Ethan Mollick on Staying Human at Work | Dreamforce 2026

Hi. Good morning, everyone. I have been so excited for this conversation. Well, first of all, thank you for coming on a Thursday morning. I don't know about you all, but I am tired and I have a headache. How many of you went to Usher last night? Amazing. Okay. Well, we are here today with none other than Ethan Mollick. We are We're super privileged to have him here to help us close out Dreamforce this morning. And we're gonna have a really great conversation about AI and work. I should introduce myself. I'm Paula Goldman. I'm the Chief Ethical and Humane Use Officer at Salesforce. I spend all day thinking about AI and work and what we should delegate and maybe what we should keep for people and how do we make AI safe to work with and so on. And I've been so excited about this conversation because Well, you all have a copy of a book on your seat. It's called Manage the Machine. It came out on Tuesday. Actually, I'm gonna ask you to make my mom really happy and hold it up so I can take a picture of you doing it. Will you do that for me?

Okay.

Sure. Thank you. Thank you. Thank you. I'm actually gonna take a picture of Ethan here. Thank you. All right, so, so while I was writing that book thinking about, you know, AI has gotten really, really good. It used to be that, We would, you know, say, write an email for me, and then we would go back and forth with it and we would check it and then we would press send. All of a sudden we have AI agents that are capable of doing really sophisticated tasks over longer and longer periods of time, something I think Ethan's going to talk about. And then it requires a lot more judgment and it kind of changes the paradigm. What do we delegate? How do we delegate well? How do we work with these new tools? And how is it changing the nature of work? And while I was writing that book, Ethan's first book, Co-Intelligence, was very much top of mind. And he has a new book coming out in—

A month.

A month.

One month, yes.

Okay. Called Coexistence, which I was fortunate enough to get to read. And we're going to talk a lot about the state of AI today. There are few people who have thought more about this and who are more influential in thinking about how we think about AI and work. Ben Ethan, who's a professor at Wharton, a very well-known author. Thank you for joining us.

Thank you for having me. This is great.

Oh my gosh, it's so exciting.

I really appreciate all the pre-applause. It really raises the— no, no, I don't. No, it's just scaring me, guys. I have to mean it after the fact. Now it feels like I'm begging for applause. This is going very badly. Let's keep going.

No, we need it. We need it. It's Thursday morning. I had a drink. We need it. Okay. All right. All right. So for years, let's start with, for years we talked about this concept of having a human in the loop, something what we— I was just talking about that, right? We can't complete work without having a person involved in every step. AI agents change the paradigm, but it means that the choices that we are making become more consequential now. How should we think about that?

Uh, so I mean, of the rules in my book, I had these 4 rules in the first book. The one that has become most obsolete most quickly is probably the idea that you have to be the human in the loop, at least in the way that we were viewing it before. Like, if you think about co-intelligence as the chatbot era, right? You ask an AI to do something and give you an answer, you get an answer back. back and forth. And like the truth is that as of about, you know, 8 months ago, 9 months ago, we entered really true agent era. Agents were mentioned all the time before that, but now we have the actual agent thing people have been talking about that I've been writing about for a while, which is ask an AI to do something, it does it. And I have all sorts of weird projects that are, you know, the AI, I can now delegate, write a unique paper in economics and I get a paper an hour later. I could say, make a video of me singing, you know, in a 1980s rock band about agents and I get that. Like, I don't know what it's doing. I don't know how it's delegating stuff. It's making lots of choices, lots of judgment calls. And so I'm in the loop the way a manager's in the loop, right? I am assigning a task. And so that intern model of AI, which is like, it's gonna make mistakes, it's okay, that non-threatening sort of angle is sort of vanishing. And in some ways that's great, much more powerful. And another way it's kind of unnerving because I can ask for things and they get done. I don't actually know how they're getting done. And is an agent swarm attacking everybody right now to get information? I don't actually even know what's happening.

So, Yeah.

It's a big change.

So, okay, so here we are where, you know, you just gave a whole bunch of examples of things that we, you know, you could delegate an economics paper. I could delegate actually creating a tool in my line of work to check our products for ethical risk, right? How do we— there's a lot of things we should be happy to hand over, but there may be some things that we shouldn't be so happy to hand over. How do we— how should we think about that?

