How Do You Run a Company When You Can’t See Six Months Ahead? - with Axios CEO Jim VandeHei

I believe every one of us will have Einstein-level genius across any topic we could possibly give a about, probably in a three-year time period. In a two-to-five-year time period, I think the open web will largely collapse and that will move most information content to social platforms and LLMs, most of them highly personalized. I believe that anything that is verifiable, repeatable, identifiable will be automated once everybody has Einstein-level intelligence. People are not going to be satisfied with sameness. Humans hate having the same thing. I'm Jim Vandehei, co-founder and CEO of Axios. I'm jazzed to talk about how AI has transformed my job as CEO, has transformed the company Axios that I run, and has transformed the country that all of us live in and love.

Well, I wrote it because I did turn myself into an AI lab rat, and I think people are deeply curious, both our readers and my own team, on kind of like, what are we doing? Why are we doing it? Luckily, I think I was early to figuring out, okay, this was going to be a big deal, right? We cover business, we cover media, we cover tech and politics. I got to know, you know, spend some time with the Sams and the Darios of the world early on. You could tell, okay, this is going to be something bigger than your traditional technology. And what I realized early is, okay, this could be like the first technology that speaks Jim, meaning I don't know anything about coding. I don't know languages. I don't know how to do it. I'm like a self-described tech dope. But it felt like, okay, this one's going to work based on language. So it really was going to be like, how creative could you be? Like, are you conceptual? Do you think in systems? Are you kind of persistent in trying to solve a problem? If so, this technology will probably work. So I decided, okay, I want to do that. I'm going to throw myself into this. And I knew that in media, I just didn't want us to be a victim of another technological wave, kind of like the Washington Post was when I was there. So I'm like, I want us to be at the forefront of using it. If we as a company are going to be at the forefront of using it, I as the CEO need to lead by example. And so I threw myself into it, spent a lot of time reading about it, thinking about it, talking to people, utilizing it. So by the time I wrote the column, I was hearing from readers who were deeply curious about it. And I felt like, okay, I'm running a media company. I'm also a reporter and a writer. So why don't I put those two together to explain what I'm doing and what I'm seeing? Because there was and still is a huge gap between people who kind of passively see the technology and they're somewhat unimpressed with it, and I think those of us who spend a lot of time utilizing it and we kind of see day-to-day the magic of it and the possibility of it. And I wanted to narrow that gap. And the reaction was amazing, like just a lot of feedback. It turned out, I think the most surprising thing was, and you probably experienced this with your show, there's a lot of CEOs who I would have assumed were more AI savvy than they actually were, which now in retrospect makes sense because they're just running their company, right? They've always had a CTO. They got the job because they were great at sales, great in operations, great at something specific. And this is a kind of a new skill set. And so I was surprised by how many people I heard from. And in fact, after writing that AI lab rat column, I ended up starting a newsletter for CEOs by me, a CEO, that largely focuses on how do you utilize this technology and think about this technology in a workplace that's changing every six or seven weeks. And that has been a really good product for us. It turns out to be a pretty good business for us. But lots and lots of CEOs, literally thousands on it.

Oh, it's the difference between success and failure, right? The fact that like I threw myself in, number one, I realized what it can and cannot do. Number two, I realized you just got to try, right? So then after I threw myself into it, I basically gathered our staff and I said, I know a lot of you are scared, a lot of you are worried. I said, this is like almost two years ago, I said from this day forward, I don't give a And I'm not saying that to be cold. I'm saying it that I'm going to alleviate that fear by giving every single person at this company access to OpenAI, to ChatGPT. We'll pay for it. I'm going to bring people in to teach you how to utilize it. You don't need to be an AI savant, but you do need to figure out how to utilize this technology for the specific work that you do. If it turns out you have a natural instinct or interest in utilizing it and you kind of want to be a lab rat yourself for other people at the company who do the work that you do, raise your hand. If it turns out you raised your hand and you're good at it, I'm going to have you teach other people who do the same thing at the company how to use the technology so they can be proficient in it. And that was maybe one of the smartest things we've done since our founding because suddenly people went from nervous to curious and they went from resistant to enthusiastic. So now everyone was experimenting with it, and I'm sure you guys have seen this with other companies. It turns out I don't think everyone's great at AI. I think it's about one to two percent of people are really getting a shitload of value from it. But it turns out there's a lot of people in that one to two percent who are not technologists who discovered that their brain works in a way that is super-duper compatible with AI. And they tend to be people who are problem solvers. They tend to have structured thinking, either as one of their skills or as one of their unearthed skills. And they tend to have some persistence or stubbornness to them in trying to solve something. And so now we have multiple people at Axios who have zero technological background, no training, no coding, who are now part of our AI enablement team that go across the company and work with people to figure out what can be automated quite easily using AI, and then they help them automate it. And so it's been magical. So what would have been something that could have displaced jobs created new jobs.

