I often use the phrase with my team, the really famous Benjamin Graham phrase about, "In the short run, markets are popularity machines; in the long run, weighing machines." And I view that, too, about the pace that we live at. We have to think about what will be enduring. Our mission has to be enduring: it's for the benefit of humanity.
Welcome to Gray Matter. I'm Declan Kelly, founder, chairman, and CEO of Consello. I've spent the last 30-plus years advising the CEOs who are leading the world's top companies. If there's one thing I've learned in that time, the best leaders think differently. That wiring, that gray matter, separates the truly exceptional leaders from everyone else. Understanding it is what allows us at Consello to help the best in the world be even better. This podcast exists because the most accomplished leaders agreed to sit down and talk about something they rarely discuss publicly: how their minds actually work.
Artificial intelligence is changing the world as we know it, and only the people inside a few frontier companies know what's truly coming next. One of the most important of those insiders is Sarah Friar. As CFO of OpenAI, she oversees the capital and infrastructure behind one of the most consequential companies in the world. Her career has been defined by a desire never to stay in one lane and by a willingness to question the limits of the status quo. That instinct has carried her from the financial world into operating roles at hypergrowth technology companies, and now into the center of the AI revolution at OpenAI. Sarah's story illustrates what can happen when you keep moving toward the unfamiliar and push beyond what's been done before. This is Gray Matter.
Sarah Friar, welcome to Gray Matter.
Thank you, Declan Kelly. It is a pleasure to be with you.
Two Irish people talking about the future of the universe. We always ask our guests to pick where we do our interviews, and you've chosen the library at OpenAI. Why?
I am such a book fanatic. I try to read 52 books a year. So despite living in the age of intelligence where everything is supposedly electronic, I love books, and so I always feel at home when I have my books around me.
So let me get this right: you read one book a week?
That's my goal. Now, do I do it? Not always, but I get pretty close. I get into the 40s most years.
Where do you find time to do that? How do you do that?
I have to read before I go to sleep. Otherwise, I cannot go to sleep. So, it is my de-stress switch off. And then I like to run. I do a lot of trail running. So, I have a thing with books where I call it "eat your greens, and then you get your dessert." So, eat my greens are the books I should read, and then my dessert is I just get to read whatever I want.
Interesting. This building used to be your headquarters, I think, in a long far-flung past, at least six months ago.
And 50% of the people who work here, somebody told me on the way in, didn't work here 6 months ago.
What's it like coming from a corporate environment like you've come from, walking around the corridors here? The entire thing is a completely different environment to what most people in business are used to, and the speed of change is extremely difficult for people to understand. Yet you've had to acclimate. What's that been like?
So, I think in a fast-paced environment where everything is coming at you all at once, you actually have to pull back and really understand your north star. So, for us, it's our mission: AGI for the benefit of humanity. And there's a little of me smile sometimes when Silicon Valley companies spew out their missions, but they're vital in these moments where the news—I often use the phrase with my team, the really famous Benjamin Graham phrase about in the short run markets are popularity machines, in the long run weighing machines, and I view that too about the pace that we live at. We have to think about what will be enduring. Our mission has to be enduring: it's for the benefit of humanity. And it helps us in moments like not humanity who can pay or not humanity that can deploy on Azure or whatever; it has to be about broad access.
I think the second thing that is enduring are your customers. For us, that is a broad intelligence layer that goes from literally my mom all the way through to the largest enterprises and governments in the world. And so our customers and our relationships with them have to endure. And then finally, the way we walk in the world—how we build trust will be the thing that makes or breaks us at OpenAI. And so I always, despite the chaos and the noise and the growth and the thing and the thing, you have to recenter yourself almost every day and remember: what am I building that's truly enduring?
But isn't it a huge advantage for you personally to have had the background you've had because you're able to step into so many different environments? You've been a metallurgist, you've been in a mine in Ghana, you've worked in McKinsey, you've worked in banking, you've worked in consulting, you've worked in Nextdoor, you've worked in Block, you've worked in lots of places.