All right. So one of the big concerns I hear a lot about is like, Like, do we stop thinking? And the answer is like probably about some things, yes, right? Like what technology does is de-skill us. My grandfather was an engineer who built the fire suppression system at Cape Canaveral, and he did it with a slide rule and matrix math that I have no idea how to use a slide rule. I can't do matrix math in my head. I can't do matrix math at all. Don't tell anyone because it's essential AI. But it is, you know, but we are in a— Like we de-skilled on that. On the other hand, there's things we don't want to lose skills on, and I think AI is absolutely going to take away some of your knowledge and ability to do things. What we have to do is be pretty thoughtful about what we want to give up, and that's hard because you know we've seen this in education first, right? The first thing everyone does is like learning is hard, and it feels annoying. Why do I have to solve the same problem fifteen different times? I'm just going to let the AI do that. Then we don't gain the basic skills. So We're not very good at deciding what we should de-skill on. That's part of what companies and schools and things need to help us think about. But I mean, we're absolutely gonna lose some skills, but that's not the same thing as damaging your brain, right? Like we get to decide what we're outsourcing and what we're not. And people who are deliberate about that are gonna end up in a very different place than people who are not deliberate.

So, and you're seeing this all the time in an educational setting. You're also advising companies who are dealing with this. In fact, actually the question I get most frequently is, How do we think about entry-level roles, right? So if actually what's becoming more valuable, at least one of the arguments in my book is that actually human judgment is becoming more valuable as AI grows in power. How do we make sure that people actually develop the expertise from which to apply it?

So, okay. And I will say that at one point I'm disagreeing with you because I think human judgment's overrated. Not because it's not good. Human judgment is important and valuable. But we— I worry a lot about shining lines in AI discussion.

Yeah.

Like, it was creativity before, it was judgment. Like, there is no— if you have a 4-hour, 4-day-long running task, the AI is exercising judgment all the time on that task, constantly making decisions, constantly exercising creativity. You cannot do tasks without judgment and creativity.

Yeah.

It may not be your judgment and creativity. That's an important distinction.

Yeah.

But to say it can't do it is a problem.

Yeah. I don't think I disagree with you on that, by the way.

Yeah. So, you know, but I think people take comfort in stuff that I wish I could give you more comfort, like, oh, this is irreplaceably human. That doesn't mean humans aren't irreplaceable. But, like, I think we have to make decisions about what we wanna give up and what we don't. On the bigger issue of training, like, you know, I worry about this a lot because I teach generalists at Wharton, and right now everything is kind of screwed up in education. We'll figure it out. I mean, education's always screwed up. We're gonna figure out the right path through. AI tutors are really powerful. We'll find the right balance of testing people in classrooms or not. We'll be fine. I'm worried about you guys because I— and all of you there because I send you generalists and they become specialists. Same way we've done specialists for 4,000 years. Apprenticeship. Right? And apprenticeship has always been a really good deal. I send you someone who's really junior who you can underpay and they'll do the tasks your middle managers don't wanna do. And they'll show you— try and show you whether they're good or not. You can choose to hire them later. And in return, they learn the ropes. Right? They get feedback. Hopefully, I'm a really good mentor. Sometimes the mentor yells at them. Either way, they learn what they're doing and it, it pays off, right? And they get paid for it. And that all broke down because if you're an intern, you've realized that the AI is better than them at, better than you at doing the work because of course it is because you don't know anything. So you, you're, you'd be dumb not to use AI for everything because you want a job. And if you're a middle manager, you've realized that you can assign work to the AI around the intern. You don't have to manage people and people sometimes have feelings and stuff and you don't wanna deal with that. So I'd rather have the AI do all the work. And companies are like, I want efficiency, and learning is inefficient. So if everyone uses AI for everything, we get higher efficiency, and therefore we don't need— you know, who cares what happens long-term? And the training pipeline just snapped entirely.

Hmm.

And the way forward in this has to be starting to realize we need to build inefficiency in some ways into what we learn. Either you're going to need to turn to universities to help you think about how you teach people and how we assess people, and we'll have an extra year of Salesforce University at various schools like the way we do with MBA programs, or else you're going to have to take time and make teaching actually central to what you do and say, we're going to have to assess people without AI and we're going to have to have moments where they're inefficient and they're getting feedback from humans. And I think the companies aren't always ready for that. The companies that do will thrive, though.

Well, it's interesting because I heard you say earlier this morning that the worst scenario would be where we all end up just supervising AI, you know, just kind of saying, Okay. Yes, no. Agree. Yes, no. Why would that be a bad scenario?

