Altman's latest interview: Don't expect AI to usher in a 4-hour workweek.

A 2-week-old startup completes a full year's workload with AI.

July 28, 2026, Zhidx News: On July 26, US YouTube content creator Relentless released an exclusive interview with Sam Altman, co-founder and CEO of OpenAI, the US leading large language model developer. Altman stated that the AI industry is entering its third phase following chatbots and coding agents: persistent AI agents.

Altman revealed that OpenAI is redirecting more computing resources, talent and product development efforts to high-priority areas such as coding agents, and has deprioritized some previously successful projects, including its earlier text-to-video model Sora. He explained that when a given direction loses its relative importance compared to new strategic opportunities, OpenAI will reallocate resources to areas that deliver greater long-term value.

Altman also shared a personal anecdote: while refining the Sora short video product, he intended to study the TikTok product mechanism, but ended up getting addicted and scrolling nonstop for 3 hours, before eventually deleting the app entirely.

This interview reviews OpenAI's evolution from a research laboratory to a global AI enterprise. Altman recalled that when ChatGPT hit 1 million users in just five days, he truly realized that OpenAI was on the cusp of becoming a rapidly scaling large organization.

Looking back on the company's development, he noted that OpenAI's progress between 2016 and 2018 was largely driven by luck. In the 2019 and 2020 period, OpenAI logically should not have survived, as Google should have leveraged its first-mover advantage to ultimately win the AI race.

When discussing the future development of AI, Altman admitted that the biggest physical bottleneck for OpenAI to achieve "Abundant Intelligence" is transistors and electricity. He acknowledged that he had drastically underestimated the demand for computing resources in the AI era. Altman believes that even as technology continues to advance, the vision of a 4-hour workweek has never materialized, and in a superintelligent world, people will be far busier than they are today.

In addition, Altman pointed out that most current AI startups are trying to use AI to pursue "easy wins" instead of leveraging new tools to challenge disruptive ideas. The interview also covers why OpenAI is betting on Codex to compete against Claude Code, as well as the envisioned automated infrastructure for "intelligence can produce more intelligence" in the future.

At the Relentless interview site, Sam Altman (left), host Ti Morse (right) (Source: YouTube)

Core key takeaways from the interview are as follows:

1. AI is spawning the "third wave": The development of AI products is divided into three phases: the first phase is chatbots such as ChatGPT, the second phase is coding agents such as Codex, and the third phase is the upcoming persistent AI agents that can work autonomously for long periods, such as AI employees and AI colleagues.

2. OpenAI deprioritized Sora to focus on Coding Agents: He re-explained the decision to move away from Sora, not because Sora was underperforming, but because Coding Agents hold higher strategic importance. This type of decision has occurred multiple times in OpenAI's history: the robotics project was put on hold when GPT-3 was launched, and Sora was deprioritized when Coding Agents began to take off.

3. AI authoritarianism is the biggest current risk: The core struggle in the AI era is "AI authoritarianism vs. freedom". He warned that extremely negative outcomes will arise if a small group of people or companies attempt to control AI. Throughout human history, every instance of trading freedom for security has resulted in a net loss in the long run.

4. The two major bottlenecks for OpenAI are transistors and electricity: The top physical bottleneck to achieving "Abundant Intelligence" is first transistors, i.e. chip manufacturing capacity, followed by energy supply.

5. OpenAI underestimated computing demand, future AI competition will enter the infrastructure era: The importance of computing investment was severely underestimated in the past. Future AI development requires not only better algorithms, but also more data centers, larger-scale energy supplies and a complete industrial supply chain.

6. Confirmation that we are in the AI singularity era: Humanity has now entered the singularity stage. On one hand, this is a continuously accelerating exponential growth curve with no single defining inflection point; on the other hand, humanity is at another critical decision window where the development curve could branch off in entirely different directions.

7. The biggest misconception for AI startups is only pursuing "easy wins": Most entrepreneurs are only using AI to chase "easy wins" (such as vertical enterprise AI Agents) instead of leveraging brand new tools to challenge truly disruptive ideas. He encourages founders to plan for outcomes that can only be realized 2 to 4 years from now, and to trust that the Scaling Law will continue to deliver results.

