Michael, welcome to Y Combinator Startup School. We've got an amazing crowd here. It's been an amazing day, and I'm really excited to talk about AI policy and your role at the White House with President Trump. But I wanted to tell the audience a little bit about your background. First of all, Michael Kratsios is the Director of the White House Office of Science and Technology Policy. He's the president's top adviser on science and technology and one of the key architects of America's national AI strategy. He previously served as the country's Chief Technology Officer in the first term for President Trump, where he led the early federal AI initiatives. He was the Chief Operating Officer of Scale AI in between government tours, so he's seen the frontier from both inside a hyper-growth startup and inside the White House. And so we're going to talk about how Washington actually thinks about AI. So, I want to get in—you know, you have gone from the government to the private sector and you went back into the government, which is actually not a common thing. I think it speaks to your character, Michael, because public service is not easy and it is a sacrifice. You could be out making a lot more money doing God knows what, your choice of roles given the level of connections you have and your background. Why do you choose to do this work?
To me, I fundamentally believe that American leadership in these emerging technologies is one of the most critical questions of our time. For the American people to have all these benefits that AI is going to offer, we have to make sure that we have a regulatory environment that allows that to succeed. And to me, I think even a little bit somewhat selfishly, I think there is no place where you can work on bigger problems than the US government. Even at the biggest tech companies in the world, you're never going to be dealing with problems of this scale. So, being able to work on that is something that I find very rewarding and very fulfilling. I just deeply believe that we have to find a way to keep winning, and the government can either help or they can unfortunately kind of screw things up. So, being there to try to put it in the right direction is something I enjoy doing every day.
Let's rewind the clock. How did you even find yourself in this place? I'm really actually curious: at what point in your life did you just know that you were going to be passionate about science, technology, this whole field? Like, 14-year-old or even earlier Michael Kratsios, up until sort of the age of the audience, early 20s, even late teens—tell me about that part of your life and how you gravitated toward this type of work.
Yeah, I mean, I had always been sort of obsessed or interested in technology. I would follow all the Steve Jobs keynotes every year obsessively. I still remember in college when the first iPhone came out—I'm dating myself here, but it was this amazing moment where we'd run around and talk to our friends about it. To me, I was not an engineer in college, so I always wasn't quite sure how to manifest my extreme excitement for technology into something I could do as a career.
Ultimately, I ended up in San Francisco and worked for Peter Thiel for almost seven years. In working with him and working with a lot of the companies that he invested in, that's when it all kind of came together. Over the course of my time there between 2010 and 2017, as we were looking at more and more companies in the portfolio, what kept coming up was this question about regulation. So whether you were thinking about Lyft and the challenges that they were having at a state and local level, SpaceX with the challenges they were having for launch permits and what the FAA and Department of Commerce were doing—no matter what kind of industry you were looking at, there was this government angle where the government's policy actions could be ones that could actually unlock technology. So when the president won and I had the opportunity to join the administration, my first instinct was, how do we look across all the rules that we have and make it easier for innovators to build? How do we get drones flying for commercial drone operations? How do we get autonomous vehicles on the road? How do we get drone deliveries to happen? And those are things that the government can unlock.
What's something surprising about how technology policy actually gets made that would surprise everybody in this arena?
Typically, what people who haven't really worked in this space don't realize is that these decisions are extraordinarily federated. There isn't just one or two people in the White House who decide something. A blessing, I think, of the US system is that there isn't one agency that does technology. We have a health agency, we have a defense department, but there is no technology department. What that means is a lot of these tech issues are spread out across multiple agencies. So you have equities from national security questions to commercially oriented questions to core science and technology research questions. To make good policy, what the White House has to do is bring all of these agencies together. Whenever—take, for example, drones—if you want to make sure that the rule is right to allow for commercial drone operations, you have to make sure that the FAA has the right rules in place, but also that the people who oversee our nuclear weapons are happy so that drones aren't flying over nuclear sites.
