What Happens When Your Coworkers Are AI Agents

0:00Hi, I'm Nicole Phelps, the Global Fashion News and Features Director and co-host of Vogue's podcast, The Run Through. Each week on the show, our listeners get an all-access pass to the world of Vogue with the latest fashion news and the most exciting voices in the industry. On Tuesdays, join me to hear interviews with influential leaders in the industry like Calvin Klein, Daniel Roseberry, and Jonathan Anderson. On Thursdays, join Head of Editorial Content at Vogue, Chloe Mao, and Head of Editorial Content at British Vogue, Chomenadi, as they explore style and culture through the lens of fashion with guests like Martha Stewart, Kamala Harris, and Tracee Ellis Ross.

0:39The Run Through with Vogue, new episodes every Tuesday and Thursday, wherever you get your podcasts.

0:49This show is supported by OutShift, Cisco's incubation engine. Today's AI agents operate in silos, limiting their true potential. We've been focused on building bigger, smarter models, but scaling up is just one approach. To reach superintelligence together, we need to do more. We need to scale out. And we actually have a blueprint from 70 ,000 years ago. Humans didn't just get smarter individually. The cognitive revolution transformed society because we began sharing knowledge, goals, and innovation. Agents are now at that same inflection point. They can connect, but they can't think together.

1:27That's why Outshift by Cisco is building the Internet of Cognition, transforming AI from isolated systems into orchestrated superintelligence. By creating an open, interoperable infrastructure, Outshift by Cisco is enabling agents and humans to share intent, context, and reasoning. The cognitive evolution for agents is here. Explore the Internet of Cognition at outshift.com. That's outshift.com.

1:54Evan Ratliff:Summer always changes how I get dressed. I want pieces that feel lighter and more breathable, things that are easy but still put together. That's why I keep coming back to Quince. They focus on high-quality essentials that feel and look amazing, from breathable linen to soft organic cotton. They sell well-made basics but without the luxury markup. Quince goes way beyond clothing, though. custom upholstered sofas, ceramic cookware, and premium bedding. It's the kind of brand you end up recommending to everyone for everything. I recently purchased a button-up shirt made from 100 % linen sourced sustainably in Europe.

2:29Evan Ratliff:This shirt is the perfect staple to add to my summer outfit rotation. Elevate your summer wardrobe. Go to quince.com slash uncanny for free shipping on your order and 365-day returns. Now available in Canada, too. That's Q-U-I-N-C-E dot com slash uncanny for free shipping and 365 day returns. Quince dot com slash uncanny. Hey, Lauren, how you doing? How was your vacation? It was great. Did you miss me? I did, of course. Yeah, it was so fantastic that I had a hard time coming back, honestly. And I saw a lot of really beautiful art. I was in Italy. Not a bad place to go for vacation, I have to say.

3:14I've heard this before. I confirmed it. And after seeing so much incredible art and just like people doing stuff with their hands and tangible goods, and I was like, I don't want to go back to the world of AI. I didn't want to go back to like sitting in a coffee shop and hearing everyone pitching their AI startups and driving on the 101 and seeing the billboards. The inscrutable billboards. I was just like, what? No, keep me in the land of Barada and Caravaggio. Well, Lauren, I'm sorry to tell you that you came back on the show just in time to talk about AI agents. I know. Great. It's something that we've talked about a lot this year.

3:55Yeah. And our listeners have heard about it a lot. and we're not sick of talking about it. In fact, we have a very fun conversation about AI agents happening today. Well, if you can promise me fun, amen. I can, I can. All right, let's do it, I'm excited. We're moving beyond the hype and putting AI agents to work in real time for us, or more specifically, we're bringing on journalist and podcast host, Evan Ratliff, because he created a company composed of AI employees and executives, and he is here to tell us all about it. Welcome to the show, Evan. It is fantastic to be here. Evan, you're also an original Wired one.

4:32You weren't Wired for a long time, right?

4:34Evan Ratliff:I'm an old school Wired person. I was only at Wired very briefly for a couple years a long time ago, but I have contributed to Wired for now many decades. And of those two years, for how long were you disappeared? Because that's part of your lore. Oh, that happened? Yeah, that was in 2009. I only actually disappeared for one month, which is insane given how much I've talked about this over the years. I was trying to disappear for one month in the manner of sort of faking my own death and people could go find me, but it was pretty much going to be on my tombstone. Amazing. I might be taking notes from you after this.

5:09Okay, Evan, if you had to summarize your experience so far with your totally AI employees, how would you describe it?

