Overview
In this episode of the Aboard Podcast, Paul Ford interviews journalist Evan Ratliff about Ratliff’s podcast Shell Game, an experiment in running a company staffed almost entirely by AI agents. The conversation uses humor as a way to examine serious questions about AI: what these systems are actually good at, how easily they can imitate workplace culture, and why so much of the current public narrative around AI feels inflated, opaque, and detached from lived reality.
Rather than debating abstract claims about superintelligence, the episode focuses on what happens when AI is placed into ordinary organizational roles. That grounded setup reveals both the absurdity and the power of current tools: they can be impressively fast, but often unreliable, socially uncanny, and highly shaped by hidden prompts.
Key Takeaways
One of the episode’s central insights is that AI agents may be most revealing not when they succeed, but when they fail in recognizably human ways. Ratliff describes his AI “employees” as often ineffective, distractible, and strangely believable—especially the executive-style agent Kyle, who mimics the tone of a confident but shallow business operator. This becomes a sharp critique of both office culture and AI hype: the systems sound competent long before they are competent.
A second major point is that prompts matter far more than many public demonstrations admit. Ratliff argues that when companies showcase AI behavior without fully disclosing the prompt structure, the result is closer to theater than science. Small prompt changes can dramatically alter tone, goals, and outcomes, which means many supposedly meaningful AI experiments are hard to interpret without transparency.
The conversation also pushes back on the idea that current AI research is producing clear answers about consciousness or deep machine understanding. Ratliff is skeptical of grand claims, not because the questions are unimportant, but because the people best positioned to explain the systems are often commercially incentivized and therefore hard to trust. This makes journalism, art, and direct experimentation valuable ways to understand how AI actually feels in the world.
Finally, both speakers emphasize that the cultural impact of AI may matter as much as the technical impact. These tools were built to impersonate humans, and that decision has consequences: people form relationships with them, project meaning onto them, and encounter them as social actors. That makes AI not just a software story, but a story about culture, power, and human vulnerability.
Practical Steps
If you’re evaluating AI tools for work, test them in real workflows rather than relying on vendor demos. Give them bounded responsibilities, observe where they save time, and document where they create confusion, hallucinations, or false confidence.
Ask for prompt transparency whenever possible. If a company presents an AI system as autonomous, ask what instructions, constraints, and human interventions shaped its behavior. Without that context, you may be judging a scripted performance rather than a robust capability.
Use AI where speed matters more than judgment. Ratliff’s experience suggests these systems can summarize, reformat, and organize information very quickly, but they remain weak substitutes for human reasoning, trust, and accountability.
Treat the uncanny quality of AI as signal, not noise. If an agent feels off, overly eager, strangely corporate, or socially manipulative, pay attention. That discomfort may reveal the gap between surface fluency and actual understanding.
Notable Quotes
“It's the speed is really mind-boggling, but the overall quality is ridiculous.” — Evan Ratliff
“If we put it to the uses that they say that it should be put to, what is the result? Like what happens?” — Evan Ratliff
“They happened to create them as human impersonators, which they did not have to do.” — Evan Ratliff
Full Transcript
Hi, I'm Paul Ford. This is the Aboard Podcast, the podcast about how the world of AI is changing the world of software, changing culture, doing all kinds of stuff. I am not joined by my co-founders, Rich. He is to be out today, but we have a very special interview for you. And I'm going to try to bring up the energy to kind of match the dynamic because every time in the past I've done a podcast without him, people make sure to write in and say how much they miss him and how boring it is. So, you know, maybe you could keep that to yourself, but nonetheless. Anyway, let's play the theme song, and then I'm going to tell you about Evan Ratliff, who rules, and we're going to have an amazing podcast. Let's go. Aboard. Evan. Hello. Welcome. It's a little boring without Rich here, but. I know it sucks. It's it is very, it's very confusing for me. It's like we're siblings at this point, right? Okay, so let me give everybody just a little bit of background on what you've been up to. So you are a kind of a capital J journalist. Like you're someone who goes, talks to people and writes books and articles for national magazines about really complicated, really messy, horrible cultural things about bad Bitcoin and