Take us into the last week or so. This week felt like a noisy week in AI policy. You saw the letter that dropped Friday morning from a lot of the larger companies. Y Combinator had signed it, but also Y Combinator helped organize a letter that you were one of the recipients on on Wednesday evening that was advocating for the government to not clamp down on open-weight models. The impetus for that, the energy behind it, was sort of this buzz and rumor, lots of speculative worry that the White House was on the verge of doing an EO that would have restricted or clamped down on open source. Can you talk a little bit about that? How connected to reality is some of the stuff that you've seen on Twitter in the last week or so? And what is the White House's policy? I actually spoke with the Secretary last night at the Correspondents' Dinner, and he said that the White House strongly supports open source. So this is a great audience to clarify: what is the White House's position on open source and open weights?
Luther over here has turned into a little mini-reporter. He's got lots of people following his Twitter. I will say I think what the Secretary said yesterday is the same policy that we had on page one of our AI Action Plan that was released last July. For those of you not necessarily tracking this, the US Strategy for Artificial Intelligence was released in July of last year. It was something that I co-authored with David Sacks and Secretary Rubio. The number one thing, the first thing that we talk about in chapter one, is a commitment to open source—this idea that if the US wants to lead in artificial intelligence, we have to have a vibrant closed and open-source ecosystem, and that's the only way they can all work together. So to me, I think that's number one.
But I think what this week really showed, and what excites me about the job that I'm doing, is we have to have these conversations. If DC is in a vacuum and isn't hearing anything from the startup ecosystem, or even from the big tech ecosystem, or from financial services or all these different industries, we can't make the best decision. So to me, I find it inspiring and extraordinarily practically helpful for weeks like this to happen, because it forces a lot of the community to come up and say what we as Americans believe in, and we as policymakers can internalize that and make sure that we're setting the right policy going forward.
That's the democratic process in action, I guess. Every founder here is building with AI, and from where you sit in Washington, what do you think is the single biggest gap between how Washington sees AI and how this room sees it?
In Washington, when people think about AI, it is a sweeping technology that covers everything from the perception of Middle America—their perception on things like data centers—to questions about AI in healthcare, to questions about job loss and how it's going to impact the labor market. Then also, there's a conversation about how it's going to impact new company growth and productivity improvements across our tech ecosystem. What typically is different in Washington is you can't separate those conversations. It's very hard to not think about the labor implications of the AI boom, or not worry about how generally Americans don't really love data centers—they poll horribly; they don't want them in their backyard. But everyone in this room knows very well that we need as much compute as we possibly can spread across the country. So for us, we end up having to balance a lot of this stuff, which is our job. But we need the input from the startup community to know, at least for that element of the conversation, how to make sure that gets done right.
Governments are usually regulating industries that have settled. AI seems to be reinventing itself every six months. How do you write the rules for something that is moving so quickly?
The answer to that really is you don't want to set very firm red lines in the sand, because they ultimately don't work. I think the best example of that in the world right now is the EU AI Act. When we were in the first Trump administration, the EU went through a lot of fanfare and spent many years putting together this EU AI Act. This rule of regulation throughout Europe was passed and finalized before ChatGPT was ever invented. So now going forward, all of these large language models, everything that's happened since November of '22 when ChatGPT came out, have to abide by this rule that was written before LLMs were even a thing. That shows how challenging the situation is if you try to set the line too firm to begin with.
I think our own administration had challenges with this—the Biden administration set a specific hard-cap compute threshold, where if you do anything above a certain threshold, you have to do a bunch of disclosures to the government. Over time, having these firm red-line thresholds do not stand the test of time. A lot of what we do is to make sure that any type of action can move along with the frontier. What's been proven very much so by the government is that once it sets a line, it's very hard to reset it. So we are very cautious in trying to set these hard thresholds.
