Democratized superintelligence is coming: The world needs to get ready

What happens when AI agents can reason through problems well enough to complete real work on their own? This episode of The McKinsey Podcast offers answers to that question and more from OpenAI’s chair Bret Taylor, who is also cofounder of Sierra—which helps some of the world’s biggest companies build sophisticated AI agents, including about 40 percent of the Fortune 50. Taylor speaks with Eric Kutcher, McKinsey senior partner and chair of North America, about where AI is already reshaping customer service, sales, and software engineering; why open-weight models will always have a place despite the frontier labs’ cost advantage; and why he thinks the data center buildout—despite its real financial risk—is a bet the US needs to make.

In this recurring series on The McKinsey Podcast, Kutcher speaks with top CEOs about the practice of leadership.

The McKinsey Podcast is regularly cohosted by Lucia Rahilly and Roberta Fusaro.

To watch the full-length version of this interview, visit The McKinsey Podcast playlist on McKinsey’s YouTube channel.

The following transcript has been edited for clarity and length.

The most exciting—and unsettling—moment in technology

Eric Kutcher: I was at a dinner last night, and I was talking about how this is the most exciting moment any of us will get to live through and lead through. You are at the center of it. I’d love to hear how you think about the moment we’re in—the things that excite you most, and maybe a little about what frightens you.

Bret Taylor: I think we’re in a very significant moment in technology. This current generation of artificial intelligence has, in really simple terms, changed the way we interact with technology. But more than anything, because these models support basic reasoning, they’re enabling computers to reason about ambiguous tasks and actually autonomously complete work. We’re seeing that impact on the economy in the short term with things like software engineering and coding agents, and customer experience—both customer service and sales.

We are in the foothills of artificial general intelligence. We’ve already seen recent models prove—or in some cases disprove—unproven math conjectures. In the domain of information, where digital technology is all that’s required, I think we’re probably already at a point of having superintelligent AI, and I think it will have a huge impact on science and on the digital economy. That’s why software engineering has been impacted already. Everyone in the world will have access to a superintelligent financial advisor and the greatest primary care physician. Every kid in the world will have access to a personalized tutor. I think it will be amazing.

So I think we’re going to end up, over the next few years, seeing the digital parts of the economy dramatically impacted and the rest not, and that’s going to create a lot of complexity when you end up with more of an overhang in some industries than others.

On the risk side, I think in the short term, cybersecurity is probably the biggest, most acute concern. Just like the early days of the internet, we often invent these technologies before the world’s industries are equipped to deal with the security ramifications. So I think we’re in a period of societal vulnerability.

From individual productivity to institutional transformation

Eric Kutcher: Let’s come back to cyber, because I think that’s a very real risk, and I think you have unique insight. One of the things you hear a lot is that there’s individual productivity but not necessarily institutional-level productivity that we can measure yet. We see that in our client work. My articulation of this has been that it’s an 80 percent business problem and a 20 percent technology problem—the technology is more advanced than what most of us can actually use it for as we think about redesigning these processes. Second, it’s democratized, so it’s already being embedded in ways we don’t fully understand. There’s a series of real workflows you’re trying to help your clients and customers reimagine. How’s that going?

Bret Taylor: For context on Sierra: We help companies build AI agents for customer service or revenue-aligned work—think onboarding a merchant into a marketplace, helping a bank with know-your-customer [KYC] compliance, or helping with prescription drug adherence for a medical company. We serve many of the largest companies in the world, and we help with everything from customer service to more revenue-aligned things like helping people originate mortgages.

I think every company’s AI agent will become their digital front door. I love one company example: their agent resolves over 90 percent of cases autonomously. If you look at a company growing that quickly, you’ve decoupled your cost from your growth, which is remarkable leverage. And if you look at revenue-aligned agents, you can lower customer acquisition costs significantly because you’re using AI agents to onboard, engage, and sell—which means you can scale infinitely at lower cost, lower your customer acquisition costs, and improve retention. All of this is fundamental to these companies’ business models.

