Sam Altman’s latest interview: when everything is changing exponentially, what mindset and judgment should you have?

Editor: Deep Think Circle

Have you ever wondered how a startup that only existed for two weeks can recreate an entire mainstream office suite—documents, spreadsheets, presentations—redesigning everything around AI? This isn’t something I made up. Sam Altman (OpenAI co-founder and CEO) said it himself in a recent interview—he had just met such a company. Ten years ago, what a ten-week-old startup looked like could almost be predicted. But now, if a ten-week-old startup still looks the way startups looked ten years ago, it actually means it’s fallen behind.

This interview is packed with information. They talked about entrepreneurship, how OpenAI has evolved over the years, and also how he deals with pressure and makes trade-offs. I pulled together the parts that hit me the hardest, added a lot of my own understanding, and I’m writing it for you.

Most people are still fighting the easy wars

Altman said that this moment in time is especially interesting: costs are falling fast, and the cycle of getting things done is shortening fast. This is exactly when startups have the biggest advantage. This is happening across many fields at the same time. In theory, it should be the best era to build startups. But he observed a pretty contradictory phenomenon: most startups are still doing the same thing—building AI agents for a particular industry. This path can work and even make a lot of money, but it probably won’t become the company people truly remember from this era.

What I find especially thought-provoking is the line he said right after that. Even though the tools have completely changed— even though model capability is still rising—people still aren’t really willing to bet on something they can’t do today but might be able to do in two years. He said the temptation is enormous: use today’s agents to solve problems you can solve right now. Everyone understands that—but it’s not the path he would choose.

When I thought about it myself, I realized this is really a matter of patience and belief. Being willing to set up plans for something you can’t accomplish yet is basically a bet that the models will keep getting stronger, and a bet that your judgment about the direction is correct. Most people can’t do it—not because they can’t understand the trend, but because they can’t resist needing returns right now.

Believing in exponential growth is harder than it looks

Altman mentioned a method he has been using himself. Whenever he meets a newcomer, he mentally places them on a coordinate system—judging where they are right now. The next time he sees them, he checks how far they moved forward and how fast. He said this is the same kind of underlying belief as how he judges model capability improving. In plain terms, it’s his deep trust in exponential growth—whether that growth is happening in a person, in a company, or in a model.

He said that if he is still giving advice to founders, this is the one thing he most wants them to truly understand. And he said it’s hard for this to be widely accepted because the market itself hasn’t adapted to the fact that model capability will keep moving forward at an exponential pace. So starting now on projects that only become viable with smarter and cheaper models is actually completely reasonable.

I think this part is especially worth thinking about. Believing that a curve will keep going up sounds like a simple principle. But to stake all your decision-making on that belief takes a level of courage far greater than most people imagine. Most people’s intuition about exponential growth is wrong: either they underestimate the accumulation from the early years, or during the middle phase—the fastest climbing period—they start to doubt whether this is about to burn out.

Enduring chaos is something you can only learn through experience

There’s a section I remember especially well. Altman said that no matter how clear the logic is in someone’s head, some capabilities only come from repeated real experience—operating while things are in chaos, and trusting that you can ultimately handle it. This won’t kill you. Even if you don’t know yet how to solve it right now, you will solve it. He said this is something you can only learn and not teach, and he believes it’s one of the biggest weaknesses of many young founders—they haven’t gone through the process of slowly learning to live with chaos.

He also shared a very blunt metaphor. The first time you encounter something big that could kill a company, it feels like the sky is falling. Then when you’ve made it through the tenth time, you’ll think: I’ve survived the first nine. This time probably won’t be that bad. Later he realized something: bad things were always going to happen. Instead of resisting them, it’s better to learn to accept this uncomfortable process. He said most people think the opposite of a bad experience is a good experience. But actually, the opposite of a bad experience is having no experience at all. And not too far in the future, you’ll eventually reach a phase where nothing dramatic happens. So even terrible experiences are worth feeling grateful for.

When I read this line, I froze for a moment. We’re too used to treating pain as something we should avoid as much as possible. But if the opposite isn’t “comfort,” but rather “emptiness and numbness,” then enduring chaos doesn’t just look like a cost—it looks like part of living.

A trustworthy company—what promise does it actually make to the world?

When Altman talked about mission, he mentioned something he cares a lot about. One of the AI risks he worries about most is that a small group of people, or a company, might believe it should control the entire world. He calls this AI authoritarianism. So what OpenAI wants is to make intelligence extremely abundant and extremely cheap—put it into everyone’s hands, instead of hoarding it in a few hands. He especially emphasized that they don’t plan to build products in every vertical themselves. Instead, they want to get the underlying capability of intelligence right, so that the whole economic ecosystem can naturally grow all kinds of things on top of that foundation.

There’s a part I find especially interesting. He said that the things needed to create abundant intelligence—chips, energy, data centers, robots—are also exactly the things humanity will need most right after intelligence becomes abundant. Even if ideas and creativity become “cheap,” we still live in a physical world. We still need things to be truly built. So energy and robots aren’t just stepping stones toward that goal—they’re also the things that will be needed immediately afterward.