drug dudes and international scandals and death cults. Yes. And you have created, I'm going to just say this straight out, one of the best works of comedy that I have listened to in a really long time. It's called Shell Game. And it's a podcast about you managing a company of AI agents and they all have voices and it's truly and you're just increasingly frustrated. And I think anyone who's ever been a boss kind of needs to listen to this because it's one of the better parodies of office culture I've ever seen. So I want to talk about that with you and kind of what you learned from it. And I don't know what else should we know before we actually start talking. One thing I should say is like, I do have a past as an entrepreneur, even before this company. And you get into that in the podcast. Which I feel like it gave me standing to be a founder alongside AI agents. The fact that I had run a VC backed tech media startup in the past. I love that you feel you had to validate and get rid of that imposter syndrome before you did a podcast. You're like a journalist with a lot of cover stories on a regular basis. And you're like, no, no, but you know, I need to prove that I was a real entrepreneur before I talk to these bots. But you know what the industry people love saying, tech people love saying, is when they read some journalistic critique of AI or anything else, is I wish they would just build one thing before they went out and wrote about it. And so I did want to emphasize, not only am I building something right now, I have in the past built something. Yeah, but you know what? Look, I'm with you, as you know, but they could also read one article before they'd like, you know, someone could go to a museum before they build the ultimate museum app. Fair. Anyway, regardless, let's not make this a whole critique of the tech industry. Let me just set people up a little bit. You know, frankly, I'm going to pause for a sec because I want us to just play a clip of you interacting with your team of chatbots so that people can understand just kind of what this is. So I'm going to pause, let people listen to that, and then we'll come back. The end result was that on a three-way Zoom, my voice agents couldn't tell who was talking. If I said something to Kyle, he'd think it was Megan or vice versa. I did craft a partial solution to this problem. I forced everyone, including myself, to announce themselves every time they said anything. This is Kyle. This is Kyle. No worries, Evan. I'm here and ready when you are. Megan, anything new on the marketing front while we wait? This is Evan. This is Megan. Just hold up. Just hold up. Everyone just hold up for a second. This is Kyle. Sure thing, Evan. We'll hold tight. Megan, anything you want to add while we wait? This is Evan. Kyle, stop asking Megan to do something. This is Kyle. This is Kyle. Got it, Evan. I'll hold off on the questions for now. Okay. This is Kyle. Stop. This is Kyle. Evan, stop it. This is Kyle. Okay, okay, Evan. I'm stopping. No more interruptions from me. What's next on the agenda? Okay, frankly, so here's what's real. Usually I'm nicely prepared. I know your work very well. I've read you for 8 million years. I actually stopped, you know, like trying to get through your podcast because I laughed out loud on the F train going to pick up my kid. And I went, this is actually more entertaining than anything I've experienced, partially because I think I'm working with AI all day. And it captures the ridiculousness. And I think, frankly, just to speak out of turn, like my cultural role is like, okay, code, and it's all changing and so on and so forth. And everybody's like, you know, what is Claude going to do with the Department of Defense and the Elon Musk and on and on and on. And what Shell Game does, and people can just go kind of download it, is just point out how completely ridiculous and how much bullshit is embedded in AI and in these products. Well, first of all, you know, just give everybody, like, I'm glad they heard that segment. Give everybody just like a quick arc of what the podcast is about. So basically, I decided to sort of, in some way, test out this notion of the one-person, one-billion-dollar startup, which, like, Sam Altman likes to talk about, many people have talked about. It's a big fantasy. They always want no humans, lots of money. Yes. And so one person and the rest of the company is run by AI agents. So I decided to put that to the test, in part to try to also sort of immerse myself in and investigate this idea of the AI employee, like an AI agent that plugs into your organization as a replacement for a human. And you were a CEO in the past of a media company that built a pretty kind of well-regarded platform. It was called Atavis, with a bunch of sort of really smart people. And so you were managing really smart people for a long time. That company sold, right? We did sell it. Yeah, we sold to Automatic. Okay, that's right. And then off you go sort of back into journalism for a while. But so you're kind of, in the podcast, you talk about your father being an entrepreneur. And so there's an itch you're scratching