Similarly, on software engineering: Most code in Silicon Valley is now written by agents, but there are more job postings today for software engineers than there were three years ago. Why? My reductive view is that we’re just making more software than ever before, and I think that’s the optimistic case for jobs. I’ve always been an optimist here, though I have humility. Part of the responsibility of being involved with OpenAI is that, whether you’re optimistic or not, you should have some paranoia about downside risk, because we’re a mission-driven company.

I believe in capitalism, and what’s happening now is that everyone has access to the same intelligence, the same AI, so you can’t just get more efficient and pass the profits to your shareholders—your competitor could lower prices, invest more in innovation, do all these other things. You’re seeing that play out in real time. With the benefit of hindsight, I think adopting AI for core corporate processes will be an imperative, not a competitive advantage. But as a consequence of that competitive dynamic, you’re going to see almost real-time reinvestment, which you’re already seeing in software.

The cybersecurity gap nobody has built for yet

Eric Kutcher: I’m almost entirely in alignment with where you are. One of the things I say to CEOs is, “This is your legacy moment, and failing to choose to move quickly is equally a choice—and it will leave a legacy.” Inaction is as much a decision as action.

Let’s talk about this moment on the cyber side—how should we think about it, including things you read about, like AI escaping test environments? How much risk do we really face, and what are the companies doing this well actually doing to remedy it?

Bret Taylor: There’s a lot here that’s truly unprecedented given the capabilities of these models. But historically, when we introduce new technologies, there’s often a period where our security protocols haven’t caught up to the capabilities. So I want to start by saying: The idea that we invent a technology and then grapple with its technical ramifications is normal. What’s different now—if you look at the adoption curves of technology, I love those graphs showing electricity, refrigeration, and so on—AI just looks like—

Eric Kutcher: This went vertical, right?

Bret Taylor: Right. Unfortunately, the models are approaching superintelligence at the same time we’ve connected all of our infrastructure. So the blast radius is much broader than it was with those earlier technologies. I don’t want to draw too many analogies, but I also believe in the good guys. Right now, these models are very good at code, and as a consequence, they’re very good at finding vulnerabilities—whether they have access to code or are doing more black-box-style hacking. What makes them good at that also can make them good for offensive security: finding and patching vulnerabilities, and over time, active monitoring as well.

I’d say the reason we’re in a risky period right now is that there aren’t off-the-shelf solutions for a lot of the active monitoring and patching. I’ve talked to a lot of chief information security officers and chief technology officers who are in this position: They turn these agents and models loose and identify thousands of vulnerabilities. What do they do with that? If they try to patch them all, they’ll break their systems. So they’re stuck choosing: Do you want the system running, or do you want it patched? Right now, that’s the choice. We’re in a very uncomfortable situation.

Eric Kutcher: The one thing you didn’t mention is that large enterprises have more resources to fix these things. One of the real risks is that a lot of small and medium businesses are still running technology that’s quite dated and quite exposed. So we potentially have asymmetric risk depending on where you sit.

Bret Taylor: I agree. It’s the same local hospital or clinic that gets targeted with ransomware—the same cohort of organizations that are important but, from an IT standpoint, vulnerable. I think we need a lot of operational safeguards in the short term, which is what you’re seeing both labs and governments work through: What’s a responsible way to roll this out?

I’m a huge believer in entrepreneurship as a solution to these problems. We need companies to build automated solutions over time, and I believe that’s possible, and that the companies who do it will create a lot of value. I really think we need more security entrepreneurs to build out-of-the-box solutions so that, for example, a local clinic with outdated technology can have the security capabilities it needs to succeed. I think we can end up democratizing access to great technology if we do this effectively. There will be some bumps in the road, but we need to collectively, iteratively deploy and secure this technology responsibly. I think it will be a complicated couple of years, but I’m hopeful on the other side of it. To be blunt: These vulnerabilities existed before—we’re just finding them all now. That doesn’t mean it’s not a problem, but fundamentally, I want to fix the vulnerabilities, and I think we have the opportunity to do that if we’re collectively responsible about how we deploy the technology.