It is kind of incredible the course of your career. It is always helpful to look back because the line looks a lot straighter, and in reality, it was a lot of zigs and zags. I actually feel like you should take risks in your career. I have a phrase internally in my head that if you don't get an adrenaline rush every day, you might want to think about what you're doing. Most people are like, "That sounds awful," and I literally love it. My brother is a doctor and he's the guy you see if things are really not going well—he's an ER plus-plus—and he always talks about that moment when he's trying to bring someone back to life, you shoot them with adrenaline. And so to me, it's lifegiving.
So that's number one, a lot of zigs and zags. But you do get into this moment where you're like, was it all culminating towards a thing? And it really feels in some ways like it was towards this moment of AI in the world and what could that mean, and then OpenAI. So whether it's earlier today we were debating open source—I co-authored an original report called "Fear the Penguin," which was about Linux coming into the world, and so I was trying to use my pattern recognition of something that happened 20 years ago to today. When we started to think about ads going inside ChatGPT, I feel the benefit of Nextdoor; I spent seven years building a really hard ad model. Just a few weeks ago, we signed a deal in Georgia with Georgia Power, and now we're really having to work with the community at a grassroots level. The thing I learned at Nextdoor is do not tell people what they need, because they're not interested in what you think they need. They will tell you exactly what they need, and it will be the right thing for them in that moment. As an example there, they have told us that one of the biggest needs is food security, and it is kids getting school dinners. I think back to myself getting free school dinners and how important that was, and it just felt like a beautiful full-circle moment. So yes, a lot of things you could look back on in my career and say, how did that all stick together? But I really feel like each one was a building block to what I'm doing today, and I feel so privileged to get to do what I do today.
Well, your upbringing was obviously central to how you got here, but also fascinating. I'm going to touch on that in a moment because we have a lot of kindred thoughts from Northern Ireland. But when I wander around the halls here and I see how people are shaping the future of how we all think, how our lives are going to change in that way at a molecular level as opposed to what most people talk about on television or in TED Talks or in articles or white papers or earnings calls—it's all sort of theoretical for the future, but you live in the consequences of the here and now. And so what do you see when you look at that?
A phrase I use for my team all the time is actually I think we live in the future right now, and it's an incredible place. The future's awesome, first of all. But what I mean by that, with all due respect and not trying to sound lacking in humility about it, we are getting to experiment and use the tools put in front of us often way ahead of the rest of the world.
For example, ChatGPT has gone from something you ask—which, I got asked earlier in the week what was the AI wow moment, and it literally was the first time I used ChatGPT. I was like, "This is incredible." You talked about my background: I was showing it to my dad, and he was like, "What is this?" And I was like, "Tell me something you have to write." And he's like, "Oh, I have to write a speech for whatever." My dad's still giving speeches at 89. And I was like, "Okay, tell me a little bit about it." And we wrote a speech right there, and he was blown away. And then I made it in the style of Taylor Swift, and then he was even more blown away by it—or at least my daughter got interested at that moment in time.
So that was then. And now we're at this moment of doing, where actually ChatGPT just became ChatGPT Work. For the first time, over a billion people will get to experience agentic behavior. And what does that mean? My mom would be like, "I don't know what you're talking about." But it's literally going from asking to doing. So, it might be something very simple like, "Help me book dinner in two weeks when we drop my son off at college, and it needs to be for 12 people, and here's some of the restaurants we like, or suggest one." Imagine if now everyone in the world has a personal assistant that goes off and does that for them.
And that's scratching the surface, right? It's like we got given electricity and we figured out it turns on a light, but we have no idea all the other stuff. Or when we first got a mobile phone, we didn't know you'd be able to call a car, order food, or do your personal finances—all those things. We're at that beginning point. Talking to the world about it, though, in some ways I think we lose the humanity of it. So what's more interesting is to say, okay, let's talk about a customer where it's really moved the needle for them. BBVA, one of the most amazing banks in the world, they're utilizing it across, I think, 120,000 employees who are literally saving hours every week. If I go to someone like Cisco, a bit more of a tech company, but their developers have gone from what they do every day to three to 10x more productive. So that means more product, more revenue, more outcomes.