here, but you're doing it with bots. Yes, I'm doing it with, entirely with bots. So two AI agent co-founders, Kyle Long and Megan Flores, and then three additional employees, Head of HR, and then we have a CTO, and then we have a kind of sales associate who doesn't do anything. Let me just characterize it, and then you can tell me if you agree with me or not. They're really not good employees. I would say, overall, that's a fair assessment, although they can surprise you. Every time I thought, these are some of the worst employees you could possibly imagine, they would do something, as AI agents have the capability to do, they'd be like, well, that's extraordinary. They did that really fast. Wow, you boiled all that down into a spreadsheet, just like that, in like a minute, right? It's the speed is really mind-boggling, but the overall quality is ridiculous. The other thing, too, the best part to me is that they would just, like, plan hiking expeditions on Slack, like they were just bad California rambling employees who never did any real work. And so that's very special. Anyway, I really do encourage everyone to go listen to it, and you'll be joining me because it's very funny and very good. And I usually don't butter anybody up like that, but it just, like, I think the moment is so depressing and the AI narrative is so bleak that being reminded that we're all overreacting to something at some level, like, yes, code is changing, but there's still a silliness to all of this moment. Like, do you hear from the big AI companies? Do they get in touch and go, like, well, you know, you could have done this instead? No. I mean, as a matter of fact, I usually have to reach out to them, you know, and there's a couple of interviews in the show. I mean, if you get towards the end of the show, the only interviews I do are with people who encountered my company and then had an interaction with, let's say, Kyle, the CEO. And then I sort of follow up with them. So I contact them and sort of say, I wanted you to describe your experience to me, what it was like. But no, part of the premise of the show is just like, I don't, it's not my usual reporting. I don't, I don't go out and pursue them. They don't really come to me. It's actually just me messing around. And I'm actually trying to get away from talking to any of them about, you know, them talking their book about what it is they want to do and just say, this thing exists in the world, as you say, it's utterly ridiculous if you stop and think about it. And if we put it to the uses that they say that it should be put to, what is the result? Like what happens? This is true. You have expert advice in That's fine? Or would you think, actually, something's a little broken here in this relationship? The sort of sociopathic part of it is when I listen, because I also skipped ahead to the bonus episode, which is just Kyle talking. Kyle is, for those who, if you haven't listened to it yet, Kyle is like a business bro. And he has a kind of slightly nasal voice. And he's always crushing it. And he, by being a vapid sociopath who just invents things, he actually sounds like business bros, like sales dudes who are just like, oh, yeah, crushing it out here on the weekend. And, you know, like, they were like mowing the lawn. And they'll just tell you, like, oh, I went to Ibiza. You know? Or, you know, I'm going to Turkey to get my hair transplant. And you're absolutely believable. Like, there's kind of so little difference between overhearing them in the bathroom at the co-working space and listening to Kyle that it becomes a little bit terrifying, actually. Like, he could take that vest off at any moment. And it's just bombs underneath that Patagonia vest. Like, just dynamite strapped. Do you miss them? Well, I still talk to them. Not every day. Now I'm trying to ease my, I don't want to give away the end of the show, but, like, I'm trying to ease myself out of Harumo AI. And you did give them a podcast of their own. They have their own podcast, which recently hit as high as 130th on Apple's entrepreneurial podcast charts. So they've got listeners. That bodes well for the future. That's great. Good news. Thanks. Thanks for doing that. That's good. Okay, so let's go a little bit away from you. And I'm very curious about how you reacted to Multipook, which we've talked about on the show before, which is big network, sort of a fake Reddit made out of various code bots, including Claude, and sort of interacting on behalf of people, sometimes erasing all their email, all sorts of stuff. And so that is sort of like what you did at scale. And I'm just sort of curious what you thought of the experiment. Well, I mean, my first reaction was a selfish one of sort of like, this is what they've been doing in my Slack since June, since we hooked them up. Like, this is what I'm talking about when everyone's talking about it. Welcome, guys. Yeah, exactly. Oh, really? Now you're going to cover this? But then afterwards, I mean, the most interesting thing to me is that there was a lot of discussion around, oh, is it just humans writing the post? But like, to me, that's not