1. Podcasts
  2. Uncanny Valley | WIRED
  3. Unpacking AI Agents
Uncanny Valley | WIRED

12 Jun 2025 · 36m · 11 chapters

In short

AI agents—how “agentic” systems differ from chatbots, why major tech firms are racing to deploy them, where they’re useful now (customer service, coding, office automation), and the risks (reliability, cost, security, job displacement, and loss of human agency).

Guests

Will Knight (Wired senior writer; focuses on AI research; discusses “vibe coding” with Claude Code and Andrej Karpathy’s term) and Lauren Good (Wired senior correspondent; discusses consumer/home device implications and concerns like privacy/security and skill atrophy).

Key claims

agents use language models to trigger actions with humans more “in the loop” than in constant dialogue; companies are rebranding tools as agents due to unclear definitions; early wins are customer service and task automation; reliability/hallucinations persist; agents that buy things are higher hacker targets; automation may dehumanize work and erode human interaction.

Notable examples

Google Duplex-style reservation attempts, Claude Code editing/moving/deleting files, Southwest chatbot flight changes, Project Mariner (cocktail recipe → Instacart cart), Google Project Astro glasses, Mariner Instacart authentication bottleneck.

Written by AI. May contain mistakes.

Chapters

0:46

Introducing the Topic of AI Agents

Discussion on AI agents and their significance in technology.

“I think, Will, you brought us croissant, did you not?”
2:05

Personal Experiences with AI

Sharing personal experiences with AI agents in daily tasks.

“I did have an experience recently with, do you remember Google Duplex?”
4:26

Defining AI Agents vs. Traditional Chatbots

Exploring the differences between AI agents and traditional chatbots.

“And it was like, talk to our chatbot if you need help.”
7:22

Current Players in AI Agent Development

Overview of major companies developing AI agents and their strategies.

“And agency actually involves more than just trying to solve an individual problem because it involves other people, which makes it complicated.”
9:11

The Future of AI Agents in Customer Service

Discussing the future role of AI agents in customer service and office tasks.

“It seems like every briefing we are on these days, a company is talking about agents.”
11:24

AI Agents in Our Homes

Exploring how AI agents might integrate into our home technologies.

“Because it involves taking what language models are brilliant at, which is talking to people and then trying to do a fairly limited number of things.”
14:00

Exploring AI's Impact on Jobs

Discussing how AI will transform job roles and tasks in the future.

“And him basically saying, well, with AI, people are just going to level up.”
21:00

Ad Break: The Political Scene

Promoting the political podcast featuring discussions on current global issues.

“This week on the political scene from The New Yorker, Trump's rupture in the world order.”
21:44

Challenges and Concerns of AI Agents

Analyzing the limitations and ethical concerns surrounding AI agents.

“So the development of agents is going forward full throttle.”
28:00

Navigating the Future of AI Agents

Discussion on the evolving role and implications of AI agents in our lives.

“that happens when you dispatch an agent to complete a bunch of different tasks online through a bunch of different apps.”
Show all 11 chapters
33:00

Recommendations and Closing Thoughts

Hosts share their recommendations for listeners and close the show.

“Will and Lauren, thank you for an invigorating conversation.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On Uncanny Valley | WIRED, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Hi, this is Mike. Before we start, I want to take the chance to remind you that we want to hear from you. Do you have any questions around AI, politics, or privacy that have been on your mind, or just a topic that you wish we talked about on the show? If so, you can write to us at uncannyvalleyatwired.com. That's uncannyvalleyatwired.com. And if you listen to and enjoy our episodes, please rate the show and leave a review on your podcast app of choice. It really helps other people find us. Hey, everybody. How you doing? I'm good. I'm hanging in. How are you, Mike? I'm great. I feel good. We have Will Knight on the show this week.

0:39Yeah, I feel great about that. I do miss Katie, but we know that Katie has gone off in search of more French butter. Literally, she has gone to France. I think, Will, you brought us croissant, did you not? I did. I sure did, yes. Freshly baked. Thank you for that. You can stay on the show now.