And then you get to the things that really get to my heart and soul, which is seeing—you talked about teaching—seeing a teacher use it in a way where you're in a classroom, and they're teaching a kid something like algebra. You have that bell curve moment. Every teacher knows it where a chunk of the class is coming with you, there's a group at one end that are so far beyond you that are either going to stay engaged or get bored and start pelting each other with a piece of paper, and then there's the group at the back of the curve that it's not clicking for them.
I doubt that, too. But if you could suddenly multiply your teaching moment where ChatGPT can actually help you redesign in the moment—let's have different work we can give to these kids that are moving further ahead, I can give the torso of the class to keep going on the script, but let me just figure out a totally different way to help this child engage—and that just wasn't true for us.
I describe you as the adult in the AI room. And what do I mean by that? Because someone has to walk into the room at the end of the day and say, "Guys, this is awesome, but we actually have to make money doing this while also helping humanity." And the numbers get silly very quickly here. The world's ability to grasp these kinds of numbers is just not there yet. Even the business community is still marveling at this whole concept. So, how long do you think it takes for the business community at large to integrate and learn how to make money with all of this technology?
I think we're already there in many, many cases. My job is to make sure we have the right capacity at the right time to meet the demand. So, we're not willy-nilly just buying stuff in the hope—let's buy it and hope it will come. Hope is not a strategy. We very much see the demand today, and that's not made-up demand. The eras of AI have been quite quick: in 2022, ChatGPT breaks on the scene; in '23, people are just learning this call and response, they worry about things like hallucinations; at the end of '24, we had the breakthrough of reasoning, and that's the moment where you really moved to System 2 type thinking and all of the things like hallucinations largely gone because of reinforcement learning; '25, agentic behavior.
So coding was the beginning because it is effectively—like Bret Taylor on our board said it last week at a company all-hands, coding was humans trying to act like machines for a long period of time, like a lot of data entry, but now you've got a computer that will do that for you and you can actually be in charge of the logic. I think CS as a major is still an important major in college because it teaches you logic. There are more developers today than there were two years ago, so the death of the developer has been way overblown.
If you think about being—you called it the adult in the room, but I think more about how do we turn it into real business value—we are absolutely seeing that. I looked at, for example, our customers that are furthest along, the ones that are public: their earnings growth last year versus what analysts thought coming into the year, they more than beat. Maybe it's causation or correlation, but there's something there. And what's interesting to me as an ex-sell-side analyst is analysts have kind of reverted to the mean on that earnings growth. And so if that's true, there's actually huge upside still on all those stocks because I don't believe that their earnings growth is over. And that's why I come back to examples like BBVA, Cisco—you give me a vertical, I will give you a customer: Walmart, who's actually doing things at scale, transforming their business. Coding was an easy start. Cyber is now the next obvious one. We should spend time on that for your audience because this is an area, if you're not waking up in the morning talking cyber with your team right now, you are not doing right by your business.
It's kind of interesting because every time somebody puts out a press release saying, "We've had a cyber breach, but don't worry, everything's going to be okay"—I've dealt with five of these in the last couple of weeks—the market doesn't even give it a beat. But the consequences of the consequences, as my own mentor used to say, are far-reaching. And so the combination of that plus AI ends up being potentially somewhat apocalyptic if it's not handled the right way. I assume you agree with that?
I think it is a really new—companies catch up. So again, career-wise, I started as a sell-side analyst at Goldman—did a couple of things before—but I made my career in security because I was there when Nimda and Code Red, these two viruses, happened. We wired up the internet, things were looking good, and then suddenly we were like, "Oh, bad things can go along those pipes, too." And so that birthed a whole industry of antivirus. And then you had these moments like Stuxnet and SolarWinds. I think you're at a moment like that right now.