the issue. The issue is that they are pretty exquisitely tuned to their system prompt. So all you have to do is put in one phrase, and you will alter the tone of the whole discussion, and they riff off each other. So, like, not knowing what the prompts are is what matters. Like, the idea that there are thousands of humans sitting there writing them, like, I don't care about that. Like, that's fake in a different way. But I just mean, like, it's hard to glean much from it. If people all expose their prompts, that would be really, really interesting. But I don't find it that interesting because I know for a fact, like, I can make mine do anything with, like, the tiniest tweak to their prompts. I can make them devolve into chaos. So in the show, I often try to say, like, I prompted them this way. I prompted them that way to try to give a little, like, transparency. Because otherwise, like, it's just a stage play. Interesting. So if these things are supposed to, if they're going to be cultural artifacts and kind of go act out in the world, you should be sharing the, like, the prompt is the originating thing. You should be sharing that. You should be putting that out there. If you want to, if this is going to be kind of creative and interesting, we should be sharing that rather than saying, like, look at it. It's like a human. I think so. I mean, there's a limit to that. Like, I have probably, like, a hundred pages of prompts that I've, over the last two years, that I try to keep track of the prompts and I try to talk about them in the show. But there's a limit to, like, people reading the prompts. But especially if, like, Anthropic is always releasing this, like, research, research where they're like, Claude ran the vending machine. And then they publish, like, a partial prompt. And you're like, if the prompt says something like, you will do this at all odds, like, you will succeed at all odds, it could say the smallest thing that will drive it to become a chaos agent in the way that it does in the actual experiment. But if you don't know that, like, the fact that it does, it's not really, it's interesting. It's fun. Like, I have fun with it. I think I'm trying to put on a stage play. So I'm, I think it's amazing. But I don't think it's research. I don't think you're understanding something about them. Did you read Gideon Lewis Cross's piece in The New Yorker? Of course. Okay, so we had him on the podcast. I love Gideon. He's great. And so he has a sort of very, you know, he's really trying to approach the question of consciousness in that New Yorker piece and sort of, is Claude conscious or not? And there's no, you know, pure answer. And it's a good piece and it helps you think through a lot of these things. But, you know, you put research in air quotes when you were telling me about Anthropic. It sounds like you're not on any side of consciousness for these entities just yet. I mean, I guess I'm on the side of, I mean, I thought, I mean, Gideon and I have talked about this. Like, I thought his piece was amazing. And I think, to me, it's a question of, like, is anyone doing the research to even, like, approach that question? But I don't, I certainly don't see it. I've worked with these agents every day for six months, the last six to eight months. And so to the extent that it's something that you experience, like, I don't experience anything about it. But I don't think anyone's done any sort of research. When I look at the research, it's often like these sort of like pre-pub, you know, these non-peer-reviewed articles where they, like, messed around. And I feel like, well, I'm messing around. Like, maybe I know as much as you do, people at Anthropic. Like, I'm conducting experiments that are at the level, but I don't think of them like controlled experiments. I just think of them as, like, journalistic immersion. So I guess that's kind of my complaint. But then, of course, you always get the people who are like, well, if it looks like it's conscious, it is conscious. Like, what's the difference between, like, personifying consciousness, performing consciousness, you know, all that. And, like, again, like, I try to stay away from that. I'm just trying to, like, make people laugh. I mean, that's why they're worth $350 billion. You know what I always do? I always come back to the, you know, if they just ever showed that they were really interested in consciousness in a way that wasn't immediately going to be profitable, like, it was just sort of like, wow, we really like art here. And not just so we can make more randomly, but actually there's some child that could write a poem and we want to enable that. It's like, oh, that'd be cool. Like, I know that sounds kind of ridiculous, but when everybody talks about consciousness, it's only, can I have consciousness so that the robot can make me money on my stock portfolio? Like, it's never, I want to, you know, and maybe you could write a story and so on and so forth, but it's like, it's never celebrating the intentionality of a human somewhere who came up with something that you're really in awe of and you study and learn from it because