1:04This is Wired's Uncanny Valley, a show about the people, power, and influence of Silicon Valley. Today, we're talking about AI agents and why so many tech companies have been betting big on them. In the past six months, OpenAI, Anthropic, Google, and others have released web browsing agents that are designed to complete tasks independently, with only minimal input from humans. OpenAI CEO Sam Altman has called AI agents the next giant breakthrough. By one estimate, nearly half of technology companies are already adopting or fully deploying artificial intelligence agents. This week, we'll dive into what makes these agents different from other forms of machine intelligence and whether their capabilities can live up to the hype.

1:50I'm Michael Kolori, Director of Consumer Tech and Culture here at Wired. I'm Lauren Good. I'm a senior correspondent at Wired. And I'm Will Knight, a senior writer at Wired.

2:03I want to start us off by asking if either of you have had agentic experiences in your life lately. I did have an experience recently with, do you remember Google Duplex? Sure. Okay, so that was kind of an early version of this. It came around around 2018. We wrote about it at the time at Wired. But I tried to make a dinner reservation recently at that place that you and I went to for your birthday. Vegan sushi. Already, this is the most San Francisco podcast ever. We're talking about vegan sushi and how I dispatched unknowingly an AI agent to try to make the reservation for me. How was it unknowingly?

2:37Because I just did the very human thing where I went to website and typed in, would like to make a reservation for this number of people at this time. And unbeknownst to me, it funneled it through the Google Assistant. And then it said, we are trying the restaurant on your behalf. And it kept trying, and then it would send me a notification, and it kept trying, and it would send me a notification. So it was working behind the scenes to try to make a reservation. It did not successfully complete the reservation. Does that count? Yeah, that fully counts. That's an AI agent? It was like working on my behalf.

3:08Will, what about you? So I've been playing with AI coding tools and primarily Claude Code, which is, I think, a good example of an agent because it does all these fun things like edits, not just writes code, but it edits files, moves them around, uses your terminal, deletes stuff if you're unlucky. That's the agentic thing I've been playing with. So you've been vibe coding is what you're saying. I have, yeah. It's been fascinating. So vibe coding is this term that a very well-known AI researcher, Andre Kapathi, came up with to describe basically conjuring up whole programs, you know, finished software just by prompting a model.

3:50So these models have always been able to kind of autocomplete code, but they've gotten so good in the last year at coding that they can create whole projects involving lots of files and lots of folders and everything. Actually, I started off doing it with my kid because I thought, how do you teach a kid to make sense of AI? And also, like, you know, you've always thought it's a good idea to get them to code. And I thought he loves games. So I thought let's code some games. Well, my experience is far less technical because I recently had to make a change to a plane reservation through the Southwest app.

4:24And - I'm sorry. Yeah, well, it was actually fine. My condolences. It pushed me to the chatbot. And it was like, talk to our chatbot if you need help. And I was like, oh, this is going to be infuriating. And actually, it was fine. It took like five minutes. I gave it my confirmation number, my flight information. And then it said, what change would you like to make? And I said, I would like to do this. And they said, we can do this for you with no fee. And I was like, great. Do you want to book it? Yes. OK, it's booked. Here's your new boarding pass. You need to check in. But here's the link to check in and get your new boarding pass.

4:56And I was like, wow. So is that the agent we're talking about? That's a customer service agent, and it's a bot replicating that. But this new era of agentic AI, tech companies raising millions of dollars and having valuations in the billions because they're pivoting to agents. That's not exactly what that is, though, right? Right. That's a good question for Will. So can we throw this one to you, Will? What's the difference between an AI agent, what we're talking about, and a traditional chatbot, which most people have experienced? Right. So a chatbot generates text and you can communicate with it through that narrow, very, very narrow sort of domain of text.