When I read this section, my first reaction was that the logic is quite straightforward: no matter how smart something is, if you want it to happen in the real world, someone has to move physical matter. But thinking more carefully, it’s also reminding us not to treat “intelligence” as something too abstract. No matter how capable a model is, it ultimately has to land on a pile of extremely heavy, extremely physical infrastructure.

A company as an invention matters more than many technologies themselves

Altman talked about something he thought about when he was a kid. He had long been curious about the Industrial Revolution: how a bunch of technologies happened to emerge around the same time, and happened to expand at roughly the same pace. He kept wondering which technology was truly the key one. He said that looking back from today’s perspective, the most crucial invention was actually the joint-stock company itself. Before that, business relied on trust among acquaintances—lots of family businesses, with no concept of shareholders. After joint-stock companies appeared, sovereign nation-states granted this completely new kind of entity an unprecedented status: not giving it the power of the state, but giving it abilities far beyond individuals. It can pool capital, take on extremely high-risk and highly speculative things, and allow different companies to specialize in different stages while coordinating with each other.

He brought up a chart that I really want to look up: the decline curve of the proportion of people living in extreme poverty in human history, and the decline curve of infant mortality. If you stretch all of human history out and mark the time when joint-stock companies were invented, he said the shape of the curves would clearly be different after that point. He said this is a wildly extraordinary, almost outlandish display of capitalism in human society.

I think this part really opened my perspective. When we talk about startups, we usually talk about products, fundraising, and growth. We rarely step back and think that the company as an organizational form is itself a technology invention that ties together the interests of a whole bunch of people. Viewed this way, entrepreneurship is basically using that invention—and then stacking your own innovations on top of it.

Trust only a few things; keep everything else flexible

When it came to long-term planning, Altman said he doesn’t really like planning by working backward from the future to the present. His more familiar approach is to first identify a few directions he believes in deeply, and then move forward step by step from the current time—thinking through what can be done now, what can be done this year. Only in rare cases does he plan five or ten years into the future. He said he has seen too many people hold onto a bunch of beliefs about the future, only to have their own rigid worldview trap them. You’ll see some rocket companies suddenly pivot to AI—that’s the kind of situation he meant. The truly useful approach is to cling to only a few things you believe deeply, keep everything else flexible, and firmly protect the core.

He mentioned a friend’s company’s core values called “critical path.” The idea is to always focus on the biggest bottleneck right in front of you, remove it, then find the next one and remove it—repeat that process. He said that he has been very clear about it over the years: his critical path in life is to make intelligence abundant. As long as no weird concentration of power happens, he believes it will lead to massive prosperity. He said he is rarely tempted by other ideas and rarely wonders whether he should switch goals. But recently, for the first time, he has been seriously thinking about what comes next. If they’re really about to build superintelligence, then what should the next step be?

Reading this moved me. For someone to keep pushing forward for years by focusing on one critical path, with almost no distractions split by other opportunities, is harder than any planning framework. In the end, planning isn’t about being accurate in your calculations—it’s whether you can, over the long term, believe only in a few things and filter out all the noise.

Would you dare to fly a plane at the edge of the envelope?

Altman mentioned a principle he has always believed in: when you’re in a somewhat risky edge-of-the-limit state, get on the plane. He told a story from the time right after ChatGPT was released. Leaders around the world were very tense then—some people were wondering if it was about to get out of control. He could feel a storm gathering. So he took Brian Chesky’s (Airbnb co-founder) advice and decided to do intense, short-term overseas travel. Chesky himself had previously done an itinerary of about the same scale—around eight cities—but they went directly to 28 countries in 35 days. He said he was basically living on airplanes. The experience felt very strange. Even though the rides were comfortable, traveling itself is still exhausting—jet lag, missing his own bed, missing his own office.

He also mentioned an interesting criterion for distinguishing a true trend from a fake one. A fake trend looks like: lots of people are hyping it, but the people who actually buy it use it for a while and then don’t like it. They don’t design their life around it. In the end, the product just gathers dust. He used VR as an example. A true trend looks like: it keeps showing up in your daily life. Like ChatGPT is for him—he uses it almost every day. Sometimes he uses it for three hours in a day, sometimes he barely uses it, but it always remains a continuous part of life. He said this method is something he summarized after seeing many startups at YC (Y Combinator). If you’re willing to spend time analyzing these data, you can really see a lot.

I like this way of judging true versus fake trends because it’s simple enough and doesn’t require looking at complicated growth curves. You just ask one question: has it quietly been embedded into the rhythm of your daily life already, or does it just excite you for a while and then get thrown aside?