I mean, it's bad in a bunch of ways. I mean, one of them is nobody likes verifying stuff. It's the worst job, like the worst job in the world. Like every time I write something and there's a proofreader, I'd be like, I feel so bad for them, right? Like there's no way. We've all proofread stuff. It is awful work. So if your job is reduced to all the interesting work's done by the AI and your job is both proofing it and then agreeing that you will be fired if it's bad, That is the worst possible outcome in an AI world, right? And it doesn't have to be that way. I worry that if you— we just let the sort of AI world progress the way it will, that's what we end up with is like humans are— you get fired and you're responsible. Like, it's just awful. And instead, we should be thinking about the way we do this with other labor-saving or drudgery-saving techniques. How do we get AI to bring us the most interesting problems and do the worst stuff so that we— our life is a series of interesting decisions and moments where we can be additive and useful ways. And I worry a lot that unless we do design work around that, we end up with the default, which is how do we automate work?

Right. And I think that's a lot of what we've been spending time on, at least on my team, as we're thinking about designing our platform, is how do we make sure that for things that would be like you're checking yes, yes, yes, yes, no, no, no, no, no, and you stop paying attention, how do you get AI to supervise that versus how do you actually put a little bit of positive friction into the experience where judgment is actually truly necessary? And/or how do we make sure that we're designing those moments where people are thinking before they're even handing it off to AI and so on? It sounded like you wanted—

did you want to say something? No, I mean, I agree. I mean, I think I just worry about everything being a temporary state of affairs. Like, you could design a world right now with the current systems are jagged enough, right? They're good and bad enough, enough stuff that You cannot hand over large job functions to AI yet, right? Like, it makes judgment— like, all the comments are sort of the same. It sort of looks the same. It's like any of you who have turned to— like, everybody who turned to AI writing 6 or 8 months ago, like, it looked good for a little while until now. Everybody gets ploddish at first, like, and your eyes start to bleed as soon as you read a LinkedIn post now that's like, you know, tells you it's carrying the weight and that, you know, the, um, and like, it's just the worst, right? Like, but we're not gonna stay there forever, I don't think. And part of this is how do we build for the future systems we want? Rather than just being reactive to what the systems could do.

You see, I'm throwing away the note cards, y'all. I'm throwing— we're just going to do it. Okay, so let's run with that. The state of AI today is not going to be the state of AI literally in an hour or tomorrow, and it's going so fast. I actually think this is a place where I feel very humble about it because my predictions were totally wrong a year or two ago. And, but like fast forward this, I think you're right actually to say we shouldn't be drawing bright lines about what's human and what's not. And AI is capable of creativity. It is capable of judgment, maybe a different kind of creativity and judgment. Actually, let's start there. What do you think about, is there a difference between AI creativity and human creativity?

So, We have a lot of studies of this. Creativity is really hard to measure. Okay? So we measure it with things like the RAT test, the Remote Association Test, where you're supposed to name, you know, 10 very different things from each other. Or, you know, how many uses can you, how many uses can you make of a mousetrap? Or that, like, those are the kind of measures we have. And, you know, they, they're good for human creativity. Like every other benchmark, they sort of break down with AI, but AIs outcompete almost every individual human in terms of the creativity we have. However, we have a lot of research that also shows, including some being done by my lab, That shows that the ideas tend to cluster more than a human, right? Like, the AI—if you're used to this, right? Like, it's like AI is actually not a bad writer; it's just one kind of writer, right? AI is not a bad thinker, but if you use AI a lot, you probably see that it has like the obsessions of a 2022 era Silicon Valley person. Like, it loves NFTs and VR, and will always like suggest those kind of like and you know and carbon neutral setting. Like, it has a bunch of stuff it's really interested in, and it comes back to those themes. Now, I can tell you. We have research that shows better prompting techniques and other approaches will make the AI more diverse, more like a set of human ideas, but it's still not as diverse as really bringing humans to the table. We generate ideas people don't. So one source of human advantage is variation. We bring more variation to the table and, you know, the whole world isn't Claude, right?

Like, can we pause on that for a second? Why is that though? Theoretically, AI has access to way more data than I could possibly keep in my brain. So why would having humans bring more variation to So there's a few things.