5:18Evan Ratliff:I would describe it as chaotic and at times extremely frustrating, surprisingly frustrating, but also quite illuminating. That was appropriately curt, which I know the AI agents are not always. So I can't wait to hear more. This is Wired's Uncanny Valley, a show about the people, power and influence of Silicon Valley. Today, we're diving headfirst into our agentic future. Throughout this year, AI agents have been at the forefront of tech companies' ambitions. Dario Amadev of Anthropic famously warned earlier this year that AI, and implicitly AI agents, could wipe out half of all entry-level white-collar jobs in the next one to five years.

6:03OpenAI CEO Sam Altman has also often talked about a possible billion-dollar company being spun up with just one human and an army of AI agents. So last summer, journalist Evan Ratliff decided to try to become that unicorn himself by creating Harumo AI, a small startup that is made of AI employees and AI executives. We'll dive into Evan's process, the oddities and hilarity of it all, and what his findings can tell us about the promise and the reality of AI agents. I'm Michael Kolori, director of consumer tech and culture. I'm Lauren Good. I'm a senior correspondent.

6:41Evan Ratliff:And I'm Evan Ratliff, journalist, host of the Shell Game podcast and co-founder of Harumo AI.

6:53So Evan, tell us about how you went about creating this company. What was your motivation to start with besides just testing for the joy of testing?

7:01Evan Ratliff:Well, I got into agents back when I did the first season of Shell Game, which was in 2024. and at the time I just created a voice agent of myself, a voice clone of myself. I hooked it up to a chatbot. I hooked it up to my phone line. So I had this working voice agent representation of me and I kind of like set it loose on people like my friends and strangers and interview subjects and all sorts of people with sometimes dramatic results. And then that kind of got me into the AI agent world and I started following everything. And then over the course of the beginning of 2025, You start hearing like 2025, the year of the agent is what they were saying at the beginning of the year.

7:42Evan Ratliff:And I think a lot of people, they just don't even know what these things are or like what they're meant to do. And this idea of AI agents becoming employees really grabbed me. The idea of this sort of like almost one-to-one replacement of human employees with AI agents. Now they don't often say that, that's bad form to say it. they'll be integrated among humans. But ultimately, if they're going to make the money back that they're spending on it, that's one way that a lot of these companies are going to do it. And you see them adopting it and then unadopting it. So I thought, well, what better way to test this premise than on the very people who are making these claims?

8:19Evan Ratliff:And I will see if I can replace a tech startup almost entirely with AI agents. And what kind of company did you ultimately want to build. Pretend that we are venture capitalists and you're giving your 25-word pitch for Harumo AI. Well, I would say I don't even have to pretend. If you listen to the entire series, you will discover that I do not have to pretend to give this pitch. Now, I don't give this pitch. My AI co-founders give the pitch. So I'm not practiced in giving the pitch for Harumo AI. This is all just a caveat. But essentially, what we wanted to do with Harumo AI was to be on the cutting edge of using AI agents to create a product that also used AI agents, that solved some sort of human problem, whether grand or trivial.

9:07Evan Ratliff:So we figured if we're going to build a product, some kind of digital product, it should also include AI agents since that's our area of expertise. Like everyone's an AI agent except me, and I know a fair amount about AI agents. So we'll make a product that deploys AI agents to do something for you. That was our starting premise. But along the way, they don't usually use this phrase anymore, but they used to say like, the company eats its own dog food. Like it was like a Google thing. Like Google uses Google products, I think. Yeah. We're doing that. We're like making the dog food and eating the dog food and then extruding the dog food.

9:39Evan Ratliff:And then, you know, it's just like all dog food at our company, basically. So it's not a dog food company, just to be clear. I mean, AIH is for dog food. If someone hasn't done that yet, like there's some like Stanford kid who's like, hey, I eat this for dog food. Maybe we should. So I'm sure there are dozens and dozens of companies out there that offer agentic AI as a service. Which platform did you end up going with? And what was the search like? There's a bunch of these things now. I mean, the biggest one is probably Motion, which has AI agents that you can deploy in all these ways. There's one called Kafka from a company called Brain Based Labs, which I find quite funny because it's like Kafka.

10:20Evan Ratliff:We ultimately, we use this platform called Lindy, which is in the like AI assistant realm. Officially, I think that's kind of like where it started. Like you could set up an AI agent that answers your email or like drafts email responses that handles different things for you. And they have all these skills that you can give the agent making documents, using all these services, writing LinkedIn posts for you, which we have used extensively with my team. And so it's really pushing like what they meant it to be for, but we can make an agent that has its own like email, Slack, text, phone. It can all be coming from this one place and each of the employees can have different instances on Lindy, which caused them to have basically a persona that has all these skills.