Ask for what you want out loud; sometimes it really can get you the impossible

Altman talked about Codex (OpenAI’s programming agent application) as an example. He said this was one of the most memorable experiences of him proactively asking for something. At that time, they were already clearly behind in the programming space compared with Claude Code (Anthropic’s programming agent product). In normal circumstances, trying to overturn a category where someone else already had the first-mover advantage would be considered basically impossible. Most people would concede, then turn to the next direction. But they thought this was too important to give up easily, so they found a team and assigned them what was essentially a near-suicidal task. As a result, the team achieved a rare feat in business history. Among the best programmers around him today, the programming tool they use the most is this product.

He said it very directly: if they hadn’t proactively asked for it and handed that nearly impossible task to the team, none of this would have happened. His explanation was that programming is too important for RSI (recursive self-improvement), not to mention the economic value behind it. They couldn’t persuade themselves to abandon that track.

When I read this, I thought: asking for something proactively sounds simple, but in practice it’s really fighting against a very strong default in society—everyone assumes the winner is already decided, and going to claim the remaining spot is destined to lose. What Altman is doing here is refusing to accept that default: first assuming it might be possible, and then trying.

Projects to kill; team emotion to reset

In the interview, they discussed a painfully real question: when a project has already been funded for over a year, has spent a lot of money, a lot of compute, and a lot of people’s hard work—and the users are using it and they like it—how do you make the decision to pull the plug? Altman said this isn’t something decided by holding one meeting. It’s more like a gradually accumulating realization: you start to discover that this compute, these people, and this product direction can create more value elsewhere—then you have to make that very painful decision.

He gave two examples. When GPT-3 (OpenAI’s early language model) truly got to a working point, they shut down the robot project that was exciting at the time and consolidated resources into GPT-3. More recently, after programming agents truly worked, they shut down Sora (OpenAI’s video generation product) and the browser—two other directions that were also seen as promising—and went all-in on programming. He said this doesn’t mean Sora wasn’t good. If they had kept investing, it could have succeeded. It was simply that at that time, putting compute and energy into programming agents was more important.

On how to get the team to accept this kind of pivot, he said people actually understand the mission and the trade-offs behind it. Even if it’s hard in the moment, the team internally knows why they’re doing it. Some people might be unhappy, but more people would say: “I understand why we’re doing this. It’s the right choice for the mission.”

I think the hardest part of this isn’t the decision itself, but how to make a group of people who have already poured a year of effort into believing that the next thing is also worth going all in on. That requires not just judgment, but also strong communication and leadership.

Focus on what you’re good at; the rest is finding the right people

When talking about becoming better at one thing, Altman used Johnny Ive (Apple’s former Chief Design Officer) as an example. He said the most important lesson he learned from Johnny was that truly great design is more about digging deeply into the problem itself than about suddenly coming up with a solution out of a flash of inspiration. If you rush toward answers too quickly, or if you lock yourself too early into a particular solution, what you build usually won’t be very good.

He also admitted very plainly that he isn’t good at product work. He didn’t agree with the idea that people should only be hired for domains you personally understand deeply. He said he also doesn’t really understand design—but if you talk to Johnny for thirty minutes, you can tell that this person is genuinely excellent. His principle is: rather than forcing yourself to patch weaknesses you weren’t born good at, it’s better to put all your energy into what you’re naturally strong at and make your strengths even stronger through hard work.

There’s also a fairly personal detail. He mentioned that while working on Sora, in order to understand product experience, he deliberately let himself get addicted to TikTok. At first it was just to learn, but later he realized he truly liked it. It went from the first ten minutes before bed to an hour, and then to three hours on Saturday afternoon, sprawled on the couch. He said the feeling was great in the moment, but he obviously knew it wasn’t good for him. Later he turned off most notification prompts for the apps he used— even turned off notifications for message-type apps—and he deleted TikTok. He said he felt it was too powerful for him and he couldn’t control it.

Reading this surprised me. Someone who builds stronger and stronger AI products every day would still get harmed by these kinds of things he himself created—and he had to rely on the most primitive method: turning off notifications and deleting the app to pull himself back. This reminds me of something: judgment isn’t a one-time thing you build once and then you’re done with forever. It requires continuous self-management. Even for people who understand these product design logics extremely well, it’s the same.

My own reflections

After reading the entire interview, what stood out to me wasn’t a specific methodology. It was a mindset toward uncertainty. Trust that exponential growth will continue upward, endure chaos until it’s no longer scary, keep only a few things you believe deeply, speak up when it’s time to ask, let go when it’s time to let go, and put your energy where you’re truly good. None of these lessons are new when taken individually. The hard part is doing all of them at the same time in an environment where every assumption is being overturned.

Altman ended with a line that left an impression on me. He said that most startups today still look about the same as they did ten years ago, because the so-called “correct way” of doing things teaches them to. At most, they change the wording: “We hire fewer people and spend more money on tokens.” But that’s far from enough. I think that line is a reminder for everyone: the tools have completely changed. If your way of thinking is still stuck in the old coordinate system, no matter how radical your talk is, what you build will most likely still be old.