Reid Hoffman On AI, Superagency, And The Future Of Work

I'm honored to be here. I look forward to the conversation.

So, my day is full of a lot of different things. Obviously, one of the primary things is around AI. So, you know, having co-founded Manas and Inflection, you know, the question about what does AI mean for the progress of humanity and society and obviously the commercial vehicles of creating companies to do that, whether it's an AI drug discovery company using AI's superpowers to try to cure cancer and other kinds of things. Whether it's kind of the question around how is our world going to change when everything that we do will have AI agents with us and around us in terms of how we operate and what does that mean for becoming more human in terms of what we do. So, that and then obviously I do various forms of content, whether it's podcast like Possible or Masters of Scale or coming here and talking with you. And then also try to like how do you help shape like society, right? So like, you know, here we are in London, you know, go and visit 10 Downing Street, which is very curious about how they use AI to help the UK economy and the UK citizenry and so forth and so they're always looking for bringing in intelligence because they got very smart people and are doing some really great things.

Well, so it's the right thing to do from a strategic point of view of how do you get to the good futures to be a strategic optimist. Right? Because you only get to the good futures by steering towards them. You don't get to the good futures by steering away from the bad ones. You must also steer away from the bad ones, but simply steer. It's like for example, if you were saying, 'Hey, we're sitting in London, let's drive to Oxford.' Well, but first we must plot out every possible thing that must go wrong. We're never going to get to Oxford. You go, 'Look, the car is working, you know, there is the road's not closed, let's go.' Now, part of being strategic about it is to say, 'Okay, being optimistic and choosing where you're going is an intrinsic part of creating a good future.' That doesn't mean that you don't pay attention to concerns. It doesn't mean you don't pay attention to them in terms of how you navigate. So, it's good to have that dialogue and think about what kind of things could go wrong as you're heading towards what kinds of things could go right. And the problem is if you have a conversation that's doomer or is, you know, you're quoting from super agency, gloomer, you're like, well, you're inevitably going to bad things if that's all you focus on. You have to go, here is how we get to the good things. And that's what I try to do.

So, a simple way to look at this is what kinds of agents, so we have them on all of our smartphones, would make human life a lot better. Let's start with a medical assistant. Today, you can have a medical assistant that's as good as your average doctor. Right? It has some things that's weaker, some things that's stronger, but like you could have that on your phone. Now, you think here we are in London, like think about if the NHS had that as a front end that said, hey, oh, you have a serious thing. We need to check you in, we need to get you to the hospital today. We need to check you in, have a meeting with your GP tomorrow or the next day. And it was doing the triage and countering, but also could say, oh, that's not so bad. You know, for that, just take a Benadryl or whatever the equivalent anti-histamine is here. Huge evolution in human life. Next, legal assistant. Most people cannot afford lawyers. But like, okay, what about the rental agreement? What about this thing? What about this issue and so on? To actually help you understand what your rights are. Interface with government. What are the government services that could work for me that could help me in being able to do that including like hey I want to report this thing that's wrong at the street corner or something else. That's really good. An educational assistant. Like I want to learn this. I want to be able to learn a new profession. I want to figure out what kinds of other work I can do. Help ideate with me. All of those things governments and you know should say we want to get those as free assistants and make sure that they're enabled on every single smartphone for everyone. And that's part of how we navigate a more turbulent future because of course AI is going to transform jobs, right? Like a certain set of tasks that used to be done by humans will now be better done by AI. By the way, that's just something really good for you. Like say for example you said, well my job was I was doing digital form entry for advertising. Okay, AIs are going to do that a lot better. But that means that now your job is okay, let me think about how to do better digital marketing. Let me orchestrate a bunch of agents in doing that. It is a transformation. Some people will find that painful. It is what it is just as when we had cars, we moved away from horses as transport. It's better ultimately for everyone in the world. But the transformations are sometimes challenging.