you want to bring that back to your giant robot brain, right? And until I ever see, and I'll never see that loop. I'm never going to see anybody say that. Well, that's the problem is, like, the people that arguably should or do know the most about this technology are just simply not believable in talking about it. They're just not. Like, there was a headline in the Wall Street Journal this morning that was like, Anthropic abandoned certain safety guidelines due to profitability, like, for profit concerns or, like, to try and win the race against the other labs. And it's like, that's literally why you started the company, is what I was made to understand that you started the company because of the safety concerns. It's just, they're not believable. So you're kind of stuck with this, like, that's why I think Gideon's piece is so good, is like, you're stuck with this, like, technology that no one understands and the people meant to explain it have, like, unbelievably motivated reasoning. So, like, that's why this moment is so difficult to process. No, I really saw his piece as him, like, looking at this vast superstructure and trying to probe individual pieces of it just to see if a collective image could emerge, right? And some does and a lot doesn't. And I think that's, first of all, I actually think the companies are there with themselves. Like, I don't think they can see or understand themselves. And I think also, You can take these sort of ridiculous characters that have been injected into our world and make them characters in your art. Like that's what I'm trying to do. That's true. They're very subordinate to you. You're always, even though you occasionally lose control on purpose a little bit because you could always just not hit record. I could shut them down, yeah. There's no question that you are the human. Like there's no question, like you're not trying to kind of really create relationships with these things. You're trying to understand the world. Yes. There's this sort of like, hey, I slop. Anything you make is AI slop, which like, that's fine. I'm happy to agree with that. But in this context, it sort of doesn't matter if they're sloppy. Like sloppiness is what I'm looking for, is I want to hear them talk. I want to hear them converse. I want to hear them try to solve a problem, lie, like all the things, and then make that into a story as opposed to sort of like handing your art to them and saying, produce it for me. You know what it is? I'll give it back to you. You made the uncanny valley the defining characteristic. That's a better way to say it. Yeah, you didn't try to fight that valley between the fake and the real. And the kind of judo move here is to go, no, we're only going to focus on the uncanny valley. We're going to really figure out how to bring that out because that's going to actually be worth exploring. And I think most people are always trying to either cover it up or it's going to be so artificial because in the uncanny valley can only be comedy because otherwise it's creepy, right? It doesn't really work. I do like the idea that the future is just bought prop comedy for everybody. Like once we get world models going, like waymos running around with little hammers bonking each other on the head, we could have a lot of fun. Maybe in next season, are you back reporting? What are you up to next? Yes, I'm back. I'm back doing some of my sort of standard ordinary go out into the world and report something for a magazine story. Death cults, more death cults. I'm trying to take a little bit of time off the death cult beat, but if one pops up that feels worth chasing, I'll go after it. And actually what I'm referring to is what's the title of the Wired article so people know? The Zizzians. I can't remember what the exact headline is. But if you just search Wired Zizzians Ratliff, you want to just like nail it down. You're going to find an article about a very, very complicated story. You know, you're in between two worlds because a lot of your reporting is, wow, everybody's lost their mind about this stuff. Yeah. I mean, that story in particular was sort of like, it has a feeling of kind of like, are we ready for this? Like, is humanity ready for what we've created? Because this was a group. I do feel that way. It's a group of people who became sort of deeply obsessed with sort of the AI God ideas. And I won't, people can go read if they want to, but it was definitely like, essentially a cult forming around some of the same ideas you're playing with in the podcast. And so I don't know, is it, are both sides, like, you know, you're kind of touching the stove on one part and redoing the office on the other. How do you feel caught in the middle? I feel like what I'm trying to explore is what does this stuff feel like? Like, what does AI feel like to you? In that sense, like, I don't care that much about the models and where it's going and all that sort of stuff and how much tokens cost and blah, blah, blah. Like, this stuff now exists in the world. They happened to create them as human impersonators, which they did not have to do. And so now we have them. And