Oh, I don't think that that's true at all. I think that, I know what that person is saying. I think it's very hard to teach someone to be almost superhuman in using AI. So someone who might be getting a 10 or 100x return on the amount of time that they're putting into the technology, I tend to be in that camp. I think it's somewhere around one to 1.5 percent of people truly can extract a ton of value. However, anybody who's introduced to the technology and then is shown like how you might be able to utilize this to do the very specific job you do can learn to use it at a very basic, somewhat passive but basic level, much like using a computer or using a phone or using a calculator, and get real value out of it. By the way, and that's all I ask of our employees. Like, I'm not saying you have to be an AI optimist. I don't think you have to be an AI anything. I'm just saying the minute that we see something that the technology could do at human efficacy or better, and we've made the corporate decision that we're going to do it, you just need to do it, and then use your free time to do stuff that you can do uniquely. And when you do that, it creates an energy in and of itself, and we're able to do a lot more. And I guarantee you, you're going to be a happier person, and that's been the end result. But I understand what that person's probably getting at. What do you think is the current state of the media landscape when it comes to using AI? And I guess both in terms of like how they are actually getting most use out of it, but also maybe just what's the mood tone amongst your peers about the space.

It's such a weird twilight zone time right now because I think there's, and I spend a lot of time because we, like, the president, members of Congress read us, right? So I'm trying to explain the technology to them. A lot of them are tech dopes. They don't really understand the technology at all. And then there's people who are sitting on X or living in Silicon Valley who are just marinating in this stuff all day, every day, every weekend, and they just know it so deeply. They know the new models. They know the difference between Astra and Opus. And they could, there just is like this fluency. And so there's this massive, massive gap, right? And so you have on the one hand, you have these people who are like, 'Oh my god, this could end the earth in less than a decade.' And on the other hand, you have like 90 percent of people who are like, 'I don't know, seems like a pretty good search engine to me. Like the other day I put something in it and it hallucinated. I don't know what you guys are all hopped up about.' Well, both things are true at once, as weird as that sounds, but they are true. Like it is only somewhat impressive if you're not super fluent in it, but it is wildly impressive if you put your head in the damn machine and stare at it all day and compare it to the benchmarks of where we thought we would be at this point in history. So I understand why both are happening. I worry a lot about this period where there's this massive gap because in the middle of that gap, I think that's why you have so much political backlash, you have a lot of corporate confusion, and you have a lot of fear, fear about what could be, because if people aren't using it, like think about data centers, like why would 70 percent of people have negative opinions of data centers when 99.5 percent of people don't live anywhere near one? Like why would you give a right? Well, I can tell you why you'd give a Because all a data center is is a stand-in proxy for AI. And all you've heard about AI is it might take your job or it might destroy humanity. Both of those seem unpleasant. Therefore, I have a negative view of it. And that's the kind of, we're just in this weird place. And I think we're going to be in this weird place for another maybe a couple of years, but certainly for the next year. It's just, I think things are going to get weirder, probably a little bit scarier than they are right now. And I don't know that the public's ready for that. I think that's why you're seeing this political backlash.

I think most media is useless for understanding AI. I get 98 percent of my knowledge out of AI, outside of like me being on the phone talking to people, from following smart people, researchers, data scientists, people inside the company on X, listening to very specific podcasts, but understanding the biases of those podcasts. Like I find Moonshot to be really useful, even though I think they're delusional about the technology and they've gone way overboard in the singularity. Useful but delusional is a perfect band, right? But I always, they always say something that, oh, I didn't know that, or they're obsessing about some obscure thing that like, oh, maybe I should obsess about that. Another good example is All-In. Like I actually, as annoying as the podcast is sometimes, I find them to be terribly useful because they're all either operators or heavy investors. So they're seeing in real time how is this technology working. And then you have obviously David Sacks is wired into Trump on the policy side. So I'm sifting out all of the bias and trying to input all of the interesting things that I might not have known or things that I should be exploring, but all of that information consumption is not happening in a traditional way.