So there's enough that we could cover the entire podcast answering that question. But I'll start with three. The first, everyone thinks, okay, yeah, coding assistance, obvious, but it's the second order effect from coding assistance. So, it's not just that now everyone can start generating code. Like, for example, everyone can actually generate a smartphone app today. But, it's like, what does that mean for now better reasoning? Like, for example, if you're a financial analysis, the coding agent allows you to do all kinds of new things in financial analysis, and allows you to have superpowers that you didn't have before. If you're, like, for example, trying to figure out, like, for example, say, doing home finances, you could actually, even just not being a financial analysis, do that. If you were kind of saying, like, for example, one of the things we do is we create AI music. We have AI assistants, agents, helping us create the AI music and kind of iterating through it as a way of, like, expressing, like, giving a birthday gift to a friend. So, there's this all this range of things that are the second order effects, in part because the coding agents are like thinking, reasoning, working machines that can not only underlying coding, but underlying everything else. So, all of the development of coding is something that I find to be extremely interesting. The second area is like memory and personalization, because part of the reason why I think part of the reason why, you know, kind of we started Inflection AI, is a notion of we'll have AI agents that should be ours that go with us throughout our entire life. They're not just provisioned by a company, but it's like my agent for me that understands what I'm looking for, what I might need, and it be a good foil or an additional voice on how do I navigate what is this new world and kind of information. So, and a key part of that is the memory of you. Now, part of that memory of you is also like, okay, so like how do you learn something? Or like kind of what is the real promise of personalized learning? Because we created the best learning technology in human history. Now, it builds on the book, but it's I think even much much better than books as a piece of technology for doing this. And then third and final is, you know, part of what we're doing with Manas AI is to say, well, actually in fact, we use these AI superpowers to transform, because, you know, like what's one of the most central things in elevating human life is human health. Right? So, like part of the reason Siddhartha Mukherjee and Ujjwal Singh and I co-founded Manas is because you say, well, cancer kills everybody. Right? And it young, old, male, female, wealthy country, poor country, the whole thing. And it's very hard, like the solutions are very challenging. You're lucky if it's micro, it's discovered early, and it's surgery out. That's the easy one. Everything else is brutal. Like chemo is we try to kill the cancer before we kill you. And we're just it's poison, right? As a way of doing it. How do we get a much more human way of identifying it, preventing it, solving it? AI is the first like major evolutionary step that we've had as a shot on that in decades.

So, I think it's a definite part of the future and it's definitely very good. The reason I kind of said these things for what is exciting me right now is these are the things that I see progress on, you know, 2026, 2027. It's not to say there isn't progress on this in the physical, but the world of atoms moves much more slowly than the world of bits. And so, and people like, 'Well, you know, you'll have a robot that will do the house cleaning and the dishes.' Like, you know, that's not near-term. Right? I do think we will get better and better autonomous vehicles. One of the reasons I was an early investor in Aurora and other things for autonomous vehicles. I think, you know, people go, 'Oh, is it safe to be in an autonomous vehicle?' And the answer is yes. You should get in a Waymo. You should see them in terms of how this... coming to London? Yes, exactly. And so, and then world models for helping navigate the world is all kinds of very good things that come from that. But I think that the first most interesting parts of world models will be things that are call it high-value things that are expensive or difficult for humans, not things that are common for people. Right? And so, you know, it's like, 'Okay, like I want robots for doing mining.' Right? Because it's dangerous for people. And by the way, we should help miners get other jobs and all the rest. And by the way, some of them will be guiding the robots. Right? Matter of fact, we might be able to do a lot more green, ecological, and safe mining if we have robots with people, you know, kind of doing that as a way of navigating. So, I'm positive on physical world AI and robotics. I just think it's not as soon as some of the press sometimes suggests.

So, lots of ways. And one of the things that I encourage people to do, which I do myself, is I try to figure out new things every week to add to my AI thing because what's almost for sure for just about everybody, like maybe there's only 200 or 1,000 people in the world who are actually using AI to its full capability. It's not very many. Or maybe if you put all the employees of, you know, certain tech companies in, maybe it gets to some thousands, but it's not a lot. And so for me it ranges from let me maybe put it in three buckets, which is professional, fun, and personal. So, professional, like for example, when I'm looking at investments, part of the thing that AI is a good catalyst and companion is like doing deep research on due diligence. Like what questions should I ask? You know, like build me a competitive chart from all the information on the internet. What are the things that you think are the top questions? You know, help me analyze this particular market space and what the competitors are, you know, etc. etc. And by the way, people say, 'Well, you used to do that with associates.' I still do that with associates. We both do that now. Right? With this. And like it expands the areas that I might invest in because previously it'd be like, 'Well, I'd be curious about that investment in fusion, but like to get to expertise in it is way too hard.' Whereas now I can, right? And you still have to think about like, okay, what's my expertise versus other people's expertise and so forth. So, if you go, 'Well, both of us are using AI and the other person's much more expert, they should be the investor.' Versus, you know, considered the investor versus me, but I think it also expands that surface in terms of it. So, that's professional. So, for personal, I'll do that and I'll do fun last. Personal is like, well, when you're you know, trying to like say for example, a friend of mine might be facing a health question. Right? Or I'm facing a health question. If you're not using your favorite frontier model as a second opinion, whether it's ChatGPT or Claude or Gemini or Copilot, you're actually in fact almost doing malpractice even as a patient. Right? Now, obviously there's still a lot of reason to have very high quality doctors as first opinions. Some people don't have access to that and then your second opinion may become your first opinion. But as a second opinion, it's really important. And I literally have known friends of mine whose lives have been saved by using a second opinion because they went, 'Oh, you went to another hospital. Hospital had the wrong diagnosis.' They said, 'Go to a different hospital.' And then literally like, 'Great. If you'd gotten here 3 hours later, we wouldn't be able to save you.' Right? So, that kind of thing. So, the second opinion, and by the way, it's both patients and doctors for it I think is very important. And that's an example of personal. And then for fun, you know, one of the things that I've been doing is I've been thinking about how AI increases human expressiveness. Okay. So, obviously it does that with words. Obviously it does that with pictures. There's various things happening in video. But like part of the opening up music to not just people who are because like I'm one of those people you would pay not to sing. Right? But like I released an animated video on March 14th for Pi Day, right? About celebrating the number Pi and the history of it with Reid AI singing in my voice, because I can then go be geeky about a subject that kind of matters, like math and pi and so forth, in a way that's fun. I can also create music for friends, for like birthdays or other things or holidays. And so, so using AI to increase your expressiveness is another thing that is completely doable and that you know, I recommend everyone do. I think it increases our humanness.