It seems like a lot of the policy conversation around AI has to do with safety and risk. What do you think is an AI risk that builders underrate, and that you think is probably overblown?
I think the risk of the day, at least today—there's two main ones. The first one is obviously the cyber risk presented by the Mythos moment, and ultimately the analysis that the labs had to do when they were putting out Fable on what kind of guardrails are put in place to make sure that the more exquisite, powerful, quote-unquote dangerous capabilities of Mythos were appropriately limited for the release. But again, the challenge with a lot of these cyber-capable models is the same model that is able to do something nefarious is the same model that can be very valuable in hardening an existing system. So there's these inherent trade-offs that happen.
The second risk that always comes up that we think is coming over the horizon is this biological risk question. My sense is, at the moment, and it has been for many years, a bit overblown. People were shouting about the bio problem back in '21, '22. It hasn't been an issue for three years. But we definitely need to build the right infrastructure to run the right test and eval processes on models as they creep past the frontier.
Got it. A lot of tech policy conversations end up dominated by the five biggest tech companies. I should say this is one of the reasons that Y Combinator, with a handful of other companies a couple weeks ago, helped spin up a trade association called the Little Tech Association. Because it seems like sometimes—I live in Washington, DC, you know that, and you do as well, obviously. We were actually on the flight this morning at 8:00 together. It feels like kind of a Google, Apple, Facebook, Amazon Truman Show sometimes in Washington. Every position paper you read or speech you hear, you think, gosh, that has some big tech influence behind it. How does a two-person company even have a fighting chance if those large companies are writing the rules? How does the White House make sure that the kind of company that's sitting in this room gets heard?
Well, the most important thing that I do is talk to you as often as I possibly can. But the reality is there's a great number of institutions that help support and stand up for Little Tech. Generally, government is quite cognizant of the influence of a lot of large players in every industry. This isn't a problem or a situation that is only occurring in tech; it happens in everything from energy to healthcare to whatever it may be. What we try to do very hard is utilize mechanisms where you can get input from a wide variety of folks—the traditional request for proposals or request for information process where everyone can submit comments.
For us, I think it's actually working. If you think about one of the major priorities that the president himself spoke about in the speech that he gave when our strategy was released in July of last year, it was this question about preemption of state laws around AI. The fundamental thesis there is that if you create or allow for the creation of a patchwork of regulations—meaning there's one set of regulations for AI in California, another one in Maryland, another one in Texas—the big tech guys can deal with that. Google can hire an army of lawyers and they'll figure it out and they'll be fine. But for all of you out here, that's just not going to work. The president stood up and said, no, we have to have one national standard for AI so we can make it easy for anyone who wants to build an AI company to know what the one set of rules is and build their company that way. So I think that's an example of us really taking that to heart.
Well, I will say I commend the White House for continuing—and really it started with the first Trump term—the bipartisan antitrust project with respect to big tech. US v. Google was started under the first Trump administration. US v. Apple is being continued; hopefully the reports that that might settle imminently are incorrect. FTC versus Facebook, FTC versus Amazon—these are cases that the Trump DOJ and FTC have continued. Coupled with the president's own usage of the phrase 'Little Tech' and championing Little Tech companies, when we met years ago, I said we just have to make sure that Little Tech has a seat at the table. As you go back to Washington, we're doing everything we can to make sure that Little Tech has a seat at the table.
It's important to me. I spent most of my 20s here in San Francisco working at a venture firm. I understand how challenging it is to build these companies. What makes the US so special is that there is no place in the world that is better to do a startup than in America. It is easier here, and the opportunities are just unbelievable. So as a public servant, what I think about every day is: how do we protect that ecosystem? How do we make it even easier? How do we create all the opportunities for all of you to create great companies and succeed? Everyone in the world is clamoring to come to America to build companies. There's something very special about here that we must preserve.
Is there a version of AI regulation that protects consumers, but doesn't hand the incumbents a moat? What does that look like concretely?