5:38And that's been extraordinarily powerful. The revolution we've seen is built around these language models, which can produce extraordinarily coherent and good answers to questions. that isn't quite the same as an AI system that is inherently an agent. So if you go back in the history of AI, actually, a lot of researchers have always worked on this idea that intelligence inherently involves agency. Humans are not things that just speak. They try and perform actions. You try and solve problems. And so that's been a longstanding goal. And it's kind of a weird quirk that we had these language models, which are so sort of limited and kind of old fashioned.

6:23And a lot of what we're seeing when it comes to agents out there are basically people taking language models and saying, you know, using the fact that they're very general and very flexible to take a complex command and trigger some actions. and the real leap that companies want to make is that the models themselves will decide and be able to formulate their actions themselves more like a like an intelligent entity so the human is still in the loop but the human puts in a command or a prompt and then steps back and the agent is doing a bunch of ai stuff basically while you go do other things right that's the promise that they're sort of going to make us all hyperproductive yes so ideally you would have an agent that would call up the airline for you this is you know you tell it you what you want to do and it goes and does the negotiation for you or you can give it much more open-ended you know commands or what you what you're trying to achieve and it figures out how to do it for you and the promise of these you know these models are so the language models are so general so capable that in theory they ought to be able to do that but there's some there are some really key catches when it comes to how you try and replicate human agency, because there's a lot more going on in the world than just language.

7:40And agency actually involves more than just trying to solve an individual problem because it involves other people, which makes it complicated. To say the least. Yes. So speaking of the models, it seems as though all of Silicon Valley is invested in developing their own AI agents, but who are the main players right now? I mean, the main players are the obvious ones open AI, Anthropic, Google. I think Amazon is a good sort of dark horse in this race. They hired some exiles from open AI who made a agentic startup and they have a lab that they're trying to build more, build a lot of agents. One of the key things for these companies is where you get the data from.

8:21So they're all desperately trying to figure out how you get data to train these agents. Instead of having just text, you want examples of things that people do. And Amazon actually has quite a lot of good examples of that, of people doing stuff on its sites. So they're a good player. Well, then in the coding world, too, there are companies like Cursor, right? Tell us about that whole world since you've been vibe coding with your son. Yeah. And so there are a bunch of startups who've really ridden on the backs of the other AI companies, but really focused on coding. And I think just because that's moved so quickly, they've been able to kind of carve out these niches.

8:56And, you know, OpenAI is widely rumored to be looking to acquire one of these companies, Windsurf, for about$3 billion, which gives you an indication of how they've been able to, you know, make this really great business for themselves. It seems like every briefing we are on these days, a company is talking about agents. Yeah. And it ranges, too. It's every company from NVIDIA to Microsoft to well-funded startups. Even just earlier this week, a company called Gleam, I'm pretty sure you guys have heard of it. It's a company that has built a search engine for enterprise apps. They just raised$150 million in a Series F round of funding.

9:37They're now valued at$7.2 billion. And their whole thing now is agents. It's all about agents. I think the way to think about it is who is really building models that are trying to learn the kind of general agency that you see in a language model. And that is OpenAI, these foundation model companies. If you can go back to a previous era in AI when we had some of these game playing AI systems like AlphaZero that used reinforcement learning, this process of experimenting. But it was fundamentally agentic. It only operates within this very limited world of a board game, but it's taking actions. It's figuring out how to take those.

10:19And what we're seeing now with these AI companies kind of mirrors that or is going back to that in a way because they're trying to get them to reason to a particular answer, to a particular end goal. If we think about what role AI agents could play in the long term, a lot of the companies that we're talking about are starting with customer service, right? They're starting with, okay, you have a simple question. You need something simple done. I can go figure that out for you. I can go accomplish that for you. And I think that's where most of the early sort of strides are being made. Gartner, the tech market researcher, has put out an estimate that AI agents will resolve 80 % of common customer service queries by the year 2029.