The full transcript of the interview is below (edited by Zhidx for better readability without altering the original meaning):

01. Powered by AI, a 2-week-old startup can deliver a full year of work

Morse: Today I'm sitting down with Sam Altman, co-founder of OpenAI. A long time ago, you gave a speech at Stanford about how to start a company. It's been almost 10 years since that original talk. What's the biggest change?

Altman: It's obviously AI. The scope of what a small team can achieve, and the speed at which they can deliver those results, has been completely transformed. Not only has the definition of "what is possible" changed, but also "what you need to do to stay competitive" has shifted, because the entire world has undergone massive upheaval.

I find it absolutely staggering what a 10-week-old startup can build today. If you look back 10 years ago, a 10-week-old startup would almost universally be considered to be in a precarious, struggling position.

Morse: If you had to name the best example of a 10-week-old startup founded today that most intuitively demonstrates this "lightning-speed progress", which one would it be?

Altman: I don't even know their names, but I met one roughly 2-week-old startup that built a full office productivity suite from scratch. The entire product is built for a world where AI operates as a first-class user, with free access to edit documents, presentations, spreadsheets and all other types of files. I can't imagine that this would have been anything less than a full year of work for a startup not long ago.

Morse: I would assume that as barriers to entry for startups lower and available tools increase, theoretically, harder startup ideas should become easier to launch.

Altman: But the very definition of "what counts as a hard startup" is changing at breakneck speed. I don't claim to have a perfect mental model that can identify exactly what the truly difficult, high-value tasks are over the multi-year lifecycle of building a wildly successful company.

I hear a lot of people say that anything related to the physical world is now becoming exceptionally valuable, because software will become commoditized. But as you know, it won't be long before robotics also becomes extremely capable, and many of those assumptions will shift. Even though building rockets is incredibly hard, that landscape could also change dramatically in unexpected ways.

I love this era. I believe startups hold the greatest advantage when change is most intense, costs are falling rapidly, and iteration cycles are shortening quickly. That is the exact moment when startups have massive inherent advantages. And this kind of change is happening simultaneously across many sectors right now, so this seems like an absolutely perfect time to build a startup.

However, most startups are still saying "I'm going to build an AI Agent for vertical enterprise use case X". That will work in many cases, but the competition will be extremely fierce, and it will likely not become the defining, most successful startup of this era — though it can still be a solid business.

But given how drastically the entire landscape has shifted, I'm surprised that more people aren't pursuing ideas like "I'm going to use this brand new set of tools to tackle completely audacious problems", and truly internalizing the fact that the Scaling Law will continue to work, and planning for things that are impossible or uneconomical today, but could be feasible 2 or 4 years from now.

So right now is a great time to start a company, the soil is extremely fertile. But at the same time, there is huge temptation (I don't judge people for choosing this path) to just apply today's Agents to grab those "easy wins", and I fully understand that.

02. Make AI abundant, affordable, and prevent AI authoritarianism

Morse: One thing you've been saying for years is that every time you meet a new person, you try to place them in your internal mental model, and the next time you see them, you want to see how much or how fast they've progressed.

Has this observation across thousands of people shaped your ability to evaluate AI models — for example, seeing that a model is at one intelligence level today, and at a different level 3 months later, so you can map that curve and more easily trust that this growth trajectory will continue?

Altman: I think there's a common core thread running through both: I've developed enormous trust in exponential growth, whether it applies to people, companies, or models. I don't think experiencing model progress feels subjectively identical to watching founders grow, but they share the exact same underlying worldview.

If I was still giving advice to startup founders today, this is the most important concept I would try to get them to understand. And it's obviously hard to do, for the exact same reason that there is still "free money" left in the market for people who bet on high-growth young founders.

The market hasn't fully adapted to this reality yet. I don't think the market has fully internalized the fact that exponential model progress will keep going. You can start building things right now that require smarter or cheaper models that don't even exist yet.

Morse: How did you get so good at navigating this constant, chaotic change? How did you learn to operate effectively in chaotic environments?

Altman: I think you can only learn this through practice. There are things that no matter how thoroughly you understand them intellectually, you need a huge amount of real-world experience to be able to emotionally withstand them. Operating in the middle of massive chaos, trusting that you can solve problems, that everything will turn out okay, that it won't kill you, even when you don't yet know the solution.