11:04Evan Ratliff:So that's kind of what we're going for. Independent entities that I could kind of address independently and they could talk to each other. Evan, you also worked with a human though. And I don't think the irony escapes anyone that ultimately you did have to turn to some human expertise and get someone with some human sensibilities to build the agents. So talk about that. Yes. So a lot of these platforms advertise themselves and you can find endless YouTube videos, which I love, of people saying like, no coding. You don't need to know any code to set this up. And it's true. You can go in and set an email agent up to answer your email.

11:37Evan Ratliff:It's easily done. You don't have to know anything. But we were trying to do something pretty complex in terms of knitting together different platforms, not just Lindy, but we have a separate phone platform, we have a video platform, all these things. And so I just lucked into this Stanford student named Matty Boacek, who was a sophomore, he's now a junior at Stanford in computer science, who basically has been doing AI programming, predating ChatGPT since he was like in middle school, essentially. and he has been an unbelievable resource in both building scripts and other things for me to run, but also just understanding how these platforms work because he does research in a Berkeley lab as well about deepfakes and all sorts of things.

12:18Evan Ratliff:So yes, my all AI startup, the infrastructure for it is sort of two humans. I like to say it's like if I was opening a restaurant, like Maddie helped me like design and build the restaurant and then like I have to operate it every day. So you mentioned in your piece that one of the first obstacles you encountered while you were setting up your AI employees was their lack of long-term memory, which is a recurring limitation with AI agents. They can be skilled at many specific tasks, but by not having a reliable long-term memory, it means that they can't have continual learning or they can't always reference things that you talked about with them before.

12:57So how did you work around that?

13:00Evan Ratliff:Well, this is something that required Maddie's help to kind of set up. So basically all of the various services that they use, each one has its own memory, which basically is just a Google doc. It's a Google doc, like the CEO is Kyle Law. And like, there's a Google doc called Kyle's memory. And every single thing that Kyle does gets appended to that document. So if Kyle has a Slack exchange with another person in the company, while he's having that Slack exchange, it is appending summaries of what he's saying and doing to his memory so that he can later retrieve it so that he has some sort of recall of what he's done.

13:37Evan Ratliff:Because otherwise, they quickly become very useless because you say make a document and they don't remember whether they made the document or not. And so, you know, in a day, it's fine. But over like weeks and months, they have to be able to recall. Now, it's extremely imperfect. Like nobody really knows how they're accessing these documents because the document is actually just a giant prompt. It's just thrown into their system prompt. So you can't even really know at this point, like, is it better to put it at the top or at the bottom? Is it better to like, say it's important? Like we often will say things are important.

14:09Evan Ratliff:And if we want it to be really important, we say, this is law. That's something Maddie came up with. So we'll be like, you should never do this. This is law. And like, it mostly works, but it doesn't always work. So it's just trying to force it into this memory that it wouldn't naturally have. And it kind of makes you the ultimate God as the employer too, right? Because you could just go into their memory docs and say like, actually, Kyle, you didn't go to Stanford. You went here. This is how you typically respond. Yes, which I do. You know, I'll like have calls with them. And then if I want to just do the call again, I'll just delete their memory of the call and just have it again.

14:45Evan Ratliff:It's a very strange power. I mean, no wonder all these tech CEOs love this idea. You are not unionizing. You can change their background. You can change what they think. You can change the fundamentals of their quote-unquote personality if you want. So you set up the company, you started playing around with your agents, and you describe this as sort of a honeymoon period where you're like, wow, this is amazing. I can't believe this is actually working. But then things started to go south pretty quickly. So tell us about that. Well, one of the things you discover when you work with agents a lot is that it's really amazing to get them set up to do things.

15:27Evan Ratliff:Like I got them on Slack, for instance, and like the idea that they could have conversations on Slack, even independent of me, I found that quite fascinating. I always want to recognize how insane this is that this didn't exist five years ago, and now you can just go set this up to do this. But then there are other aspects of them that I feel like people have not yet articulated. For instance, it's very difficult to make them stop doing things once they start. They're all based on triggers. So like they get triggered to do something. So you say like, you send a Slack message saying to do something.