One is it has access to the data, but it's trained on that data. And then there's a series of training steps the AI goes through to enforce the way it thinks. One of the weirdest things about AI, by the way, about these systems is whether you're talking about, you know, OpenAI or Anthropic, or you're talking about, uh, the, the Chinese open weights models like KIMI, or you're talking about Grok, which was specifically built to be very different, they all have almost the same kind of political views and value system. And we don't 100% know why that's the case. Some of that has to do with the training system. Some of it has to do with the data. It is possible to push AI from its sort of centroid area. And by the way, the wrong thing to think about is that AI gives you the average answer of the internet. It does not. It, it gives you its own thing that's much better than the average answer of the internet for a bunch of really interesting reasons. But regardless, it tends to be fairly similar. Now you could destabilize that a bit by being, giving interesting prompts, but then you're also adding your own variation to this.

Okay.

That might be a solved problem at some point, like, but right now that is a very large jagged frontier gap.

But you don't think though that, so there's a difference between saying AI gives an average answer versus AI can revert sort of to the mean. Is that?

It does not revert. So, I mean, part of what's really interesting about AI systems is you could think about it as like a group of experts working together, giving you an answer. And so because the AI can essentially boiling it down, can route answers to the right experts. If basically if law experts have written more on the internet about law than non-experts and marine biologists have written more about marine biology than non-marine biology experts. The AI kind of learns the biases that if it's a marine biology question, it'll answer it with the view of those experts versus the view of another set of experts. So it actually is not pulling the average answer. It's sort of taking polls of the right sets of people or the right sets of its training data.

Got it.

So it makes it hard to sort of say it's an average thing.

Got it. Okay, well, I'm gonna push on this one more thing and then we're gonna change direction. But so there's a story in my book about innovation experts at IKEA who were trying to break down the concept of a couch and come up with a different prototype for a couch. And they kept being like, hey AI, give me a different answer. And it kept reverting to boxy, cushiony. And they realized before actually asking AI, they needed to challenge their own thinking. So they sort of started to come up with metaphors like campfire or gathering spot. And they ended up coming up with this prototype that was lightweight called Couch in a Box. It was like a very different thing that ended up getting exhibited at all these museums. And I'm curious, thinking about that example and thinking about your LinkedIn example of how people start to sound the same when they use AI, if that's not reverting to the mean, what is that? Is that like, that's just—

So I'm gonna give the world's worst answer to this, which is, it's a prompting skill issue to some extent here, right? Like, you are not really, like, there is a lot of stuff on LinkedIn that is absolutely terrifically written by AI, and the pangram will tell you it is, right? Yeah. for now, a good detector, but like, that is, doesn't read like AI writing. So there is like a little bit, like, by the way, this is where we can talk more about wide knowledge, but like knowing writing is helpful because you can tell what's good and what isn't.

Yeah.

In the same case with the IKEA thing, the, by the way, the right way to do any kind of ideation session looks the same as with AI. A brainstorm was invented in 1955, and by 1957 we knew it was a terrible idea, but we keep doing it because it feels good. But like any ideation session should always start by you writing down your ideas individually. Before you talk to any other person so you don't pollute your mind with their idea. Same way with working with AI. You should absolutely write down ideas on your own, then you have the AI do it. Then it turns out if you push the AI, we found you can generate like 1,000 unique ideas from the AI. You just have to keep pushing it. And if you say things, make it more extreme, how would a superhero use this? You actually get much more diverse ideas. You're still doing selection. I mentioned wide knowledge. Taste is another unique thing about humans that were at least semi-unique for right now. That we have good— that our taste matters, right? Picking the bad LinkedIn answer versus the good one, that. But it's not the average answer. You can get a huge diversity of answers. People are picking the average answer or they're satisfying themselves with one answer. Why would you get one answer outta the AI when you get 3,000 and you can pick the right one?

Okay, so it's a prompting problem. It's a way we should be addressing. Okay.

The most annoying thing you could ever say is a skill issue, but this is a little bit of a skill issue.

Okay, okay. So, so you're making a pretty compelling case that AI is getting better and better, that we shouldn't have bright lines about what we think that people should do and AI should do. But you also just have a book coming to market about the case for staying human at work, if I'm not mischaracterizing it. So if you project out, I mean, there is so much anxiety about a possible jobs apocalypse or what it— why are you optimistic about it?