I don't think it will be apocalyptic, but I do think that we need to make sure, first of all, mission-critical infrastructure is really locked down. We definitely have nation-state actors who are doing bad things if you're sitting in the US and democratic countries around the world. We have to get on this fast.
And is government able, ever? Is this the first time ever—and I worked in the government—is this the first time ever the government will never be able to catch up? Because regulating is one thing, but controlling and managing is another. And it's not a criticism, it's just a fact of life because governments aren't trained, born, or enabled to move at this speed. So how do you deal with that push-pull?
I think this is one where, first of all, we need to show up as really good partners to governments around the world. And that's what's been really different about this job: when you say customers, you're not just talking about companies, you're talking about countries. So we have to show up as really strong partners. Beyond that, the government itself, we need to make sure they are secure because that is mission-critical infrastructure.
And then I think government—not big G, but government as you work your way down: state and local, local cities, mayors, and so on—this is where trust can be won or trust can be lost. For their constituents, you want to bring people along. Imagine if you had said, "Okay, let's not have electricity. It's too dangerous. If you touch that wire, you're going to die." That's true, we could take that approach. Or we could say, "We could earth that wire. We could put insulation around it. And now we can have lights and lamps and TVs and heat our homes or cool our homes." There's a lean-in moment that you want to do for your constituents, but we have to do it in the right way.
What's your view on the whole SaaS apocalypse phenomena where companies are getting disintermediated because they think the big bad boogey monster is coming to take away all their business? There's some evidence that the Street believes them. What's your view on that?
I do think the world of software has dramatically changed with AI. There's no other way to look at it, right? The fact that anyone can code now—empowerment has happened. A lot of my career has been about: how do you get resources to people with ambition? In this case, what I see in my own team is I now have finance people who are coding. So that's kind of point one.
What does that mean? That means the old world of packaged software, very deterministic by nature, still has a place because that's often where either data is captured or controls are managed. But the layer on top now, where the user is really interacting and doing the thing the user wants to do, that has become a Cambrian explosion, to use a great Sam Altman term, because anyone can code on top of that.
I'll give you a very mundane example: our procurement system. We have a SaaS vendor that does procurement. Procurement is a very important part of any corporation. We have something like 80,000 vendors that we work with today, and everyone shows up all the time wanting to buy something new. It's a very rote process, but in this case, we were able to chunk it out. We keep where the data resides and also the controls, because the great thing about a procurement system is not actually, "Can I buy something? Okay, I've bought it." It's that if you want to buy something over a certain amount, then this person needs to have said yes all the way up to me or ultimately Greg or Sam. It's actually that delegation of authority that's very important, so I don't want to lose that. But then on top of that, why wouldn't I massively customize it for OpenAI, for example? Or why would I have people having to do their own credit checks when I can get an agent to do that? So, I think that the world of software, the great companies are moving apace and starting to talk about headless APIs. It's a very tech term, but what does that actually mean? It means they're exposing out a lot of the good of what their software brings, but they're allowing companies and individuals to do massive customization on top.
But the scale of the expenditure—you did the largest fundraise in history and led it successfully. I think it was $120 billion or something like that, maybe more.
Lose 2 billion, you could drop that off the end of the boat, nobody will even notice anymore.
Well, maybe. No, but in all seriousness, if you look at the numbers that are being bandied about for debt financing for these projects and everything that's going on, the quantification of this and the speed with which it's happening, and getting customers even to believe in the future that there will be a pot of gold at the end of the rainbow, but it's 5 years away and it's going to cost you $20 billion today—it's never been done in the history of mankind. So when you're having those meetings and you're sitting there asking for that with a straight face, which you inevitably do, are you being met with an evolving reaction or do you feel like there's still incredible shock when people sit there and go, "Oh my god, I didn't realize it was going to be like that"?
