The Atlantic’s Josh Tyrangiel sits down with host Jeff Berman to dig into his reporting on how Elon Musk and Jeff Bezos are trying to vertically integrate their empires beyond Earth, and what it means for the rest of us. They also discuss Tyrangiel’s book “AI for Good” on how real people are already using AI to solve problems and save lives.
About Josh
- 12-time Emmy and Peabody Award winner
- Executive producer of Netflix’s InstaDocs
- Former Editor of Bloomberg Businessweek
- Chief Content Officer who reinvented Bloomberg Media
- First AI columnist at The Washington Post; author of AI for Good
Table of Contents:
Transcript:
The new space race: Musk vs Bezos
Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.
JOSH TYRANGIEL: We are in this bizarre situation where we have essentially supranational actors. It is even more bizarre because we actually really kind of need Jeff Bezos to be successful to counteract Elon Musk. But neither is Lex Luthor, right? If we can depoliticize this, if we can get it away from everybody thinking each AI product is Sam and Dario arguing over the fate of the world, there is a lot of ground to be gained, and it comes back to competence. If you just demonstrate to people that something works, we’re in an environment where people are so starved for effectiveness that they don’t care if it’s AI.
[THEME MUSIC]
BERMAN: This is Masters of Scale. I’m Jeff Berman, your host. In today’s episode, Atlantic staff writer Josh Tyrangiel joins me to discuss his brilliant book that, candidly, I tore through in one sitting this summer, AI for Good, and his latest reporting on the modern space race between Elon Musk and Jeff Bezos. This conversation took place live onstage as part of the Masters of Scale twenty twenty-six summit. I hope you enjoy it.
Josh, it’s been a busy week of news. I want to get to the book and some of the stories in the book, but you’ve just written a cover story for The Atlantic about the one billionaire and one trillionaire engaged in a space race, one of whom, it seems, deliberately slowed down communications technology to hurt Ukraine and advantage Russia. I’m curious how you understand the reporting on that and what you understand about the motives behind it.
TYRANGIEL: Yeah. I just wrote a cover story basically about the two richest men on Earth and their attempts, in competition, to vertically integrate the galaxy, OK? I’ll explain what that means. Come on. You guys wanted to do it, too.
It’s a four-step process. It begins with rocket launch capability. Elon Musk obviously has SpaceX. SpaceX is responsible for 80 percent of the mass launched into orbit in twenty twenty-five. The next closest competitor is the government of China, and it’s not close. Jeff Bezos has Blue Origin. Launch is actually a bad business, not a very good business. The second step is satellites, largely government surveillance and high-speed internet, which is a really good business. Obviously, Elon has Starlink. Bezos is trying to create his own. The next two are the theoretical levels: data centers in space and on the moon. It sounds ridiculous. Both think they’re five years away. Where it all comes back is unfettered compute to Earth-based AI models that they control.
It all sounds a little science fiction-y. It is not. With just the first two rungs, what Elon has been able to do is control the outcomes of things like conflicts between nations. As I was writing the story, I spoke to Jake Sullivan, who is a former national security adviser, and we were talking it through. He said, “We are in this bizarre situation where we have essentially supranational actors.”
He also said, “It’s even more bizarre because we actually really kind of need Jeff Bezos to be successful to counteract Elon Musk.” Then he turned to me and said, “But neither is Lex Luthor, right?” The reporting we have is that one of two people absolutely gave an order to shut down the high-speed, defense-capable internet during this attack by the Russians. Whether it was Musk at the behest of Trump, or Trump ordering Musk at the behest of Musk, is a little unclear. But we are existing in a world where individuals have the powers of nation-states, and that is a relatively clean segue to AI.
BERMAN: Gonna be an easy 20 minutes, pal.
TYRANGIEL: Yeah, we’re going to fly through it.
Copy LinkWhy Europe lacks a real Starlink alternative
BERMAN: There was some follow-on reporting that some EU nations and Canada are exploring alternatives. Is that even viable? Is that an option?
TYRANGIEL: The reason the four stages of this vertical plan are so important is: How are you getting your satellites to space? There are 11,000 Starlink satellites in orbit right now. By far, the majority of things orbiting have been launched and are controlled by Elon Musk. Currently, both SpaceX and Blue Origin launch satellites for other satellite providers, and there are many. But in just the past couple of months, there’s been a lot of reporting about the fact that those external companies are beginning to get their dates on Starship pushed further and further out, and availability is going down because there’s more captive demand from Elon’s own companies. We do not have diversification for space launch. So the notion that suddenly Europe is going to jump in, not at scale. That is not happening in the short term, and so there needs to be some sort of government response. Again, good segue to AI, but that’s the universe we’re in.