So, okay, so there's the jobs issue and then there's the human issue, right? And the jobs issue, I mean, all of you work for organizations and you know the world is not nearly as easy as people think. There's all kinds of frictions. The model that I find very compelling from the economics on this is what's called the O-ring model of jobs. And it's based on the idea of the space shuttle Challenger, tragically, which had 1,600 pieces work right and one thing failed, which was an O-ring that kept gas in and that exploded the Challenger. One failure in a complex system is not the same thing as a small minus. It can stop the entire system. And because AI is jagged, it's good at some stuff and bad at some stuff you wouldn't expect. It matters that it's not equally good at everything that you do, or that parts of it they can't do because it's a machine, or that it's, you know, trapped in, in your environment. And so as a result, human jobs will shift. The value is gonna shift to different parts of that O-ring really quickly. You can see a sample of that happening in coding right now, right? We went from what were you hired before? You were a good, elegant coder who could write really good code to then it became, are you a good prompter? Can you prompt AI really well to get code? And then it became, Are you like— the best engineers I know don't write code anymore. They now— their job is, can they give a good requirement to the AI with the right kind of checkpoints and the right imagination of the outcome? That is, the job hasn't gone away, right? But it has shifted what the value is really quickly as it moves along that O-ring. Now start stacking those in an organization and you realize there's lots of like points of friction.

Yeah.

I've been, you know, in my little things at Wharton and elsewhere, We've been hiring more coders because we have more ideas for code. But the last little bit, which is not always, by the way, boring verification work, but all of the kind of pieces we're missing, we need actual coders to do that work. I don't think we know what 10 years from now looks like, right? But I think that this transition is not going to be as clean as people think it is. And that's separate from the idea of what we bring human to the table, which we could also talk about.

Yeah, well, I would love to talk about both of those, but I actually think the coding example is a great one because I think that's right. When we say or when I say, you know, people are gonna be managing machines, the idea is what you mentioned earlier, like, yes, yes, no, no. But actually what we see in coding is that, in fact, what people are doing now are they're sort of upleveling the, it's sort of, we're thinking about the architectural questions and how the code fits together and what the values are that go into it and all of that kind of stuff. And I know that's probably not applicable across the board. I think it will be uneven what the impact of AI will be in different roles and different jobs. But it does seem to me like pretty good example of how AI could uplevel what people are doing.

I mean, I think it's going to be chaos. I mean, yeah, I mean, I mean, for example, on coding, there's a nice paper out of Anthropic that suggests that managers are actually better coders now with AI than coders are because they're better at specifying what the requests are. Like, job— jobs are already blurring in every company I talk to. It's like a complete mess. Like, the senior leadership often doesn't know or know what to do about it, but like, You know, designers are now coding and product managers are doing everything. And like, what do you do with a, like, ev— all these jobs are breaking down. So like, Yeah. In some ways, like, part of what happens is our jobs transform. I'm a professor of entrepreneurship. It has never been a better time to launch a company because companies were all about the fact that you were really good at one thing and then you were bad at everything else, and you had to hope you were good enough at your one thing and lucky enough that the other stuff you were bad at didn't kill you. And now the AI's really good at all that other stuff. So suddenly, Whatever you're best at, you have an accelerator for. So I think we'll see that in companies as well. But again, it's messy, right? Companies don't know what to do with this. There will be job losses. There's going to be transformations across industries in ways that are really hard to predict.

So, well, let's talk about what should stay human. I mean, I'm just gonna open up that to you as a broad question since it's a super confusing time, hard to predict. Like, what are the things that we shouldn't delegate to AI?

So I think there, I, I have a whole list in the book of different times to not delegate. Some of these are about personal improvement, right? You shouldn't delegate to the AI if you wanna learn because you're not really gonna learn unless you're using the AI actively for learning, right? You shouldn't delegate to the AI when, you know, if like, if the human touch is the thing that matters, cuz then when someone finds out, they actually are mad at you as opposed to using AI for other circumstances. There's a lot of times not to use it. On a broader societal level, we have a lot of weird choices to make cuz humans have been the gate for so many different things. Like, One of the things I do as a professor is write letters of recommendation for people, right? Nobody actually cares about the letter. What they care about is that I have set my time on fire as a symbol that I care about somebody, right? And so it takes me 45 minutes to write a letter or whatever it is, and that's an indicator. And, you know, I— and when I ask my students, would you rather have a better letter I write in 1 minute or a letter that's not as good that takes 45 minutes? Everyone's like, the 1-minute letter. And in fact, one of my students actually sent me the prompt they wanted me to use for the letter of recommendation and just say, put this in. And I did because they asked me to. And then I mentioned it. But the— but I think that that's a human system, right? There's a whole bunch of these kind of gates where we assume that the work involved was an indicator of thought or effort. Essays are the same kind of way, right, as a grader. And so I think one of the things we have to reconstruct is what does it mean with all of this human effort stuff? I mean, nobody wants more PowerPoints, but that is the first outcome of AI in every organization, right, is you're drowned under work slop. And that's because A, you're incentivized to create PowerPoints, but the PowerPoints were never the point. It was supposed to be that that PowerPoint was a debate and an argument between people and a research effort. So we have to think hard about that. We can't just say PowerPoint should stay human or this policy stays human. We have to instead be really deep about what is the output of the job that mattered? What did we want a human to add to it? And I mean, that's tough because organizations don't change quickly. So I think we're going to have this gap period where people are automating their work and doing much less work or whatever, right? Or worrying they're going to get fired. But the fundamental work is still, you know, is different from what you're being measured on.