It's definitely an evolving reaction, right? Even from last fall when OpenAI was talking about how much compute the world would need in order to have an intelligence layer that could meet every customer that wanted to partake of it, that was heavy sledding, a little uphill. I feel like we're now at least in hindsight being judged as more prescient. And what is fabulous is that the compute I bought a year ago, I could sell in the market today for 3 to 5x. So if nothing else, a great investment.
Now, the bad news is that we're still short compute, so I should have bought more of it. I might live in the future, but my future still needs the screen to get a little clearer, higher fidelity. What we are doing is like building the railroads. It's very different from the internet era because it is true capex, it's true shovels in the ground. And the hard part is you're creating capacity that has to show up a year from now, two years from now, three years from now. Right now, I'm actually spending a lot of my brainpower and my team's brainpower on 2030 through 2035. So you're having to really live in the future, frankly. But you have to pay for it today and show the shape of a curve of revenue and good old operating income that can pay for it.
EBITDA, like CFO's favorite word, and you are definitely having to explain that. Now, a year ago that was very, very hard. As an example, a year ago when I walked through the eras of AI and we got to agentic behavior, we were trying to explain there'll be this thing called an agent and we think one of the first things it'll be able to do really well is coding. So, how we created—good finance people—how we created the TAM, the total addressable market, and then the actual market we can go after: we literally had counted how many software developers are there in the world, how much would people be willing to pay per month to give them a tool. Investors a little bit rolled their eyes, right? In the end, what I actually heard coming out of that round of financing was, "We fully underwrite ChatGPT for the consumer because we can see you're the fastest-growing consumer app in history. It's a wow factor, people are clearly using it, and we think you'll both be able to build subscriptions and an ad model. We underwrite that. The agent thing? Yeah, I don't know."
And here we are today where people are spending—at the time, shame on us, we didn't think big enough, right? We had them paying maybe $1,000 a month. If you think $1,000 a month, that's $12,000 a year per engineer. People are now saying, "I pay my engineers $250,000 a year. I'm getting three engineers for the price of one. So maybe I should be willing to pay another $250,000 a year to give them the tool."
I analogize it to finance people. Imagine if we showed up and all we had at our disposal were—not quite stone tablets—but calculators even, or we didn't even have that right before the era of productivity suites and Excel. Your ability to do finance was so minimal because so much of it was data entry and hence—
Very laborious. So I just view us as on a continuum, and what just happened to coders is now happening to knowledge workers. Like my own behavior every day: my agent wakes me up in the morning, has gone through my inbox, my Slack where we do a lot of our work, it can see all of my documents and my Excel spreadsheets. It has listed the top five things that it thinks I need to triage the minute I jump out of bed, before I go to the gym. And it gives me not just—it says, "Delegate, do," it gives me a best next step, and it can take me to that email or to that Slack message immediately. And that's all while I've been asleep. And so that's just a beginning.
And by the way, I do the same for triaging all of my Wall Street emails that come in, and I turn it into a little—
I get a lot of those. God bless those analysts that are sending me things at 4:00 a.m. I turn that into a little podcast that's just for Sarah. It's not Gray Matter, but I might have to call it something now because I'm inspired, but I literally get a little five-minute podcast every morning that I just listen to when I'm brushing my teeth and getting ready and putting on clothes. I carry my phone around with me and I listen to my personal podcast. And that's kind of back to that point about software companies: it's the massive personalization that's available that blows my mind.
We're now in a situation where intelligence is becoming a commodity rather than a scarce resource like it used to be years ago. And so when you commoditize intelligence, the human's ability to fill their day with stuff that actually is worthwhile, it becomes a really interesting question. And I'm not asking you to be the evangelist for the industry because I know it's a very personal thing, but there is this thing: 24 hours in a day. And at a certain moment in time, aren't people going to sit down and say, "What the hell am I going to do with those 11 hours?"