BERMAN: This is the argument for vertical integration, right? If I control the whole thing, I’ve got a moat. I can defend it.
TYRANGIEL: Yeah. In my conversations, Mark Kelly is a good example.
BERMAN: Senator from Arizona.
TYRANGIEL: Yeah, and former space shuttle pilot and frequent combatant with Musk.
He says point-blank, “We can’t begrudge them the success. They got there fair and square. They’ve invested for 25 years. They’ve blown things up. They’ve tried again and again.” But government does have a role in ultimately deciding, if you have captive demand and you have concentration of launch, what do we do about that? There are mechanisms that government can use. We’re not using them currently. But I think the last couple of years, and particularly this last incident, have drawn attention in a way that is impossible to ignore.
Copy LinkWhat government can do about concentrated power
BERMAN: So Josh, Mark Kelly is one of many who are reportedly considering running for president. Imagine he or one of the others calls you, and I know you’re going to say they should call someone else, but they call you and say, “Well, OK, I understand this. I’ve read your story. I get what you’re saying. What do we do about it?” What are you telling them?
TYRANGIEL: Yeah, look, it’s a super complicated answer because, on the one hand, the reason they have this ability is that the United States, in the early 2000s, created commercial crew. Even the most ferocious critics of Musk and SpaceX will tell you, “This is great.” The reason we’re making so much progress in space is that we allowed the private sector in to collaborate in a different way.
But we do have mechanisms. You can talk about antitrust. When you control launch and you control the satellite market, think back to the days when we did not allow car makers to own dealerships or movie makers to own theaters. This is elementary stuff. The challenge is that we are in an era when institutions, federal institutions, are not interested in that level of enforcement. So the political opportunity is communicating very clearly to the American people why government exists, what actions it can take, and why it’s in the national interest to have control, at a certain level, over things that are going to impact our national security.
BERMAN: OK, but a president takes office who is interested in these things. Their Justice Department files a lawsuit. It goes to a court. There is a decision. Maybe it’s adverse to the company.
TYRANGIEL: Right.
BERMAN: But eventually that’s going to go up to a Supreme Court.
TYRANGIEL: Yes.
BERMAN: And I’m not putting my money on the result that I might want to see there.
TYRANGIEL: For sure. Look, invoking the Defense Production Act, all the things you might do, all of a sudden you’re playing politics, right?
Long term, we need diversification. We need more launch capability. But if you continue to see an individual actor basically step in and say, “Your policy to Ukraine is what? No, I don’t think so,” you will have to get down and dirty and figure out what you’re going to do more immediately, and it’s not going to be easy.
Copy LinkSeparating AI technology from tech company hype
BERMAN: So why is this a good segue to the AI conversation?
TYRANGIEL: Okay, so I wrote a book called AI for Good. I started writing it a while ago, and when it was published, it turned out something in the public mood had changed a little bit, and people were like, “What good? What are you talking about?” What I encountered, really, was that the same impulse a lot of people have toward Elon and this incident is similar to the impulse they have toward the handful of people pursuing AGI, which is, “Hold on a second. I didn’t ask for this. You don’t communicate well about how it’s gonna benefit me.”
You’re entering in a context in which there’s very little trust in institutions and corporations. Some of the corporations bringing me this product are the same ones that brought me social media, which seemed great for a little while and has sapped all of the reservoirs of public trust. So we’re entering this world where there is no good faith to rely on. I’m not a Cassandra about that. I am very aware of it.
I think, to tap into our theme, we have agency in relation to AI in two things. One is the AI we don’t want, right? What don’t we want from AI? I’m pretty sure, having watched the debate, that most people don’t want agentic AI that takes their data and gives it to Mark Zuckerberg, right? We probably do want regulations around AI with a mandatory duty to report misalignment, right? We want there to be a regime. We don’t want AI that runs amok.
It is equally important to assert what we want AI to do. What should it do? Let’s create the market for it. Let’s encourage the kind of AI products we want. So The Washington Post called me and asked me to write an AI column right after GPT came out. My friend said, “Look, it’s a hellscape. We don’t know what we’re doing here. Our readers are furious.”
So I spent a couple of weeks doing the thing you do, which is you come out to the Valley. At that point, I think I sent a note to Sam Altman, and we had a two-hour conversation a day later. It was very easy to access these guys. For six or eight weeks, I had these conversations, and I was like, “Wow, this is interesting.” They’re definitely all talking about money. They will tell you off the record how much they hate each other. They will say, “Well, I can’t tell you my valuation.” I was like, “Wrong guy. Don’t care.”