Yeah, no, I agree with that. And I actually think that type of analysis is too rare where we're actually talking about the work itself versus just AI. I want to talk about what happens when we introduce AI into teams, which is happening very quickly. I was very excited when Slack introduced kind of the multiplayer AI piece of things and I've also, like, I have a whole chapter in the book about human-AI teams and about how, for example, when AI is not well suited to a particular task, it can disrupt human coordination and decision-making. I'd just like you to talk a little bit about how you think about that. Like, is AI a teammate? Should we think of it as such?

So I think if you think about that, like, this is the big question, right? And I think you could think about this as like there's a couple of ways to view AI. You can either view it as a person Or as a tool, right? And then you can either view it as sort of sustaining what we already do or disrupting what we do.

Yeah.

I think of the Slack bot, right, as an example of sustaining. Like, it's just like it's a person on the team acting like a person, right? And that is a sort of sustaining kind of innovation. And it kind of makes sense, except it starts to break down when you think about it. If the teammate is doing 18 days' worth or 18 weeks' worth of work, and there's a paper out of Epoch that shows for $240 you can get You could have gotten Claude Opus 4.8, not even the most recent models, to replicate software that take a human engineer 18 to 52 weeks to do. Like, that's not an analogy of, like, assign this on Slack. But we also have problems like in Slack, we know who's in charge. We know, like, what the dynamics are. Until the AIs are more socially aware, which, by the way, is not an AI problem, they can be, but we're not thinking about them that way. We're still thinking of them as, like, tools or interns. When you start thinking of them as teams of people that also have to respect boundaries and know who they're doing it to, That starts to change things. Then it starts to become a world where we go to this sort of dark factory, Twilight Factory setting, where we think about the AI is running things and reaching out to humans when it needs help or when we want humans to do things. So I worry a little bit that we're kind of an interim step a little, which is like AI as a teammate is not the ultimate form of multiplayer, right? Because if it's doing all the work, then we're having a chat conversation. And, you know, and also when it becomes cheap enough for it to fail, who cares, right? It failed. It just took you a few minutes to to do a task that would've taken you 3 weeks. Why not have it make the whole prototype? Why not have it open Blender and do all the work for you and give you a video of what this would look like in the end? Because it's cheap and easy to do. Yeah, that changes how work operates. So I worry a little bit that we are thinking short-term about that.

I— yeah, we're probably thinking short-term about a lot of this, right? Because it's hard for us to conceive of exponential technology and what the impacts are and so on. You also mentioned earlier that one of the things that we shouldn't hand over is when it's really about human relationships. I tend to believe we're probably underestimating how much of work is about human relationships. However, that is not to say that it necessarily should be that way. But like, as I was, you know, navigating what happens in sales or customer service relationships across the board, what are your thoughts? Like, do you— how do we use AI in such a way as to strengthen human relationships rather than undermine?

I mean, so I think that part of this is that there's a very plausible retreat to human kind of part. Like, what we're gonna end up, I think, doing, if in a good world, right? And we can talk more about how to get a good world.

Yeah.

Is that you're gonna outsource the parts of the human relationship stuff that is about— that's transactional in nature, right?

Yeah.

Like, I, you know, I, I mean, I will tell you, Astra and Fable have gotten good enough that they— I don't let them write to any of you, so don't worry, but I let them read all my emails kind of coming in, and they give me incredible— like, they— I mean, they found it. I mean, they find all sorts of, you know, This was an opportunity or an angle I didn't think about that was interesting to talk to someone. Here's somebody you really would want to speak to and you would have missed it. But even things like I sent in a tax form, information about it, and the AI sent me a message the next day saying, actually, I checked it and you put the wrong number in your tax form that you sent in. Right?