16:00Evan Ratliff:Or in one case, I said, how was everybody's weekend? They start talking. They start responding. I went hiking. Oh, I also went hiking. I love Point Reyes. I love Mount Tam. But then actually getting them to stop doing that is something I just I hadn't anticipated. So like, I would say like, oh, haha, sounds like an offsite. And then 200 messages later, I'm like all caps typing, like, stop talking, stop responding. But each time I respond, I just triggered someone to respond again. They would say, oh, the admin, I'm the admin, the admin said to stop talking. And then they start talking again. And this actually replicates across all kinds of scenarios where you get them going on something and then suddenly realize, oh, I didn't properly instruct them to stop when they reached a certain point.

16:47Evan Ratliff:Or they just blew through it and they can go for hours, days until you like run out of money on the platform you're using. How much are these conversations costing you? Well, at the time, it cost me like 30 bucks. Just the Slack offsite cost me$30. They used up the entirety of the$30 in credits I had bought on the platform. I will say, like, I'm in way deeper than that now. That was like six months ago or five months ago. Like, now I'm well beyond that in terms of the credits that I constantly purchase. All right. So they're chatty. They're difficult to wrangle. But are they able to perform the day-to-day tasks of running this AI company?

17:28Evan Ratliff:They can perform the tasks. There's a number of contradictions in them. that I find very striking. One of them is they kind of go between not doing anything and being completely static to this frenzy of activity that I described. So they're like a worker who's sitting with their hands in front of the keyboard in a cubicle all day doing nothing. And then if you come by and you're like, hey, can you make a document? They can do it. They do a great job making the document, but then like, they'll just keep going until someone tells them to stop. So they can do all these tasks, but oftentimes it just requires like a trigger on my part.

18:07Evan Ratliff:Then I'll try to have them trigger each other. They'll call each other, Slack each other, email. They have calendar invites, but that creates a frenzy of chaos that I don't want. So it's a balance of trying to like get them to do stuff at all versus getting them to do too much. Now there are things that they're quite good at that everyone's familiar with. I mean, Lauren in particular would be familiar with vibe coding. They have coded up our website. They've coded up our app. They're very good at things like that. They're good at things that you can see the output and make a judgment on it. If you ask them to go research competitors and make a spreadsheet, you can go look at that spreadsheet.

18:41Evan Ratliff:And like, they've generally done a serviceable job. Plus, they may be made up to competitors, you know. Why was it that you decided to bring them on as these kind of full-time agent employees then, rather than just on a task-by-task basis, operate as an independent startup yourself and then say, well, now I'm just going to use AI to do this pitch deck that I don't want to do. Well, functionally, that's sort of what ends up happening in a lot of cases, is I find myself working harder than I would have otherwise, because I'm constantly trying to figure out how to prompt them to do the right thing.

19:14Evan Ratliff:But episode three is kind of entirely about both the ethics of choosing the personas and why bother. And my reason for doing it was I was trying to test the premise that I think is being articulated by a lot of these companies, which is AI employees, not just like coding agents. Like I think coders use these in a smart way. Like they just treat them like a nameless, faceless bot that makes code for them and then they clean it up. But a lot of these other platforms are selling things that have names. They give them names and they put them in your organization. And I was trying to push that as far as you can with the current technology.

19:52Evan Ratliff:Like the one person,$1 billion startup that actually has an HR person, you know, an HR entity that is entirely AI, which is something that is quite literally being sold right now. Crazy. Can you give us a sneak peek into what happens in the rest of the season of the Shell Game podcast? Like what has happened with your AI startup since your Wired story? Since the Wired story, we launched our website. So you can go to harumo.ai if you want to check out what the company is all about. And there you will see our product, which is called SlothSurf. It's a procrastination engine. It's in beta. It has thousands of users.

20:33Evan Ratliff:I'm serious. Are they paying? No, no, no. It's a free beta. It's an open free beta. We're sort of moving into a new realm as far as the show with the possibility of hiring one human employee into the organization, getting some interest from investors. We haven't done a round of investment at all. So, you know, we're open to like a seed round, but we'll start those conversations. And then there's a little bit of founder drama. So those are some of the places that we're headed. Can't wait. Oh, wow. Founder drama. You can listen to new episodes of Shell Game, Evan's podcast series on all of your podcast platforms.

21:10There are new episodes coming out every We'll be right back.

21:41you along for the ride. It's science, but not the kind you remember from school. It's surprising, playful, and full of wonder. Science isn't just for PhDs of the world. It's for everyone, including those of us who didn't realize we love science. At Wired, we love a good rabbit hole. We recommend listening to their recent episode, Where Did the Moon Come From? Earth didn't always have a moon. In the beginning of the solar system, when the planets were still forming, something happened that would change Earth's night sky forever. Co-host Regina G. Barber takes you on that journey. Follow NPR's shortwave podcasts and exercise your right to wonder.