That, to me, is something that can only be learned through firsthand experience. I think that's actually a real weakness for young founders, because they don't have enough professional experience to reach that state of emotional calm in the face of this kind of pressure.

So it will be extremely difficult for them in the early stages, and eventually they will learn, but often at the cost of enormous pain, unnecessary mistakes and wasted resources. If you end up in a high-pressure, high-impact role, you will definitely learn how to navigate chaos and operate well. At least most people do, but I don't think this can be taught, I think you can only learn it.

Morse: You said something interesting a long time ago: the first time you go through an event that "could kill your company", it feels like the whole world is collapsing, but once you survive it, by the 10th time, it's not nearly as bad, and you think "I survived the previous nine, this one probably won't be that bad".

A year or two ago, you also shared another idea: things will always go wrong, so you have to internalize an enjoyment of going through painful or bad experiences. How did you make that shift?

Altman: I was just thinking about what you mentioned earlier, when I was doing Office Hours (one-on-one consultation sessions) at Y Combinator (YC), I could always tell which founders were new, not because I knew how new their companies were, but from their emotional state when they talked about problems. I could easily tell if someone was a founder from a recent YC batch, or someone who had been running their company for several years.

I think most people see the opposite of "bad experiences" as "good experiences", and they would much rather have good experiences because they seem more fun or more pleasant. But if you think about it, the real opposite of a bad experience is "no experience". At some point in the future, you will reach a state where you have no new experiences left, and at that point, you will be grateful for all the bad experiences you had.

Naval Ravikant said a line I really love: If you had a remote control that could fast-forward through your life, your life would be over. So the boring parts, the bad parts, are infinitely better than no experience at all. They're all part of the fun, emotional depth and richness of life. I find it very easy to be grateful for the hard days.

Morse: When you think about hard problems, what motivates people to solve difficult challenges? How does the company internally decide which problems to tackle, when to tackle them, when to focus on core competencies, and when to expand scope?

Altman: I think a clear mission paired with deep understanding of the problem is a great guide for what you should do. You won't get everything right, sometimes you'll overextend, sometimes you won't be ambitious enough.

But we are extremely focused on the fact that this technology will massively empower people. It will have twists and turns, but it will be wonderful. It is extremely important to us that power in the world becomes more decentralized and distributed.

In fact, one of the biggest AI risks I'm worried about right now is AI authoritarianism. A small group of people or companies decide they need to control the whole world. That would be incredibly bad. But guided by that mission, we feel we need to figure out how to make AI extremely abundant, extremely affordable, extremely powerful, put it in everyone's hands, so that everyone has access to massive amounts of AI.

When we look at the barriers to that goal, we see a whole new set of constraints: chips, energy, data centers, robotics, and all the pieces that need to come together to build this platform, so we can deliver on that mission.

You can build countless things on top of that platform, every vertical use case, every new startup. We have no interest in doing that ourselves. I believe a decentralized economy is important and beneficial, and we want to be exceptional at producing these intelligence units to make the world a better place.

We will embed intelligence into every product and service. That gives us very clear direction on what we "must do". Interestingly, many of the key inputs to delivering extremely abundant, high-quality, low-cost intelligence, like energy and robotics, are also exactly the things you would want in a world with abundant intelligence.

If ideas become plentiful and all good things are available, but we still live in the physical world, we still want things to happen in the physical realm. So we need the ability to make things happen in that world.

I keep thinking about what deep truth this reveals: that energy and robotics are so critical to sustaining this growth of intelligent infrastructure, and they are also the very first things you need immediately once you achieve that abundance. Or maybe it's just the most boring, obvious fact that to produce anything, including intelligence, you need to manipulate physical matter.

But it's interesting that the work we're doing right now is so dependent on these new domains, and the demand for them will be even greater once we get there.

03. Transistors and electricity are capping AI's continued expansion

Morse: What do you see as your biggest bottleneck? If you say "we want this to keep scaling indefinitely", what's the biggest bottleneck standing in your way?

Altman: First transistors, then electricity.

Morse: For example, people like Jensen Huang are extremely good at aligning all his suppliers around his vision for the future. I think you're very good at that too. How do you think about not just keeping Open