11:06So that's four years from now, AI agents will be doing all of our customer service stuff for us. And based on some of these examples that we've been talking about, it does seem like the automation of tasks currently performed by employees who are human is a common thread here. Yeah, I think the customer service is an obvious place to start, right? Because it involves taking what language models are brilliant at, which is talking to people and then trying to do a fairly limited number of things. But a lot of these companies, like the big AI players, are looking at going from coding to how you automate office work.

11:44So there's a ton of repetitive tasks in office work. And in theory, you could be able to use these models as the kind of glue to figure out what someone is trying to do and then replicate that. I mean, there's already a huge amount of worry in the world of coding that this is going to eliminate lots of jobs. And then I think that there's a good chance it's going to come for much more kind of routine white collar work as well. Well, it's been fun doing this podcast with you all. So we've mostly been focusing on how AI agents go out and do things like surf the web or call a restaurant or fulfill some sort of task in a Microsoft Office product for you.

12:28But I'm curious about what the world of agentic AI looks like in our homes on our physical devices. This has been at the top of a lot of people's minds recently because OpenAI just announced that it made a deal with Johnny Ive, the former and longtime head of design at Apple, to acquire a new startup that he launched to build a new class of AI devices for the home. How do we feel about this development? It's hard to say because we don't know what form it's going to take exactly yet, but this is, I think, representative of a new generation of startups and hardware startups that are going to exist to serve these AIs.

13:10Like, they're built from the ground up with the future of artificial intelligence and ambient computing in mind. And so we don't know exactly what form factor it's going to take yet, whether it's going to be on your desktop or it's going to be on your body or maybe both. but clearly the idea is that well one I do appreciate how I believe it was Johnny Ive who said like some version of you know we can all agree we're a little too attached to our iPhones so like the thing that they've invented is now like we're gonna we have to solve for the thing that we invented but the idea is like can AI actually do enough so that it takes us away from having to be heads down or faces and screens all of the time actually performing the actions can technology do some of that for us and then free us up for other things?

13:55And that's always, in Silicon Valley, been one of the promises of AI. I remember talking to Astro Teller from Google X years ago and asking him, well, what does this mean for jobs? And him basically saying, well, with AI, people are just going to level up. And that's always been a really sunny Silicon Valley optimistic view of AI. And I think that there will be a new category of jobs and also tasks created that we haven't maybe even thought of yet, but we still just have no idea how it's actually going to work day to day into our lives. Right. Because, you know, I feel like people have been trying this for a long time.

14:28You know, whenever we saw the demo for the next version of Siri or the next version of Alexa or the next version of Google Assistant, there was always a demo where they would ask the speaker, the voice controlled smart speaker to like order a pizza. Yes, it was the pizza and Uber. Those were always the examples. These are the deals that the device makers made with the businesses and they like coded that in. So when you ask Alexa, can you order a pizza? It wasn't like calling Domino's and Alexa's voice ordering your pizza for you. It was just doing it like on the web in an app through an API basically, right?

15:05Right. So I think when you talk about like what the next generation of that is, it's you can ask it to do anything. And what is that thing you're asking? It's some device that sits next to your computer and your phone. Why does it have to be a separate device? Why can't it just be your computer and your phone is the big question that I have. Do we need something that's going to probably cost several hundred dollars that's probably going to be very beautiful that does a lot of what the other two devices that we already own do? I don't really know. So everybody's excited about it, though, because it's Johnny Ive and Sam Altman.

15:36Right. I mean, it just becomes the delivery mechanism for whatever this new era of computing is. I do think, too, the examples we're talking about are very much like a WALL-E kind of future. Like, how lazy can we possibly be while we dispatch AI and machine learning to do the things for us? Whereas there's also going to be space for how much more can I learn? What kind of knowledge can I obtain? What can I do differently? because AI is helping me. There's been this promise sort of on the consumer end for a long time of like, this is gonna change your day-to-day life by sort of tapping into these small inconveniences that we have now.

