Reid Hoffman, LinkedIn Founder: 99% of People Are Using AI Wrong

Most people are really underutilizing today's AI capabilities, blinding themselves with emotional reactions like AI slop or other kinds of things. But the really interesting use is this agentic framework. There won't be individual human contributors anymore. There will be human managers of groups of agents. It can work while you're asleep. But here's the thing that's a little strange. AI has a negative PR perception in the US and Europe and a positive one everywhere else. And one of my greatest worries for the US and Europe is to say, 'No, it's really important we slow this down.' And in Asia, you go to Japan, you go to China, they're all like, 'Oh my god, this is the future. We want to be on it.' Part of what made the US and Europe so strong globally was the industrial revolution, embracing it early. And so, it's kind of like the reverse revenge of the industrial revolution. Maybe the rest of the world is going to embrace AI and the cognitive industrial revolution earlier. And so it's a huge moment for this kind of wakeup call. If they don't reinvent themselves, then the dinosaur decay has started. Part of the reason why people have negative is because they say, 'Oh, the tech companies are doing this to me. I'm going to lose my job.' And that's the entire story is 99% of the human race. So I understand. But the question is, how do you get to the 99%? Show me benefit. Show me how I can be amplified and using. So you go, okay, AI can be for me. And I think that's the way you persuade people. And it isn't by a story of words. All right. How you get trust is by seeing value. And so the unequivocal thing that I said to all people is it's super important to have reads.

Thank you very much and a pleasure to be here.

First, most people are really underutilizing today's AI capabilities. There's everything from the people who aren't doing it at all for a whole variety of reasons. Lack of curiosity, blinding themselves with emotional reactions like AI slop or other kinds of things. And so, they're not really digging into the pretty amazing capabilities. Then there's a relatively not small group that are kind of using it as a search replacement, which is hey, I get a more coherent answer like is it okay for my dog to eat coconut, or other kinds of things, which gives you a very precise and detailed answer that you can follow up on to check because these things always try to be pretty helpful, so they hallucinate sometimes, although the hallucinations have gone way down in the last couple years, and that's a bulk of use. But the really interesting use is for today and capabilities which most people wonder is this agentic framework of actually doing work, because what all of the labs have been doing is how do you get amplified in terms of how you operate. So for example, given coding has been the massive thing. All of the software developers who are using this actually have a large screen that they have multiple agents running on and they're kind of doing more orchestration. They're like, okay, you're doing this, you're doing this, you're doing this. This is the handoffs. Look, this one's got stuck on something. I'm doing, oh, this one's finished. This one needs to get the next one. You know, this kind of thing. Including and this is obviously one of the things that has worked for the economics and capitalism is it can work while you're asleep. So one of the most classic things these coding developers do is they have big problems that they set when they're going to go to sleep. So they get up in the morning and the thing has been working for eight hours and has a substantive piece of work left. Now you'd say because it's agents and coding and it's only agents coding, but actually coding is a lens into all knowledge work. It's a form of reasoning and any place where you can close the digital loop, and that isn't just writing, that isn't just code, that isn't just images, that isn't just video, it's other things too. Part of the reason why there are startups in law and all the rest is a place that you can be amplified today and you can actually be doing that, and the underdeployment of that capability. I can tell when people are starting to get into it because it's like, oh gosh, I'm using cloud code or Codex and I'm staying up to two in the morning getting stuff done, or I've been really diving into ceramics and now I have a ceramics app because I can do that, and it doesn't have to be like, it's now doable as an amplification for the things I'm doing in ceramics. And that's the kind of thing that you're beginning to see. But that's still a very tiny percentage of the working people. And that's what today looks like. It's all changed so incredibly quickly. And I don't think many of us realize just how quickly this amazing technology was going to advance.

Well, not as a surprise, part of that is the acceleration of capabilities and the digital closing loop. So not only will coding continue to get a lot better, but more and more people who are not coders will be deploying coding to do things, the ceramics app, bunch of other things. And those coding capabilities will then go into all kinds of workflows because you create agentic workflows. So what you'll begin seeing, which you see in a small number of people now, is people going like within 3 years, and I think one year you'll begin to see the broader based adoption of this. There won't be individual human contributors anymore, there will be human managers of groups of agents. It doesn't mean that the people aren't contributing some pretty important things that are not just managing the agents but also context awareness and everything else, but that management of a group of agents will then become an understood developing pattern of work, and every place where you can do the digital closing the loop will be there, and that's part of what will be growing out. Now, in particular, I think you'll begin to see because we've begun in the last year to really see takeoff in coding. I think you'll see the coding go to the next level and then it bringing all the other knowledge workflows into the same kind of things that coding has been doing this year.