A little bit, but I don't, and obviously we do a lot of reporting ourselves, and I do a lot of firsthand talking to these folks. So I get a lot of my information that way. I do have like a pretty sophisticated setup where I basically scan, I've pre-ordained about 200 different sources and people that I'm doing a constant scan of what's happening globally on tech with a heavy emphasis on research and academic papers that I otherwise wouldn't see. And I get that summary every morning emailed to me at 4:00 a.m. And that's useful. It just points you, useful not in that like I've taught it to write exactly the way I do in Smart Brevity, and so it's very efficient, but where it's really useful, it's just flagging things I wouldn't otherwise see. And so there's just this information ecosystem around technology that's different than politics. Like I could just read the New York Times and the Wall Street Journal and I'd be fine understanding politics. If I just read the New York Times and the Wall Street Journal, I wouldn't know, I'd know very little about what's actually happening in AI. So it's just a different world. And actually I've been writing a lot about this for CEOs, trying to explain to them like the FT is fine, the Journal's fine, but if you really want to understand, listen to these podcasts, follow these people on X, read these Substacks so that you can really understand what's happening. Because what's interesting, as you guys know because you live in this world, is if you follow them, you're kind of two or three weeks ahead of the public. All of these debates, all of these weird things that trickle up into the kind of, now it's public and people are debating it. You see that stuff happening and playing out on X all the time. So you kind of understand it before other people do. And I find that cool.

Yeah, I would give a little caveat to the way that you describe that. I would say I agree somewhat with Jensen. I think that the limitation is imagination, with an understanding that there is a reality that no matter how good your imagination is, taking this technology and plugging it into existing data pools and existing processes and existing security dimensions in any company that existed pre-AI is hard. There is actual work you need to do. It is not plug-and-play by any stretch of the imagination. That said, I agree that this is a great time to be imaginative. And in terms of us, like the way that I try to talk to my staff about it is I have made some assumptions about what the world's going to look like in three years, right? And those assumptions include, I believe every one of us will have Einstein-level genius across any topic we could possibly give a about, probably in a three-year time period. In a two-to-five-year time period, I think the open web will largely collapse and that will move most information content to social platforms and LLMs, most of them highly personalized. I believe that anything that is verifiable, repeatable, identifiable will be automated. And so I've now thought, okay, if those things are true, how do you create a media company that not just survives but really flourishes in that environment? Well, the assumptions that we have made that I think are accurate are once everybody has Einstein-level intelligence, people are not going to be satisfied with sameness. Humans hate having the same thing. It goes down to our hunter-gatherer instincts. Like yes, we want to survive, but then we want what other people don't have. So that if a content company can provide things the machine can't, there'll be real value to that. I think actually more value to that. And what a machine can't provide is somebody who has subject matter expertise who gets people to tell them something they don't want others to know, or someone who could have a conversation with you where maybe you're nodding your head but I'm looking at your facial expression. I've dealt with you long enough to know that you're telling me something different than what you're actually saying, or that I've been covering an issue long enough and the people long enough that I can capture the nuance in a way that a machine never can. And so we're putting a huge emphasis on, okay, let's automate all the things that we can automate. Let's figure out how do we get more high-quality intelligence across topics that any smart professional would care about. And then how do we set ourselves up to make sure that we can stay one step ahead of where we think revenue will go. One of the bets that we're making is that not just the big LLM companies like Anthropic and OpenAI, but that individual companies and ultimately individuals will have their own LLMs and that you will pay a premium for certain types of content. Maybe it's news to do your job. Maybe it's a comedian, or maybe it's a type of content that you just like for a passion that I think there'll be real value there. Well, that then means that I might need to re-architect my content to make it really easy to ingest into an LLM. We're already seeing demand from that. So I do think that that is going to be correct. So that's one part.