16:16But I think the real gains stand to be one, and probably, and I don't mean to sound too optimistic about this because I'm not entirely, but like, can we sort of close gaps of knowledge because AI is actually going to help us do that in some way. Yeah, I think when it comes to the open AIs and Johnny Ive thing, the models are getting increasingly really good at interpreting the visual world and interacting through speech. And so there are sort of these, I think there's this idea in the ether that you can have completely new, a new interface. You reimagine the interface entirely, right? And you don't have this graphical user interface and God knows what they're going to come up with.

16:56But you saw like Google has demoed this with this Project Astro where you wear these glasses and you can talk about what's around you. I've tried that. It's kind of impressive. And yeah, I mean, maybe there's new ways you can order pizzas with it. But also, I can't help thinking about the fact that they need the data for what people are doing increasingly to train these models as well. And so I feel like the hardware makers have so much interesting data. They don't really tend to tap into it because of privacy issues, but maybe OpenAIR will find a way. And then on the issue of how you train, you know, the knowledge gap, I think that's a really, really important point.

17:38because automation often causes people skills to kind of atrophy. You see it in autopilot and aircraft. And then you're definitely seeing coders saying that they're losing skills because they rely too much on that. And so what does that mean for the workforce? Right. If all of these AI agents free us up to do other things, what exactly are they going to free us up to do? How are we going to fill that time? And in our, you know, ultra-capitalist, some people may call it end-stage capitalist society, the answer is probably, well, you can just be more productive. You can just do more, more, more.

18:11And I think it's going to be up to us, the humans, to push back against that and say, here's how we actually want to fill our time. There are also going to be these little bottlenecks for a while that I think are going to make these more complicated than anything else. When I was at Google I.O. a few weeks ago, I got a demo, Will, of their Project Mariner. Have you seen this yet? Yeah, I've tried that. You have tried it. I don't know if this was the demo they gave you, but the experience I saw was scanned a cocktail recipe, and then it automatically just grabbed the ingredients from that cocktail and put it in an Instacart shopping cart.

18:45But then it hit a bottleneck because there was no authentication for the Instacart account. And so it was like, oh, hold on, we have to log in. And then some demoer was trying to log into Instacart. But look, it's so seamless. And I'm like, no, I literally could have just added that to Instacart. Yeah. Their examples always seem to involve buying something for some reason. So why do we think that there's such a laser focus on AI agents at the moment in Silicon Valley? Is it the Valley's tendency to always just need a new goal to race towards? Is it indicative of something larger? Is it what VCs are most interested in funding?

19:20What's the reason here? From my perspective, one of the main reasons is that this, you know, going back to the history of AI, this is sort of really the next step you would take in building something that's getting closer to human-like AI or AGI, whatever you want to call it. And so a lot of researchers are very interested in doing that. And I think there is also, you know, a ton of potential going, you know, if you're not just chatting to a chatbot is useful if you want to do lots of things, answer lots of questions, but the idea that you could automate a lot of work or mundane chores is potentially valuable.

20:01Yeah, I think that if you're on the startup side of things, I mean, this is where the money is flowing right now. And we've been in this era for a while where interest rates are really high. For some startups, it hasn't been easy to raise capital. There's been a bit of a pause on the IPO market that has cooled a bit, although now there's signs that it's picking up again. And I think the threat of tariffs is still looming over people's heads in Silicon Valley. And so if you've been iterating on your software and suddenly you can say to your would-be investors, but wait, we're pivoting to agents.

20:32We're doing this agentic AI. We're going to be able to sell this software into other businesses. And also we're going to streamline our own processes and reduce our own overhead costs. It's probably a path to raising right now. So is some of it hype? Sure, it is. But as Will is saying, because Will, you're so steeped in the research and development side of this, there's actually some pretty notable advancements being made too. Let's take a break right now. We'll come right back.

21:06This week on the political scene from The New Yorker, Trump's rupture in the world order. Europe caught between two adversarial great powers. That's basically dialing back the clock to not only pre-World War II, But really, it's a pre-20th century view of the world. And I would say it's a world of permanent insecurity that we're looking at. Join me, Evan Osnos, and my colleagues Jane Mayer and Susan Glasser every Friday on The Political Scene, available wherever you get your podcasts.