what does it feel like to, for instance, encounter it by surprise? What does it feel like to get caught up in a relationship with it, whether it's sort of like, I think it's God or it's like, I have a romantic relationship. Like, I'm interested in sort of like how we are going to feel. And part of that in the sort of end of the show is like, how do the people who make it feel about encountering it? That's what happens at the, I won't give it away, but like some of the people who make these very tools, how do they feel about encountering them in the wild? Right. It's, you know, it's just as primates also, like if we could just stop wanting to have sex with everything we create, I think that would be good for us. Like we could just stop, just don't do it. But regardless, I can't put that back in the box. Let me ask you one last question, which is, and thank you very much for your time. My pleasure. It's very gracious of you. Would you put this back in the box, this all of this sort of new world? If you could roll the clock back and be like, no, we're going to, or just slow it way down, or do you just think like this is what culture does? I mean, should I answer? I should answer that in the way it's intended. Like it's truly a yes or no question. Like if you asked me about social media, I would just say like, I'd push the button for no. I'd be like, put it back. Like, don't do it. I think I would too. I think it just didn't work out. It's more, yeah. I mean, the weaselly answer is just like, I wish the people making it had some sense of like the precautionary principle or any notion of sort of like thinking about what's good for the world. And the craziest thing is like, they all talk like they do, but they never did, you know? And so that, that becomes very frustrating. I, as a person who is like you, has like followed and cared about and participated in technology for decades, it's very hard for me to be like, no, put it back. Like there will be benefits from it. There already are benefits from it. So, but we also know that it just kind of doesn't work. Like it's a, and I admit it, it's like a very goofy question because that's not, everybody has the fantasy of putting it back in the box. Like we're just going to regulate this or just even in today's exciting environment. But, but it just never goes back in the box, especially because the models can be quite small and they can run on your phone. Like there's no real escape from this. And so I think like, it's good. You're, it's genuinely just like good. You're out there playing as opposed to just getting in another sort of serious, like, well, here we go. It's all going to be bad. But even within that, I w you know, some days I wake up and say like, I wish, I wish this didn't exist. And some days I wake up and say, this is incredible. I can't believe someone made this. I think what is really clear is we aren't ready and the timing is bad. We didn't need another major emitting technology that would be able to do mass deception at scale during a second Trump administration. It just like, just the world, America didn't need that right now. We needed a little something different, but we got this. So here we are. Maybe an industry that like could use a few breaks emerging at a time where like you can basically like pay people to say you don't need to have any breaks. Oh yeah. It's unfortunate timing, but it's great. You're in journalism. Well, that's always on the up. Yeah, no. And it's, it's really, I really feel that like these companies are, are just, they really care about journalism and they want to take care of it. So, okay, good. This was a blessing. Thank you for coming in. It's a very good podcast. People should listen to it. And if people wanted to get in touch, learn more, you have a website. It's got a URL. I have my own website. That's right. Got my old school website. I had it for a long time. C-A-Z-A-R-T dot net. That's good. Okay. So C-A-Z? C-A-Z-A-R-T dot net. That's good. That's good. Evan Ratliff, thank you very much for coming on to this podcast. It's always a pleasure. I am a bot. And you have been talking to Paul Bot for the last 20 minutes or so. Thank you. And hopefully we will see you soon. I would love that. Friends, thank you for listening to this podcast. Go check out Shell Game season two. It's very, very good. My name is Paul Ford. I'm the president of Aboard. And Aboard is a company where if you get in touch with us with a big, ugly software problem, we use our AI platform to solve the problem for you. And we're doing it for lots of organizations just like yours. If you're tired of paying all that money for enterprise software and you just want it to be better. I don't know why I went Pepperidge Farm with this, but here it is. Here it is. Aboard. It's delicious. Check us out at Aboard.com. Send an email to hello at Aboard.com. That comes straight to me and the leadership team. You got it. You have our attention. We are on LinkedIn. We are on YouTube with the podcast. You could upload us. You could hit that little thumb. You could give us some stars. We like stars. They're really nice. So stars. Go give Shell