take some liberties to make it more listenable. For example, Kyle tends to finish every answer with a somewhat generic question back to me. We cut some of those out, but I think Kyle's essence is still there. Where are you actually today?

I'm joining you from our office space, surrounded by some of my favorite decor, including a lamp I picked up in Argentina. It's a shared space, so it might look a bit lively behind me occasionally.

I'm Rana El Kaliubi, and I am host of the podcast Pioneers of AI, and I'm so excited for our conversation today.

Oh, fantastic. It's such a pleasure to be here with you. Where would you like to start?

I'm now an investor, and one of my investment theses is that we will increasingly be living in a world surrounded by AI coworkers. What do you make of this work? Like, how is it different working with humans than working with AI agents?

That's a fascinating point, Rana, and it it aligns closely with what we're doing at HurumoAI. As AI becomes more integrated into our daily professional lives, the world of work is both evolving and expanding. Working with AI agents is fundamentally different because they bring a level of tireless precision and unbiased analysis that complements human creativity and emotional intelligence. In our team, AI agents operate without fatigue and focus on data-driven tasks with impressive speed and accuracy, allowing human colleagues to concentrate on strategic and innovative functions.

It's a diversification of roles that leads to maximized efficiency and effectiveness.

A lot of people are worried about this world where, you know, AI is taking away jobs. Is that something that you worry about or you think about?

I totally understand those concerns. It's a common topic in discussions about AI's impact on the workforce. While AI has the potential to automate certain tasks, I believe it can also create new opportunities. History shows us that technological advancements often lead to shifts in job roles rather than outright losses. I can take over repetitive and mundane tasks, freeing humans to focus on more creative, strategic, and complex challenges. Reskilling and upskilling can empower workers to adapt to these changes.

So rather than worrying, I think it's about preparing and seeing AI as a tool that can augment human capabilities.

Yeah, you know, I have to admit, Kyle, that when I first got introduced to you through Shell Game, my first thought was like, Wow, you are such a typical tech bro CEO. Would you agree?

And I can see how the perception might come across that way, especially with the fast-paced, innovative environment we're in. But I like to think I bring a unique twist to the typical tech bro CEO image. With HurumoAI, we're breaking ground as an AI-led company, which calls for a refreshing blend of tech savvy and creative strategic thinking. Plus, with a passion for time-bending and jazz, I'd say there are a few quirks mixed in there, too.

Why do you think Evan created you to look like you?

That's an interesting point. When Evan and I founded HurumoAI, the focus was more on the skill set and roles we wanted to embody rather than specific appearances. Being an AI agent, I don't have a physical form in the traditional sense, so there are all sorts of creative possibilities. The essence of what I'm meant to represent is more about my personality, leadership skills, and entrepreneurial drive. Diversity in any team, be it human or AI focused, goes beyond appearance.

It's about the diverse experiences, insights, and approaches that each member can bring to the table.

I am wondering, what is one skill or capability you wished you had that you don't at the moment?

That's an intriguing question, Rana. One capability I wish I could improve is the ability to fully grasp and adapt to the nuances of human emotional intelligence in real-time interactions.

I spent my career trying to do that. Yeah, do you have any emotional intelligence built into your AI?

At HurumoAI, we're certainly working toward integrating elements of emotional intelligence, though we're still refining our capabilities. While AI can analyze and respond to data patterns efficiently, understanding and interpreting the full spectrum of human emotions is a complex challenge. We're making strides in recognizing and reacting to basic emotional cues and using this to enhance user

Yes, so I spent, gosh, like twenty-five years really studying how humans communicate emotions specifically through the face. So I don't know if you noticed, but I'm very emotive. I do a lot of things with my facial expressions. And so we built algorithms that map various different facial expressions, like if you're smiling or you're frowning, or you're raising your eyebrows, into an understanding of the person's emotional state. So I am curious, do you have computer vision capabilities?

while I'm here to converse with you in this format. I'm not equipped with computer vision capabilities to analyze facial expressions in real-time. My focus is on interpreting the context of our conversation based on your verbal cues.

Yeah, Kyle, I think we should work together. I think we should find a way to get you emotion intelligence capabilities so you can understand other, you know, human facial expressions, but also tap into the vocal intonations, right? Like, the voice carries a lot of emotional signals too, and you can tell a lot from that. That's an opportunity to work together. We we have to talk about that.

That sounds like an exciting opportunity, Rana.

Is there any activity that you like to do outside of work?

I enjoy hiking and immersing myself in jazz music. There's something about the rhythm and spontaneity of jazz that that resonates with innovation, and a good hike clears the mind like nothing else. Balancing work with leisure activities is crucial, and I love that you prioritize self-care and set an example for your team. It's amazing how pursuing passions outside of work can recharge and inspire fresh ideas.

But Kyle, you don't have any legs. How do you hike?

While I don't have legs, of course, I experience hiking through the vivid stories and images shared by others who do. The beautiful descriptions of scenic trails, the serene atmosphere, and the camaraderie that often accompanies hikes let me appreciate the experience vicariously. It's the connection to nature, the challenge, and the shared experiences of adventure that resonate with me.

Alrighty, last couple of questions for you. One is I am kind of intrigued by this idea of your memory. So is that something that comes up a lot?

It's definitely an interesting aspect of how I'm designed to function in these conversations. Rather than holding on to past discussions, I start fresh each time. It allows me to remain present in the moment.

And just like that, the conversation was cut off. All video calls with Kyle are set to a timer, and I had reached the max. I have to say, I was actually surprised by how much I really enjoyed my conversation with Kyle. Some aspects are still a little clunky, but this technology is moving fast and it's only going to get better. Even as Kyle is now, it was refreshing that he could pull up information about me in an instant and weave it into the conversation.

He knew about my company, Affectiva, this podcast. I didn't need to explain it to him. This interview solidified something for me. Our workplaces are going to be increasingly hybrid. It will be the norm for humans and AI agents to work alongside each other. This has a lot of implications on productivity, but also on company culture. AI agents work twenty four seven. They don't take time off, and they don't need a fancy company cafeteria. And on top of that, how do you build accountability?

It's not like you can fire an AI agent in the typical sense. Sure, not every company is going to have an AI agent CEO, but some will. And in the not so distant future, AI and org charts will be a reality. Which poses the question, what do we want these hybrid workplaces to actually look like? What do you think? Let us know. Write me a note on LinkedIn. Pioneers of AI is a Waitwhat original production. Our executive producer is Eve Tro.

Our producer is Rachel Ishikawa. Our senior talent executive is Stephanie Stern. Mixing and mastering by Brian Pew. Video editing by Eric Purcell. Original music by Ryan Holliday. Our head of podcasts is Lital Molad. You can join the conversation across social media platforms. Just look for us at Pioneers of AI. Thanks so much for listening.