21:44So the development of agents is going forward full throttle. But as we were just talking about, the current agents that we're able to use have some significant limitations. But quickly, what are some of the other concerns behind the growth of AI agents? Not just about usability, but about the other things that they do. Well, I think reliability is certainly one issue because they're still subject to the same kind of hallucinations and errors and even sycophancy that, you know, the LLMs are. You're also introducing much more complex queries into it because you're not just asking it to run a search for you.

22:20I mean, you're asking it to perform multiple steps under certain conditions. There are also concerns around high cost and compute power. And then there's selling the product itself. The information recently reported that OpenAI might charge up to$20 ,000 per month for some of its more specialized AI agents. And supposedly, these are agents that would be able to perform at the level of expertise of a PhD graduate, you know, supposedly. and on the one hand, that's really, really expensive. On the other hand, our colleague Paresh Taveh has been reporting how some people in the industry believe that's actually not that much money because ultimately if these agents replace jobs, which is what they might do, then you're saving on all those salary costs and so might actually be something of a bargain.

23:10Right now, though, there's a bit of sticker shock around that. Yeah, I think that's very true, and there are a bunch of other issues, things that could go wrong and probably will go wrong. There are a bunch of researchers I spoke to, Zika Coulter at CMU, who's also on the board of OpenAI, who's been looking at the security of agents and just introducing agents that are going to ideally in future interact with other agents. You have this kind of this sort of amplifying effect that could be very difficult to predict. As Lauren says also, the idea of these things sort of buying things inherently makes them a much more, a much juicier target for hackers.

23:52And if you have something like Mariner using the web, it's really interesting just to think about the idea that people will, you know, maybe try and reinvent SEO to work for agents. And one of the more sort of fundamental problems, I think, is that these aren't really actually replicating human agency. There's a really interesting book on this to go into the weeds called The Evolution of Agency, where if you look at animals and how their agency has evolved, when you get to humans, there's a very fundamental difference in the way human agency works. and it involves working with other humans and involves collaborating and knowing that you have to try and you often have to sacrifice things to work together.

24:38You have things like guilt, you have empathy that evolved. And these models, we're seeing examples of models doing really like crazy things in contrived examples where they're like, you know, told they're going to be switched off and then they blackmail someone. That is not actually that necessarily example of like an evil model. It's just, it could be just an example of a model that doesn't have any, hasn't learned what to how we'd expect a person to behave all the time. And yeah, so that could have a lot of unforeseen consequences. I think my biggest concern is something that you touched on, Lauren, which is the fact that, you know, as the hype and the excitement around AI agents ramps up, that a lot of companies are going to be looking at them as a way to reduce labor costs, right?

25:25There's nothing inherently wrong about reducing labor costs, but I feel like there is something inherently wrong about just like replacing an entire class of workers at a company with machines. It is like quite literally dehumanizing. And I mean that like not only in the jokey sense that you're getting rid of humans, but also just that the people working there and the customers that are interacting with your company are now going to feel alienated from you because they no longer have that human interaction that they're used to. So, you know, we're already seeing that happening. And that's the kind of thing that excites people who are very concerned about the bottom line and worries people who are concerned about things like income inequality and, you know, all the problems that come along with this sort of rampant push towards mechanization.

26:15We've been dealing with this as a society for 150 years. and I feel like as the technology gets more powerful that those worries just accelerate. In the past week alone, I've heard a venture capitalist and a CEO say in separate interviews this quote about how six out of 10 jobs, 60 % of jobs that exist now didn't exist back in 1940. So things change, things shift, technology enables new kinds of jobs. futurists, CEOs, venture capitalists, they love this because it creates this blank space for what we haven't been able to imagine yet in terms of what jobs will exist. And do I believe that there are some jobs that Will's son, who he's vibe coding with right now, could potentially have that we haven't even thought of yet?