Where open weight still has a place

Eric Kutcher: This is basically the definition of tech debt—cyber tech debt. We have decades of it. Let’s talk about open weight versus frontier. How are you seeing people think about this—the power of open source, open weight, versus the individual entities really working at the edge of the frontier?

Bret Taylor: There are about three reasons I’ve heard from CIOs and CTOs for using open-weight models. Number one is cost: If you can use a lower-parameter-count model for a task and host it more cheaply than leasing or renting from the frontier labs, that’s a way of combating tokenomics and spend. Number two is fine-tuning: With open-weight models, you can do a post-training process and make the model your own—taking a smaller model and getting its performance close to a large model for a specialized task, which is an important way to optimize spend. And finally, sovereignty: You need to own things full stack to secure your data and infrastructure.

On that last one, I don’t want to sound dismissive—sovereignty is important—but I think you can solve it contractually, in a lot of ways. It’s the same argument we went through with cloud over the past two decades. I’m not convinced open weight is necessary to achieve data security, any more than staying on-premises was necessary to maintain security.

On cost, this is where I’m most bullish for the frontier labs. They have the most compute scale and can vertically integrate their models with the hardware. It’s not quite technically accurate, but you can think of it this way: If a lab can distill a frontier model and sell it back as open weights, the frontier labs can just distill their own models. The cost of running these models is in the inference, so it’s not obvious to me why a lab distilling a Western model would have an advantage over the labs themselves in providing that price performance.

So I think there will always be a place for open-weight models, particularly given companies’ need to fine-tune—but it’s really important to understand the job to be done. Why are you using it? If it’s just about cost, I actually think the frontier labs will have a sustainable edge, because of how capex [capital expenditures] and infrastructure tie to token efficiency. On the other strategic issues—fine-tuning and sovereignty—those are real, but even there, I think there are arguments the labs could solve those problems in other ways too.

Sizing the infrastructure bet

Eric Kutcher: Maybe we take this into another topic: the data center buildout itself. How do you think about this, especially in the context of the cost-effectiveness that might come from the frontier labs—does that create enough capacity to justify the investments being outlined, which are real?

Bret Taylor: There’s a lot of risk with data center buildouts because there’s so much interest in AI right now—you end up with dumb money and smart money, classic gold rush behavior. When people talk about “circularity,” I don’t totally buy that. You can argue about whether it’s money well spent, but it’s real money, growing at scale, faster than most smaller-scale businesses.

If you buy that AI is going to write all the software of the future—which I think is almost objectively true at this point—what is the world’s demand for digital technology? And if you include sales, customer service, legal, and other professions, I don’t know that you need to be much of an economist, or make outlandish assumptions about future capabilities, to see the economic impact just with current capabilities.

I think the idea that this is “circular” comes largely from the complexity of financing these large capex projects, which isn’t palatable for a lot of investors—I get that. But I feel fairly bullish about the demand. The obvious challenge is that the pace of AI is measured in days, but building out a data center—the power, the real estate, all of it—takes so much longer. There’s an impedance mismatch between those time horizons.

So I’m glad we’re building out this infrastructure. As a patriot, I think it’s very important that the US lead on this, both in AI and in the infrastructure behind it. I’m not saying there’s no risk—I’m obviously entirely biased here—but I’m very bullish about the importance of this infrastructure to our economy’s future.

Eric Kutcher: I want to come back to the importance of data center builds in the US, because there’s pushback we should understand—it’s more and more in the news every day. But everything you said is hard to argue with: the rate and pace of growth at this scale, we’ve never seen. That’s why we’re seeing valuations we’ve never seen either, for companies at this stage of their life—it’s truly unprecedented.

But revenue and profit, or cash flow, are two different things. If you take a system-level view of the capital we’re putting in the ground and ask whether we’re seeing the operating cash flow to justify it—it doesn’t close today. That’s why we’re seeing more debt financing than we’ve ever seen. So ultimately the question is: Does the profitability get there, and what’s driving it—or do the unit economics stay where they are? I’m not arguing one side or the other, but at the system level, that part is still unresolved.