Well, I think people are both taking it not seriously enough and too seriously. The not seriously enough is a lot of people not actually engaging, realizing that we've created the best learning and amplification technology in human history. That we're entering into the cognitive industrial revolution and this will be faster and larger impact than the industrial revolution. So that's the not seriously enough. And the too seriously tends to be a little bit of the science fiction thing like, oh my gosh, in 2 years half of white collar people will be out of work, or that kind of thing as ways of not understanding human organizations and what humans can bring in and all the rest. I do think there will be a lot of workforce transition, but what I actually think is that a lot of people's jobs will get replaced by people using and managing groups of AI, and so that transition is difficult, but it's not like AIs are doing everything three years from now.

Yeah, there might be a three-year thing where that happens, but my hazard bet will be that we will actually have decades still of humans fitting in because if I see the pattern of where we're going, we already have super intelligence. There's already things that ChatGPT, Claude, Copilot, Gemini can all do that we can't do - speed, depth of access to the internet, kind of exhaustiveness of research, a bunch of different things including working on certain problems. For example, I haven't coded for decades and I can create interesting things with code now given AI. That's all kind of superpowers. On the other hand, what you discover is your ability to have context awareness, to understand when things are going off the rails or what real true excellence looks like, is still substantially better, and I think we'll likely be there for a while because while these things have this jagged edge of super intelligence capability, they also have weirdnesses. Even the super intelligent great models today still screw up on various kind of prime number problems.

And part of that screw up is that you ask it a simple prime number problem and it gives you an answer and you say that's wrong. Give me another one and it goes, oh sorry, yeah, here's the answer. It'll keep doing that 20 times, 100 times. Whereas human beings, and this is part of what I mean by context awareness, the third time I give you the wrong answer, I go, look, I know I don't have this and I stop. I know I need to change my game. And the models currently haven't really demonstrated any non-brittle, clunky way of doing that. And that's just one instance where I think human capabilities may add in for quite a long time and maybe for a very long time.

Well, as I think you know, it's even more relevant although the pattern changes a little bit. So a lot of blitz scaling was lessons from Silicon Valley about why within a 30 mile radius of the peninsula of Silicon Valley is over half of the NASDAQ market cap. And one of the things that I was told recently that I thought was interesting was that 91% of the non-Chinese AI market cap is in Silicon Valley. And actually if you remove Anthropic and OpenAI, that only drops to 70%. And so it's a huge amount of what's being created there. And blitzscaling was the lessons by which these technology companies of the future are built so that as many places in the world can learn, adapt, participate. I want to help the creation of Silicon Valley. Now Silicon Valley is everywhere. So blitzscaling was in part one of the things that was most uncomfortable was once you discover scale product market fit, how does it go to the moon and to Mars and beyond, like Toy Story to infinity and beyond. How does it go that way? And the answer was by really scaling the organization and taking a whole bunch of risks in product market fit, in scale of hiring and business operations and technology development and so forth in order to realize that scale opportunity at speed before others. Now how blitz scaling changes in the era of AI is that it goes from pure scaling of organization, by the way it will still have some scaling of organization, to the deployment of AI doing and redoing work at intensity. Because one of the many different things that AI brings to changing the game for how you build scale technology is you might build something for a couple weeks and then throw it away and build something new, and that might be your way of doing it. Whereas previously, you really were trying to build on top of each new thing and then refactoring it. And you'll still be doing that with AI, but you'll be building at faster, higher uncertainty, higher risk, trying things, throwing it away, going again. And you'll be still scaling a human organization. Obviously, the number of agents that you'll be scaling and how you'll be operating will be a multiple of human beings because each human being will be managing teams potentially up to thousands or even tens of thousands of agents.

Well, generalizing the last 10 years, I do think the two places that get the scale technology story most correct are Silicon Valley and China. I'm trying to help Europe, Japan, other places build us out in various ways because I think it's better for the whole world and better for those regions. But basically Silicon Valley continues to have the network, the information, the learning, the venture capital. It attracts talent from everywhere in terms of building these software technology scaleups and obviously that bleeds into other things. So there's also biology and other kinds of things happening. But I think one of the things that's funny is if you look at the last 20 years, there's always a new article about so-called peak Silicon Valley and it's always proven to be wrong 3 to 5 years later. Like, oh, housing's too expensive or the taxation thing has killed it or it's too woke and we're all moving to Miami, and then within a couple years it's like, nope, Silicon Valley has continued to massively grow, and Miami is still basically just Cryptoville.