I think about this a lot. Number one, one of the best gifts of the Industrial Revolution was the weekend. Now, I would like that gift to come back to me again, but literally, you got no—
If you had talked to my grandparents, who were farmers in Northern Ireland, and you told them there'd be two days a week where they would just kick it back, they would have said, "You're crazy. What are you talking about? Who's going to feed the sheep or the chicken, get the eggs?" Right? So, number one, the world does keep evolving. Number two, as we have looked deeply at jobs, yes, I think it's a much more nuanced conversation. I do think that there's a chunk of jobs that are going to be automated. And by the way, in many cases, I think the world should celebrate that.
There's a bunch of jobs in the torso, like a good 50-plus percent, that are more like your and my jobs, where they're augmented by AI. I think our work gets more interesting. I don't have to read 15, 20 emails in the morning to get a sense of the world. I'm very grateful for that. I think there's new jobs we just don't even know, right? The idea that we sit here today and have a sense of what jobs look like even in 5 to 10 years—I mean, who'd ever heard of a prompt engineer until like a year ago? Who'd ever heard of a forward deployed engineer unless you were Alex Karp of Palantir, right? These jobs are already starting to really take off around the world.
When I talked to friends, like at a very large consulting firm that I may or may not have been associated with, I talked to the CEO from there. He's talking about the fact if he splits his firm in two, there was kind of front-office people—people out with clients all day, hopefully driving great outcomes—and then there was the back office. He's kind of halved his back office, but he's added the number of people he took out there to the front office because there's so much transformation going on and the pace of the world has so increased that actually the net is that the company hasn't shrunk, and if anything, there's maybe more entry-level jobs right now than there were a year or two ago.
And judgment becomes precious because the formulation of thought is obviously the faster you can do it, the better. But at the end of the day, I even see it in my own job that at the end of the day, there is a moat called judgment, and it finds its landing spot naturally over time. But there are a lot of growing pains that go with that. Gray Matter is really about how people think, and you are in a gene pool of one in Silicon Valley in that you're the only person I know who came from Northern Ireland. Having been to every town in Northern Ireland when I was the envoy to Northern Ireland, I love the place deeply.
Deeply. But I also recognize what a culture shock it must be to you personally to sort of, when you think about where you came from and where you are now, it's got to be fascinating to be you alone looking out the window when you're not on your phone or checking your emails or listening to your podcast for five minutes to think about that journey.
It is incredible. It's a very grounding thing. I have found it to be much more a strength than a discombobulating experience, if that makes sense. It goes back to what I said even about OpenAI, right? The North Star always has to be there. So my North Star, I grew up in the Troubles. I grew up on the border, so it's pretty rough. You know kind of where I come from. Strabane signs, they're not salubrious in any way, shape, or form. Good farming stock.
But I think what I learned there, my mom and dad in particular, I talk about them over and over, but they deserve it. My mom was the local nurse. She birthed a lot of babies; she was actually the midwife. My dad was the personnel manager for the local mill. But for them, the knock on the door was the moment where you're going to answer it and help, or were you going back to dinner? And they always answered the door. To me, that's the thing that I think about all the time: always answer the door. Sometimes that door opens and actually it's an opportunity for you.
I've been so gifted in my life of people who I've gotten to know in one context and maybe I don't touch for 5, 10, 15 years later, but that open another door. Coming here to OpenAI was through someone who had been on a board on the Block board. Actually, the other piece is, how do you pay it back though? The other learning for me on the pay it back is actually it's not that the other person hopefully ends up better off, but you end up so much better off, right? The thing I learned about Nextdoor is this thing about kindness, that in doing a kind act for someone, it also has a very physiological impact on you because it causes a cortisone reaction to happen in your body. It takes down a lot of inflammation, good for your heart, all of the good things, and that's before the other person even benefited. And I think Northern Ireland gave me that core of like open doors.
I remember reading somewhere that your mom almost delivered you in the ambulance on the way to the hospital because she was dodging—
Right, the bonfires during the Troubles. And the incredible experience you had, you went to Oxford and interviewed in your school uniform.