Back to what I was saying with this one national framework: it's all about having regulatory clarity and ease for all sizes of companies. If we can work with Congress to do something like that, I think that would be the biggest boon for startups.
The United States's advantage has always been that anybody can start something. To your point, we are unique in that respect. This is the greatest place in the world to build a company. What would you tell a founder here who is worrying about compliance costs locking them out before they even start?
I would tell them that there's probably no administration in history more committed to reducing regulations than the current one. Our Office of Management and Budget, which runs our regulatory process—the guy who's running it is Russ Vought, and his life's mission has been to eliminate as many regulations as possible. That is what we think about every day. From the tech standpoint, when we talk about regs, the biggest question to me is how do we remove barriers to innovation?
There's two ways to think about the world of regs regarding technologies. A lot of people talk about this in Washington: there's technologies that are either born free or born in captivity. For each of those categories of technology, you have a different set of regs to look at. Born-free technologies are ones where there aren't regulations on the books—things like the internet when it just started. Those are the types of technologies you have to preserve. You have to be very, very thoughtful about whether or not you're going to introduce new regulations into that domain, because people can build anything and thrive in those areas. Born-in-captivity technologies are technologies where you're building something, but you can't commercialize it or take it out to market unless you get some sort of government approval. Think about commercial drone operations: you guys could go build an amazing drone in your backyard, build the most amazing software to connect a vendor to a customer, but to actually close that transaction legally and have the drone fly, you can't do that unless you get a waiver from the FAA. Those are the types of technologies where we relentlessly think about how to remove those regulatory barriers or make it much easier to do.
Earlier, some of you may have heard from the founder of Boom Supersonic. I've been obsessed with supersonic flight for years. I think it's the most obvious, in-your-face example of technological stagnation. We had the Concorde flying years ago; we're flying slower today than we were back then—an absolute tragedy. In that situation, he can't get his supersonic plane to fly unless the rules are set such that there is a noise limit instead of a speed limit over the United States, for example. In those born-in-captivity technologies, that's what we relentlessly try to figure out: how do you remove those barriers and make it easier for these technologies to work?
I want to go back to a question that I probably forgot to ask at the beginning, which is about your day-in, day-out role. You're not only the Director of the Office of Science and Technology Policy, but also a special adviser to the president. If you had to average the days across, what does a typical day look like? What time do you show up to work? A lot of people don't realize the White House is a sprawling complex. There's the Oval Office, obviously, but then there's something called the EEOB, the Eisenhower Executive Office Building. How big is your team? Just help us visualize what it's like working at the White House and what your job is actually like day-to-day. Is the role of Director of the OSTP always an Assistant to the President? I think that's an extra, additional role you take on, so talk about that a little bit.
The federal government is made up of all of these agencies. You have HHS, which does healthcare; Department of Defense, which does defense; Department of Energy, that runs energy and national labs. For any type of policy that you do, you essentially have to get concurrence or have some sort of conversation among all of these different agencies. There's only one building in the whole world that can get all these agencies to come together, have that conversation, and bring some resolution, and that's the White House. Generally, there's four categories of policy that get sorted out in the White House: national security, and there's a council that runs that process; domestic policy, that's healthcare and immigration-type stuff, and there's another council that runs that; economic policy, that does tax and other econ; and then there's us, that does science and technology. Essentially, there's four policy councils and four policy leads, and each of us run policy processes on the topics at hand.
In our portfolio, we obviously have AI, but we do things like quantum, civil nuclear energy, biotech, space. For each of those portfolios on my team, I have one or two people that help run that portfolio. I have a space team of about three guys, and they coordinate space policy across the government. They bring NASA in, they bring the Department of Defense, which has a bunch of satellites for national security purposes, and they all sit together and sort out the policy. On a typical day, the general things that you do are: one, stakeholder meetings. Luther comes and says, 'Hey guys, you got to look out for Little Tech,' and I listen to him, and then Google says—