27:02Is everyone in Silicon Valley thinking about AI agents in the same way? I don't think they are. I think there's a lot of enthusiasm, but there's also a ton of different definitions. Recently, I spoke to Sonia Huang, a partner at Sequoia, who tracks AI, and she was saying that most agent startups and most of the companies offering agents are really just kind of rebranding existing AI tools as agents. Yeah, to Will's earlier point, you know, there's no real clear definition of what an AI agent is right now or what it's supposed to be, there's really a wide spectrum. And so companies can sort of move the needle and define it kind of however they want to, or say they're offering it in a certain way.

27:51And I think that we're going to also see, I think something maybe of a cottage industry pop up around the spaces in between agentic AI. Like let's say, for example, there is an authentication bottleneck that happens when you dispatch an agent to complete a bunch of different tasks online through a bunch of different apps. There needs to be some kind of connection there between those apps, and it needs to be safe and secure. And so I think you're going to see solutions emerge for that. I think you're going to see some version of like last mile shipping, but for coding. Because right now, there's a lot of vibe coding happening and people talking about vibe coding and look what this bot can do for me.

28:27But not all of that code is actually shippable, right? You have to be able to debug it, and you have to be able to ship it. It has to run, it has to work. So I think you'll start to see more and more startups and companies trying to offer solutions to all these little gaps and bottlenecks that exist between the AI agents. And then our jobs will be taken fully. Good times. Well, I mean, for people like us who are consumers of this technology, I mean, you know, we're not vibe coding like Will, but we are... Not yet. By and large, we're encountering agentic AI in some form in our daily lives. And as it becomes more prevalent, how can we assess its potential and its risks in a level-headed way?

29:11First of all, I think you should all be vibe coding. I'm making all manner of garbage. When I'm vibe coding, I very consciously do it on a machine where I don't mind if it deletes a bunch of files because it is really unpredictable and will sometimes do that. I don't really know how you do that for something that you're going to let loose on your shopping. Do you have a budget that you just let the agent spend? And I mean, I think that's a great question. Like it's not, and the risk is obviously going to all be put entirely on us, I should imagine. Yeah. Yeah. But I think as these agents get better and better in doing tasks for us, we have at least a few responsibilities.

29:50And one is that we do have to sort of disrupt ourselves and our habits and learn how to use new tools because they're going to take over our lives. Two, we have to figure out where we fit in as humans and retain some of that human agency and still participate, stay in the loops, still participate in our own lives, not just let tech take it over. And three, we have to think really hard about the ethics and the privacy and security of what these AI agents are doing as they take over people's lives. The promise from all the optimists is that this is going to close the equality gap. Once again, it's going to be great for everybody.

30:25We know almost for certain this is not going to be great for everybody. Yeah, and that's why I am sort of the opposite of the optimists at this point. I think that great caution and great skepticism are things that are given here, right? We need to make sure that when we adopt these technologies, we do so in a way that is like very, very thoughtful and cautious. And this is something that feels familiar. Probably the most recent example is the phone, the smartphone, something that causes friction in relationships. It's something that you can pay attention to and you can ignore humans. You can use it to interact with other humans.

31:04You can use it to not interact with other humans. You know, this force that has really kind of eroded a small part of our humanity over the 15 or 20 years that this has been part of our lives. And I can see that happening with AI agents as like the things that we used to rely on people to do and the things that used to force us to interact with each other go away. I have the same experience whenever I ride in a Waymo instead of riding in an Uber or taking a taxi. The fact that now there's no human in the equation. There's no person to interact with. And by doing that over and over again, I am slowly eroding my own sense of what it means to be a human being.

31:46I know it sounds really esoteric, but you can chart it over time. And you'll probably find that you're feeling the same way in your own life. So I would encourage people to make that decision about how they want to use AI agents. And if they don't feel like this is something that belongs in their lives, that they should vote with their dollars. Maybe we also need a kind of renewed open source movement. So we're not just using agents that, you know, belong and funnel data to these giant companies. These open source codes and models and that sort of thing. That's a very good idea. We should get Signal on that.