Yeah, RAGging and using. So basically not necessarily training, but imagine that you're Goldman Sachs and you have your own LLM. You have your own business imperatives. You're providing financial analysis, financial advice, making business decisions. You rely on Axios content because we have people who understand the Fed or understand banking or understand some dimension that you care about. Our theory and what we're already seeing is you will pay a premium to have a clean, LLM-friendly feed, perhaps that you get even seconds before other people get, that goes directly into your LLM, and that that will be a new way for us as a content creator to extract value from the content that we're creating. So that's on the national side. One of the problems I've always wanted to try to help solve is local news. I think that people being connected to their community and understanding what happens matters. It's probably the hardest media problem to solve. And I am now certain that we can solve it using artificial intelligence. Why? Because now that you don't need a newspaper, you don't need print, you don't even need a building because post-COVID people are fine not working in an office. And now you can use artificial intelligence to do all the back-end work. All of your content distribution, your marketing, a lot of your sales prospecting, all of that that took people and took money can now be automated so that every dollar of investment goes into the journalism. And you want to use artificial intelligence to not do the journalism, but to support the journalism and then distribute the journalism. So sometime later this year, we're going to unveil what we call the local super system. And I think people will be pretty impressed at what we've used technology to build to automate most of the process other than the actual reporting and the writing of that reporting. And I think it's going to allow us to go into communities big and small with a business model that would work, that is durable and scalable. So that's a lot. Yeah.

Yeah. Well, one, thank you for the compliment. So a couple of things. Like one, you know, trying, I obviously on the technology team having people who are super AI savvy who have turned the coding and stuff like that already over to artificial intelligence, which does it better than people. Another thing that's worked well for me that might work for other CEOs. I went out to hire one AI fellow to work for me just to do projects outside of our existing system. I ended up hiring three. And what I looked for were people graduating from college who were super AI sophisticated but had no technological training. The reason that I looked for those attributes is I wanted someone that could prove to me that they were entrepreneurial in mind and spirit and that they could self-teach. They self-taught themselves how to utilize a technology. So two of them are now working for us on projects full-time. One of them is just like a total savant who just sits there and takes my ideas and together we start to prototype things outside of our system. So we're not distracting people, we're not slowing it down, but then it gives me this ability to come up with new products, have them prototyped, even test them before I have to bring anybody else in the company into the loop.

And I would recommend, and I also have that person who knows what's happening all day, every day, so that they can brief me if I don't have time to figure out exactly, I want to know the intricacies of this Hugging Face hack and I want to understand like each step. I don't want to do that work myself. I don't have time. I got to run the company. But they're thinking about this all the time so that they can brief me. So here I am running the company, but I'm also understanding how the technology is working. I'm understanding kind of the weaknesses or the security things that we might need to know. And so it allows me to both build and to understand. And because I'm building and I understand, then it makes it easier for me to teach. Then I could teach my team, I could teach the rest of the company. And they all know that I come at it as somebody who doesn't understand technology, which I think then liberates them to feel like, well, if Jim can do it, maybe I can do it at some small scale. So I think that has helped. And then the last thing on that is we do have, anytime any individual at the company that I hear has built something that solves a problem, I instantly email them and say, let's do a Zoom, and I have them walk me through it. And several of those have now been productized. And I do that one because I'm actually interested, but two because I'm trying to incentivize people that, holy, if I come up with a solution, I could get on a call today and maybe that's put into the company's like DNA and I'll probably be recognized for it. So all those things happening together now, they create like this positive energy utilizing technology.

Okay, now we're getting into the weeds. So, but I love this, but I'll go there. So right now, we don't know a ton about exactly like how to write content so that it matches the mind of an LLM. What we have discovered is that the way that we write at Axios in Smart Brevity style, which is what is the most important thing, why does it matter? Then create a hierarchy of the most important points in order and do it by bullet points shows up disproportionately well inside Gemini, Claude, and OpenAI. The best that we can tell from third parties is that probably the architecture of our content does match the way that an LLM probably wants to ingest information. So right now we think that the standard format actually works really well for LLMs. We are starting to work with companies to convert their content into Smart Brevity to see if they will show up better within LLMs. Interesting. I do think there's probably a future where you're writing a story for humans and in the background that content is being repackaged in a way that is ideally and exquisitely balanced for the way that LLMs want to ingest information. I think that can happen in the background and then it can be connected into those LLMs. You don't need people necessarily doing that and I don't think it's that hard. My personal view is that I don't think machines for most information work that different than people. I think most people want their information stacked in hierarchy. Most people want it delivered efficiently. And so I think the two for a lot of information will converge. But you could see sort of, you're at The Atlantic, you have this media article. There's a lot of like real news and information in it, but you want to write it beautifully and you consider it like a work of literature and of writing that you might give to humans. But in the background you might have a version that is just stacking the most interesting new facts in order, written in a much more digestible way.

Ethan Mollick is the world expert on this subject. Please, everything that you've read on this page in a certain way.