Well, the normal thought will be can we get an exclusive partnership or access by which we leverage that in return, and I don't think any of these AI models are really going to provide that. And what's more, partially because a lot of startups start with something that's thin because that's part of the reason why software startups are so good. Look at the technology we started LinkedIn with, it was very thin. It's gotten deeper over time, but it's still mostly a network effects business. So I think that the notion of partnering with them will broadly be beyond the scope and irrational for most startups. But partnering is different than building upon. It's a little bit like you might build upon the mobile app stores of iOS and Android and not partner with them, but still build something that's pretty effective. And so similarly in terms of these models, you want to do what is classic around the parallel of cloud services, which is you might start intensively with Azure or AWS or G-Cloud, but you're ultimately going to want to have some diversity across them so that the pricing and dependency and your ability to do that. You'll want the same thing in terms of models ultimately, not necessarily at the beginning, but that'll be part of your strategy. And then what you're going to be looking for is how do you build something that once it's built is very valuable that the models can't just go, oh, we added that as a feature and your whole business goes away. There are various techniques to try to do that, some of them are network effects, some of them are enterprise integration, but you're going to want to be doing that. So building on top of them can make a lot of sense, but you're going to have to build a substantive business that isn't just a wrapper on the model.

So, I think it's a good thing for tech. It's part of the Schumpeter creative destruction, which is it isn't that once you've innovated or created a business, you're good for all time. It's part of the S&P churn rate which might be much quicker in this accelerated universe, and it's a thing to be smart about. Entrepreneurs, 90 plus percent of entrepreneurial businesses fail without really getting off the ground, and that's because you have a bad theory of the game, you're unlucky in your market timing, you're unlucky in your competition, all kinds of different things. You can't pull together the resources to execute against it. And that's fine. For example, I saw this back in the day at GPT-3. Hey, I'm building a copywriting ad business on top of it. And you're like, well, but its writing capability is very good. What are you doing beyond just saying I'm giving you a front end to GPT-3? And so when GPT-4 came out, the whole business got disrupted. You have to be having an entrepreneurial theory of the game. That's why it is that the AI companies, the hyperscalers and AI companies themselves doing it, or other people just go around you and use it directly. For example, it's like, well, I built a great app with cloud code and now everyone's going to buy my cloud code app. Well, could they just point cloud code at it and say build me that app? And the answer is yes, in which case you're like, okay, so why would they pay you a lot of money for it because the token cost of cloud code or Codex doing that is not going to be that high. So you have to have that more detailed theory. But by the way, I have continued to invest in various AI companies along with Greylock and others, and so there are smart people doing this, but you have to have that thought on the go to market, that thought on the predictability of it, the thought on network effects, the thought on enterprise integration, the thought on what kind of loops like data loops and other things are really something that creates unique value. Then if you're one of the AI companies that say I really need to get into this, they go, hey, we're going to come and offer to buy you for a huge premium because that's how we get into the market, and that's how we compete with our competitors. And that's what you have to be having as an entrepreneur.

I do. And I think part of the thing that AI causes is it changes a landscape in which skills really matter. So for example, when you're doing graphic design, it used to be, oh, I have a lot of hand-eye coordination to be able to draw very well. And it's like, well, AI can draw very well. That hand-eye coordination doesn't need to go down. But your visual thinking and what your bar is for what is really good goes up. So it changes the landscape of which skills are really important. The landscape is now like, oh, it's super important to have the taste, to have the judgment about what excellence looks like, and not just in visual but obviously in code and writing and then the hand drawing skill goes down. So if you were just a hand drawing skill person, you're going to have to figure out how to bring that into AI and have AI amplify, and say, okay, I take these skills on and I still add something on top of AI. But like the taste, the orchestration, the judgment is another one in addition to context awareness that I think will persist for decades.

Well, one of the things to understand about the internet is that offering products through the internet, the other product is only a URL, a click away, an app download, etc. So that can be a small advantage that they found you versus others, but the other thing is discoverability. So that's part of the reason why you want to have a very good sense of what a great product is for that problem solution. It's kind of like one, you have to ask the right questions. Two, you have to have a sense of what real quality looks like. And by the way, one of my aphorisms that I'm famous for is if you're not embarrassed by your first product release, you've released too late. That is still correct because it's the trajectory towards great. It doesn't necessarily mean you wait for four years to have great coming out. It's like no, no, no. You want to be learning and engaging to get scale product market fit and which things really work and not. So you want to have a fast learning curve towards that. And then of course, one of the perennials for entrepreneurship is your go to market, your distribution. You have to have a differential theory of that. One of the most common failings of quality, tasteful knowing what the right problem is and the right great solution is, is understanding that the go to market is as important as your product in order to succeed.

Distribution is great and it's extremely important, and it's one of the reasons why by the way in existing companies it's not necessarily game over for them. If they don't reinvent themselves, then the dinosaur decay has started. And this is actually part of what has gotten the SaaS apocalypse wrong, because it's which SaaS companies are embracing and doing AI will be most likely part of the future, at least on a risk coefficient, and those that are not are almost certainly not part of the future on an accelerating decay. And it's the similar thing in terms of distribution, which is you say, 'Hey, I've got distribution that allows me to use AI and to amplify and to do it.' If you kind of go, 'Oh, I'm not going to use any AI,' then my guess is your distribution may persist for a while, but you will start arcing down some. And that's part of the reason why the AI reinvention, the unequivocal thing that I say to all people is start playing with it, start using it and understand how it helps you get to better upside. Not just better cost savings, that's interesting too, but better upside. And that is key. And distribution helps you get into that loop much faster.