32:20They should start doing that. They should make their own model. I think that sounds great. OK, let's take another break and we'll come right back.

32:54minute podcast where a team of award-winning reporters helps you make sense of our evolving economy. Listen to Marketplace Morning Report on your favorite podcast app.

33:09Will and Lauren, thank you for an invigorating conversation. We're going to put AI agents to the side for a minute and get human once again. Let's talk through some recommendations. Will, as our guest, please recommend something to our listeners. Okay. So I talked about this earlier, but I want to make this my recommendation. I'm going to hold it up, which is great radio, which is The Evolution of Agency by Michael Tomasello. And I found it just fascinating and really kind of really revealing about what's missing with AI when it comes to agency and the importance of sort of understanding human social interaction, human culture, when it comes to the big picture of intelligence and AGI, which everybody talks about, but they don't really talk about that so much.

33:56Nice. Lauren, what's your recommendation? My recommendation is local news, and in particular, Mission Local. Mission Local is a local nonprofit news organization here in San Francisco that covers the Mission neighborhood, but also San Francisco broadly. They do a fantastic job. There's a lot going on in U.S. cities right now, in particular Los Angeles and also in San Francisco. There are ICE raids happening around the country and people are taking to the streets to protest them. Mission Local has been doing a great job covering what's going on in San Francisco so far. And so I recommend that you support them and support your local news if you can.

34:33Mike, what's your recommendation? I want to recommend an essay in the current issue of Harper's. It is by the Norwegian writer Karl Ove Knausgaard, who, Lauren, I'm sure you're sick of me talking about. But it's a fantastic essay. It's called The Re-Enchanted World. And it is Karl Ove reckoning with technology. He tells you about the first time he encountered a computer, which was like 40 years ago. And then he just has not paid attention to computers at all. And this starts to bother him. There's a great quote near the top. To keep somewhat informed about the political situation in the world is a duty, something one has no right to turn away from.

35:16Shouldn't something similar apply to technology, given the immensity of its influence? And I love that quote because it really gets to the center of what we're talking about today. In order to engage with the world, you need to understand how these systems work. He goes to a Greek island to visit the writer James Bridle because their book, Ways of Being, is like a really good introduction to intelligence and artificial intelligence and natural intelligence. So they talk a lot about artificial intelligence and how it's developed and how the various ways of computing intelligence have shown up in our lives.

35:54So it's a really wonderful reported piece about the current state of technology in our lives. Great. Yeah. So it sounds a lot shorter than his books. Yeah. It's like you can read it in under an hour. All right. Sounds good to me.

36:12Thanks for listening to Uncanny Valley. If you like what you heard today, make sure to follow our show and rate it on your podcast app of choice. If you'd like to get in touch with us with any questions, comments, or show suggestions, write to us at uncannyvalleyatwired.com. Today's show was produced by Adriana Tapia. Amar Lal at Macro Sound mixed this episode. Jake Loomis was our New York studio engineer. Megan Herbst fact-checked this episode. Jordan Bell is our executive producer. Katie Drummond is Wired's global editorial director. And Chris Bannon is the head of Global Audio.

From the publisher

In the past six months, OpenAI, Anthropic, Google, and others have released web-browsing agents that are designed to complete tasks independently, with only minimal input from humans. OpenAI CEO Sam Altman has even called AI agents “the next giant breakthrough.”  On today’s episode, we'll dive into what makes these agents different from other forms of machine intelligence, and whether their capabilities can live up to the hype.

You can follow our hosts on Bluesky—Michael Calore is @snackfight, Lauren Goode is @laurengoode, and Katie Drummond is @katie-drummond.

Join WIRED’s best and brightest on Uncanny Valley as they dissect the collision of tech, politics, finance, and business, from Alexis Ohanian's newest tech venture to the effects of inaccurate information from artificial intelligence (AI) chatbots on social protests.

Learn about your ad choices: dovetail.prx.org/ad-choices

More from Uncanny Valley | WIRED

All episodes