After about eight weeks, I ended up calling a guy named Danny Hillis, who I’m sure is a familiar name to many of you. He’s one of the great men of computer science and really created cloud computing, if you’re gonna boil it down. I’m sort of venting to Danny about, “This is a financial story, and it’s a rivalry story, all of which is great, but I don’t understand.” He sort of smiled at me like the Buddha, and he was like, “You have to separate the technology from the tech companies.”
It had not occurred to me that one could do that in the year of our Lord 2024. But it’s true. What I did then was say, “All right, here are the things I care about. I care about education and healthcare and strong government institutions and human connection.” And per Danny’s advice, I went out and found people who knew those problems, who were domain experts in those problems, and were stubborn enough in finding solutions that they decided to import technology.
What I found is, actually, it’s not easy. It’s not a silver bullet. But in a bunch of realms, proper deployment of AI can improve things pretty dramatically. I think that’s a part of the story that hasn’t been told a lot, but I found it very rewarding. It made me optimistic, and it made me want to fight on both sides of this agentic debate: what we don’t want from AI and encouraging more uses that we do.
BERMAN: When you dove into the reporting, did you expect to find what you found, or were you curious whether it just wasn’t out there?
TYRANGIEL: I didn’t know what I was gonna find. Very quickly, I discovered that there were, you know, our federal government is quite large.
BERMAN: It is quite large.
TYRANGIEL: I did discover fairly quickly pockets of incredible competence, of people who care, who understand their problems, who imported AI to solve a specific problem. And look, I think we all recognize competence has become a moral value in a way that we probably didn’t appreciate. What was pleasing about the reporting is competence is everywhere. There’s AI everywhere.
It’s not all suited to the problem, but if you find the right people between the problem and the solutions, you can actually get somewhere. So I did find it pretty quickly. I went deep on the IRS, which, as you guys know, is famously the behemoth of mainframes and old-school computing, and the CTO of the IRS is one of the most inspiring people you’ll ever meet. He was able, over the course of time and tremendous political disdain, to begin the migration of the individual master file, which holds not just our tax returns but every change ever made to our tax returns, for all of us.
He began to migrate that using AI, using pair programming, and got it to a place where it’s now basically capable of interacting with any new software or AI product you can imagine. The last time I spoke to him was the day before the 2024 election. DOGE came in and was basically this project of vigilantes. He stayed.
He’s still there. He’s still making progress. What they were able to do is incredible. It’s just sort of sitting in front of us, and because the politics are so unpopular, they’re not talking about it.
Copy LinkHow AI can modernize the IRS quietly
BERMAN: I hadn’t thought about this before, but it strikes me, when the reporting came out, that all this Treasury Department, including IRS data, was effectively being hoovered up.
TYRANGIEL: Yeah.
BERMAN: Elon has talked for years about wanting not to launch a bank, but to replace the entire banking system. What he would offer is like X Bank, which, by the way, he’s licensed in like 40 or 41 states as a money transmitter now, so it’s a prelude to that, and use all that data to offer free optimized IRS returns for any new banks with him. You talk about scary vertical integration. If he’s got all of our money —
TYRANGIEL: Yeah.
BERMAN: He can freeze it, he can share whatever he wants, et cetera. What I hadn’t thought, and I’m embarrassed that I hadn’t, was competent people at the IRS could do this for all of us for free.
TYRANGIEL: Yes.
BERMAN: Is that something we should expect is coming if he stays and a next president empowers that?
TYRANGIEL: So the capability, particularly at the IRS, because all they have is data, to make pretty rapid improvements is there. It needs to be funded properly, right? The thing that I learned, that I think many people in this room know, is the friction point for most things is human beings: bringing the culture along, explaining to people why you’re modernizing, why their job is not immediately under threat, what we’re gonna do if it becomes under threat.
The care that you have to take in order to get there is particularly hard in the federal government, where your funding is constantly at war, where people are always singling you out. Is it coming immediately? No. Are we capable of it? Absolutely.
One of the lessons — I mean, I was really struggling and trying to figure out how to navigate this in the federal government. I took a quick vacation, right? So I went to East Lansing, a great place to vacation.
BERMAN: In December, maybe.
TYRANGIEL: In December.
BERMAN: Yeah.
TYRANGIEL: And I found this guy named Cliff Walls, who’s the environmental manager of East Lansing, which basically means sophisticated stuff, but also trash collection. Cliff was like three days into the job, and one of the trash collectors came in with an email and was like, “Hey, these people say they can really help us with recycling, and there’s some AI.”