Wow.

So there is a removing some of the friction of human relationships that we don't want to retreat to a more kind of human space. I mean, there is a lot of humanness that we need to kind of embrace here. I talk in the book about 4 things that you do. There are 4 things that are human traits that are worth kind of building on, not necessarily irreplaceable, valuable to us, which are, which are, we talked about one of them, right? Which is the idea of taste and selection. Like it matters. Yeah, I can do many of a thing. You want to pick among them, right? The bad LinkedIn posts are a result of bad taste. You should be able to reduce many of them and read. But if you're wide, and that fits into, you know, the wide knowledge and deep knowledge. So deep knowledge is deep expertise in a field. That lets you both compete with AI because it's not as good at you at what you're very best at for now, but also lets you make judgment calls along with this. Like one of the most expensive things about AI as we were talking about is the cognitive dissonance of, and we have, there's some great research on this too. It gives you a huge wall of text and you are like, I'm just like distracted and confused. And a lot of the time that AI saves you gets eaten up by the fact that it is really bad at keeping you on track cognitively. Right? And so, but deep knowledge lets you easily check the AI's work without getting distracted. Like, Your manager, you can look at that spreadsheet. There's something wrong with it. That is very useful. Then there's wide knowledge. We've talked about this a bit too, like knowing many things about many topics lets you better decide how to use AI. So if you know many forms of design and you can invoke Bauhaus by a word, or, you know, or, you know, or, or, you know, successionist or whatever it is, you're in much better shape. You know many authors and their style and their styles of writing a little bit like this. No, you're going too Hemingway. Dial it back. Like that kind of stuff is very helpful. And then there's agency, like having permission to do this, which is not something everybody has, but we need to do more of. And, you know, gives you some options as well.

Okay. Well, I've been— we're running out of time. I've been looking at this audience kind of glancing over and rapt attention on everything you've said this whole time. It is, it is a confusing moment. I think you've, you've sort of left us with this impression of just how quickly things are changing and how blurry the lines are. But throw us an anchor. Like, give us a reason for hope and agency.

I think there's a tremendous amount of hope and agency. I mean, first of all, AI is incredible, but not as good as everyone thinks it is. Right? So, like, people are, you know, like, you've been talking AI all day. I talk AI all day. It's really easy to get kind of caught up in, like, it'll do everything. It does everything. That jaggedness is real. And the edges and the O-rings are very real. And, like, if you actually use AI for tasks, you realize that your humanness is what makes it good at a task or not. But let's say even that that's evolving, right? And changing. The thing I wanna tell you is nobody knows anything. We are all making this up as we go along, but that includes the AI labs that are shipping this stuff to you. There's not a plan here. The systems are really good, and that gives us a unique moment because the systems aren't good at everything in the same way, and the AI companies don't know what they're good or bad at. We get to decide how they're used. And the thing I wanna charge you all with is we need more good examples of AI for augmentation, not automation. The default view is AI replaces human, it does human work. That doesn't have to be the case, right? We can build worlds where AI augments us so we do the work that's interesting, the human work, the work we care about, the work that expresses our variation rather than just being, you know, slaves to the bots. And we're in a really wide open space where everything is in the air right now. And people who show use cases that work, those will be copied and used, right? And so obviously there's policy and other ways we shape what happens, but I really think that the idea that You shouldn't just take this as received wisdom that this is being done to you, that this is something we could do something with. And the more you think about how to do it with it in a way that amplifies what we're human, and the more you share those examples, and I'm— it sounds corny, but I'm absolutely true about this. Everyone's desperate for these examples. And I worry that a lot of people, I was speaking to a couple nonprofits before, are sort of sitting back and being like, no, you know, like there's some of the people who think most about being human are too worried to use AI or show these examples or share them. I think it's really important that we get out there and give examples of how AI is transformational for good, because the labs will jump on that. Software companies like Salesforce will jump on this, but we need to be inventing those ourselves and not hoping someone else, else invents it for us.

I love that. So no one knows what's happening. Everyone's making it up. And your humanness is what is going to make it great. And we need more examples of positive use cases that are genuinely augmentative. Will you please join me? This has been such an amazing conversation. Please join me in thanking Ethan for bringing— Thank you. Thanks, everyone.