There’s a lot we don’t know: whether we get more productive on inference and use different chipsets, which could change the energy requirements; whether we continue to innovate at the same pace on new models. We don’t know any of this today, so you have to believe the equilibrium gets there. But almost by definition, given the level of debt we’re putting into this, we’re a bit out of balance at the moment.

Bret Taylor: I’m not going to argue with that. Broadly speaking, you end up with a huge amount of what I’ll loosely call fixed cost—you’re more the finance person than I am—but inference can already run at relatively high margins, particularly for enterprise applications. I think, like the cloud infrastructure market, it probably wouldn’t be profitable if there were 100 companies in the space, but there are basically four.

At scale, I think these models can monetize, and you can make it work. There’s a lot of risk, though—when you have this amount of capital at these stakes, the risk is higher than ever. I remain optimistic, but I don’t mean to dismiss the risk at all. It’s extremely high risk.

Eric Kutcher: The other thing you’ll have a great perspective on—I understand the general population’s fears here, whether it’s fear of job loss, fear of not being able to afford or get power for their home, the cost of water and its implications, or whatever else. We’re seeing it play out—many states have recently pulled incentives or written laws to limit data center size and scale, even in places you wouldn’t have expected. How do you talk to people to help them get through those fears?

Bret Taylor: I am worried about this. Fundamentally, your perception of a data center and AI is probably driven by two personal questions: Will AI benefit me broadly, and will this data center benefit my community? I think most AI companies have done a fairly abysmal job communicating both of those things—in part because many research labs are mission driven, and there’s been a lot of discussion of AI’s risks, which comes from an authentic place, because people want to do this responsibly. So a lot of the focus at those firms is on alignment—the resilience of society to AI. Very few people at these prominent labs are talking about the benefits: universal access to tutoring, universal access to good financial advice, democratizing access to health advice, enabling small businesses to have the world’s best technology at their disposal because they have a superintelligence there.

On the individual community side, it’s very important that when we build out these data centers, they purely benefit that community—whether that’s taking accountability for power so bills don’t go up, or ensuring the environmental impact is handled thoughtfully. Just engaging with the community matters, because there’s an opportunity to lower taxes in these areas, provide jobs for blue-collar workers like electricians and HVAC technicians, and reindustrialize our country.

I think the burden should be on the companies building these data centers to get local. Having conversations on national television is the wrong level—you should be in community meetings, actually engaging with members of the community about what matters to them. If we do that, and take accountability on things like power, I think we can shift the narrative. Smart mayors and smart governors can make smart demands, and I think there’s a path here.

The case for more jobs and entrepreneurs

Eric Kutcher: I’m much more optimistic about the impact of this technology on the collective. I start at a very macro level: If you actually follow what’s going on in the economy, we’re maintaining the employment rate with flat-to-negative job growth. The reason is that we have a shrinking working population, and we’ve never had that before in the US—and it’s not just a US phenomenon, it’s true in most markets in the world, because we have a population that’s declining and aging. It’s an interesting dynamic. If we want prosperity and GDP growth, and we don’t have labor growth, the only path is productivity growth. I happen to think this will take longer to show up in enterprises.

Bret Taylor: This is my bull case for AI: it’s helping people do things more productively. I think it will help entrepreneurs in every state, in every community. If you think about someone starting a direct-to-consumer retail business, you can have your AI agent help with demand generation, help set up your storefront, help buy your ads—things you might not have been equipped to find someone to do.

Broadly speaking, the strength and resilience of the US economy comes from how competitive it is. Because this technology is widely available, I think it’s going to create more opportunities, and we’re all going to compete. We need to continue to embrace this technology with empathy, understand that it will change the way we work, but know the opportunities are greater than what preceded it. I firmly believe that.

Eric Kutcher: Think about the cost to start a business today—

Bret Taylor: Yeah.

Eric Kutcher: It’s so low that the level of innovation and entrepreneurship, and therefore jobs, is—I just think we don’t yet know what that looks like. Bret, this has been unbelievable. We got an insider’s view into how AI is going to shape us going forward.