Cliff doesn’t know anything about government at this point. He’s out of school. He looks it over, he’s like, “Oh, that’s cool. It’s a pilot program. We should just do that.” It was like an $8,000 pilot. They somehow found the money, and the pilot was basically visual AI. This company in Saskatchewan — two guys who used to work at the dump — created a visual AI trained on tons and tons of photos of contaminants and proper recycling, attached it to the truck, and what made it more than AI, what made it public policy, is that when they found a contaminant, they did an A-B test.
They would send out one of two postcards, because it’s all connected via GPS, to your home. One would say, “Hey, we noticed you’re putting the wrong thing out at the wrong time. Download our app. We encourage you to do it.” And the other actually took a photo of the offending item, and it said, “No, you did it wrong. Next time, do it this way.”
I’m hearing this. My wife, as Jeff Berman knows, used to run the New York City Transit, so she ran the MTA. And she hears this story, and her eyes roll back in her head like, “What? They took photos of their trash and sent it to them?” But because Cliff rolled this out very slowly, they never said it was gonna solve every problem. They never even said it was AI. They just said, “We’re trying a new thing.” It worked.
What happened is they improved the recycling rate by something like 40 percent. They did, I think it was, 12,000 test runs, and they got two complaints. Two. Cliff and I were just like, “How did you slip one past the goalie?” What we discovered is, again, it comes back to competence. If you just demonstrate to people that something works, we’re in an environment where people are so starved for effectiveness that they don’t give a shit if it’s AI. They don’t care. What they want is stuff that works.
So I mentioned this to people inside the IRS, and they were like, “One, go Cliff. And two, yeah, if we can depoliticize this, if we can get it away from everybody thinking each AI product is Sam and Dario arguing over the fate of the world, there’s a lot of ground to be gained. There’s a lot of effectiveness.” So I really gravitated to these quiet uses that are not perfect, but they are better than what we have.
Copy LinkWhy small pilots can unlock public trust
BERMAN: I love that Cliff, and $8,000 in a municipal budget is no joke.
TYRANGIEL: Yeah.
BERMAN: I mean, that is a real thing to find. And also he had autonomy. He had the ability to say, “Oh, we’re gonna try this,” right? And if you are the Los Angeles Unified School District or the Cleveland Clinic, that’s much harder. So when you looked at education, healthcare, pick your system —
TYRANGIEL: Yeah.
BERMAN: Were you able to find pockets of that as well?
TYRANGIEL: Tons. And it really, in the end, came down to leadership. So the guy who runs Cleveland Clinic is a cardiac surgeon. He is very stern. The first thing he told me — and he invited me in, I was free to wander around — was, “There are many healthcare organizations that are getting into AI because they think they’re going to be the Microsoft of healthcare. That is not us. We have two goals. The first, improve patient outcomes. The second, save money so that we can plow more into improving patient outcomes.” Very, very focused.
Even then, in a group of 80,000 people, they brought in a CTO, and he said maybe eight to 10 people here know how to actually run a project and motivate people and persuade them to try it and love the tech. But with those eight to 10 people, they made huge progress.
One simple example is sepsis, which kills more than 3 million people per year globally. In the United States, about 300,000. It’s the body’s response to infection. Everybody knows that it’s a terrible, terrible thing. So the Cleveland Clinic had lost something like, I think it was more than 3,200 people in a year in its hospital. Totally normal. They decided, we need to stop this.
So they created a sepsis prediction engine. They brought in a vendor. They spent a year and a half working through it. It was really hard to test, hard to pioneer. Not everybody in the organization was thrilled with its results. But at the end of the year, they had reduced sepsis mortality by 41 percent. Is it perfect? No. Is the work of culture change hard? Yes. But man, that’s 1,000-plus people who would be dead. So we need to figure out where it can help and really encourage the people doing the work of implementation, because our lives can really be better.
Copy LinkHow healthcare leaders make AI actually useful
BERMAN: Josh is a staff writer at The Atlantic, which we should all be subscribing to. He’s the author of AI for Good, which you all should read. Thanks for being here.
TYRANGIEL: Thank you, man.
Episode Takeaways
- Jeff Berman opens with Josh Tyrangiel’s stark view of the Musk-Bezos space race, where launch dominance and satellite control now give private actors power that can rival nation-states.
- Tyrangiel argues the real policy question is not whether private space succeeded, but how government responds when one company controls both launch and the infrastructure others depend on.
- Turning to AI for Good, Tyrangiel says public distrust is less about the technology itself than about who is selling it, and urges us to separate AI’s promise from tech-industry hype.
- His reporting found quiet pockets of competence inside government, including at the IRS, where AI-assisted modernization is already making old systems more flexible even if few people notice.
- The most hopeful lesson is that small, practical pilots can build trust fast: from East Lansing recycling to Cleveland Clinic sepsis detection, people embrace AI when it simply works.