Here's the thing that's a little strange. AI has a negative PR perception in the US and Europe and a positive one everywhere else, including Asia and so forth. And one of my greatest worries for the US and Europe is to say, 'No, it's really important we slow this down.' I mean, this gets to the idiocies in the US of like, 'Let's ban data centers.' And it's like, look, there's nothing that says any particular region should adopt a data center if they don't want it. And I very much get that the local community should ask for good things that help the local communities, economic upside and jobs and other kinds of things, and also noise pollution and other things that are important to get right. But that's part of the future. And in Asia, you go to Japan, you go to China, you go other places, they're like, 'How are we part of the future?' But not just Asia, you go to Central and South America, you go to Africa, you go to the Middle East, they're all like, 'Oh my god, this is the future. We want to be on it.' And so you've got this kind of like maybe it's the reverse revenge of the industrial revolution. Part of what made the US and Europe so strong globally was the industrial revolution and embracing it early. And maybe the rest of the world is going to embrace AI and the cognitive industrial revolution earlier, and then you'll have a pattern of modern reverse colonization from the US and Europe from everywhere else adopting AI much more fervently. So there's a tale of two regions in AI perception. Part of the reason I state this bluntly is of course I want the US and Europe to embrace and to be a leader in the cognitive industrial revolution. I think it's great societies, great sets of social and societal values, and we want those to be at the leading table in the world if not leading the table. And so it's a huge moment for this kind of wakeup call. You would think, especially in the US, if they're leading the turbocharge and the storytelling and the big picture vision and making AI the most amazing technology, they would be very positive about it. Europe, where I'm originally from, I'm originally from the UK. Sorry to the listeners listening to my British voice every single day. But they're having a few more positive signals around AI and what it can actually do.

The ultimate thing is to engage and see value. Part of the reason why people have negative is because they say, 'Oh, the tech companies are doing this to me. I'm going to lose my job and AI is going to be doing the job.' And that's the entire story. And by the way, if that's the entire story, you don't tend to be broad-minded and say, 'Hey, my society and my community needs to work.' Most people tend to go, 'Well, if it doesn't work for me, then it doesn't work for anyone else.' Obviously, there's a certain amount of narcissism and egoism in that perspective that is 99% of the human race. So I understand. But the question is, how do you get to the 99%? Show me benefit. Show me how I can be amplified in using it. Show me that even if it's a job transition, which I don't really want to sign up for, it feels like a reduction in

Agency. This is part of the reason I wrote the book Super Agency because it's like how you can gain agency and lean into it and do that. However, the big tech companies are deploying their AI agents. But to you say, okay, even if I don't want to be doing this job and am now being forced into it and doing it, what are the benefits I get? Well, one benefit is obviously a huge exoskeleton of the mind and the capabilities, which is great, but also part of the reason why I would like to see it deployed with speed and vigor is like a medical assistant that runs for free on every smartphone because you go, 'Oh, I'm doing this job transition.' But unlike where it used to be only the 0.1% or 0.01% of humanity could have a doctor on call 24/7 to answer something for their kids, their parents, their partner, their pets, etc., that exists now and to deploy that effectively. And by the way, the same thing for a legal assistant in terms of, you know, most people, they sign a lease, they don't know how to read the lease and deal with it. AI agents can do that, can help with that today. You know, educational tutors and helping do jobs, all the rest of this stuff. So you go, okay, I get it. I can gain agency. AI can be amplification intelligence. AI can be for me. And I think that's the way you persuade people. And it isn't by a story of words. It isn't by a hey, just trust me and I'm trustworthy, so you should just follow me. It's by getting them to engage and getting them to use it and seeing value for themselves. And that's part of the reason why it's like engage, engage, engage, engage is the short answer. And by the way, when you're a leader in a company or a society for doing this, it's get people to engage. They see the value.

Yes. Well, and also part of how you engage a thousand percent and part of how you get trust is by seeing value. Obviously, we live in very chaotic times. People tend to trust institutions, leaders a lot less, you know, across not just the political spectrum, but also the corporate spectrum and institutions even, you know, you see nutty groups of people not trusting doctors and so forth, which is insane, you know, like the antivax stuff and all the rest. But part of how trust is built is by this has been helpful. This has built useful things into my life. A medical assistant helping me with my work. And that's the kind of thing where people start going, 'Oh, now trust is being rebuilt.'