Well, so one of the things is like one of the classic ones is traditionally the way that enterprise will try new technology is run a pilot. Problem is a pilot with AI only tells you something useful in very narrow circumstances. Because it's kind of the question how you work and what do you do? So, so classically they say, 'Okay, group X do it.' Whereas, for example, almost any company should say, 'Hey, we should start using AI in our meetings.' Right, we should not just record them, not just have transcripts, but like the AI should be the, you know, like for example, part of the process through this is, all right, which questions should we've asked that didn't? What are the follow-ups? Are there people who weren't in the meeting who should know summaries or certain like the whole meeting or parts of the meeting? Are there people that we need that have we have action items assigned to? You know, what is the follow-up plan on that? And then like literally AI's transforming this. So, if you're like any enterprise, the short answer is you should go all in on meetings right away. And that will start giving you the lens into what other kinds of work that might be right for your company for doing it. As opposed to like, 'Oh, well, we'll send off Sarah and Bob and they'll go write a report on how we should be using AI.' You're like, 'That's not going to be that interesting.' Right? And what's more, when you get to these questions like, 'Well, I guess we should be using it for financial analysis.' You're like, 'Okay, well, then you say, Well, we need to have a course.' You're like, 'Actually, you don't.' Right? Like AI is the thing to do. You start engaging, start using. And so, one of the things that I recommend people do is to, you know, many work groups have a cadence of a weekly meeting is to say, 'Add into your weekly meeting what's something that you tried with AI and what was the result? Like was the result great? In which case everyone else should be doing that. Was the result mixed? Interesting here, but limited here, etc. Or is the result bad? Not useless.' So, you're sharing. Now, part of it is don't overly conclude useless because, you know, one of the things I love about Ethan Mollick's line is the worst AI you're ever going to use is the AI you're using today. So, what was useless today may not be useless 6 months from now. So, it's worth tracking because it like why repeat doing the same not thing, but it's this question of doing that. And I think that's what makes corporations start succeeding in deploying it. Now, what happens is you get a long list of, 'Well, we don't know about legal liabilities, we don't know about information security, we don't know about hallucinations. We don't know about any of this.' And you're like, 'Well, but what you do know is that AI will be a more and more significant part of the future, such that every company that is not fundamentally using AI will start arcing off.' So, don't use excuses to not take small risks, small challenges, small problems to start the learning curve within your own corporation. And, you know, that takes a little bit of more boldness than most managers, executives have because the classic thing is we keep doing it the way we have. No one ever questions me. If I change it, I get questioned. But, you should seek to be questioned, that's kind.

Learning what those are now and how to integrate it and like, for example, for me, in investing, due diligence, competitive analysis, etc. Great. If I said, 'Should I make the investment or not?' That's not currently useful at that. That doesn't really get to the like, if I proxied all of my decisioning over to, you know, ChatGPT or Copilot or anything else, it wouldn't work. Right? And I tried it with all of them just to see every so often. But, and it doesn't mean that it won't get there in ways that will be much more helpful than they are, but like, because it's not useful there, doesn't mean I stop using it in investing context.