Former Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe Coding

Spent a long time at Intel.

Yeah.

Only 34 years.

34 years.

Yeah. Probably one of the greatest American companies ever, and then it absolutely went off the rails and got absolutely demolished by NVIDIA, TSMC, and, I guess, Apple to a certain extent. You had this incredible Intel Inside moment. We bought our computers based on, “Hey, the Pentium,” and that sound.

Intel Inside, baby.

Intel Inside. Dum dum dum.

The dum dum dum.

So let’s talk about how things went wrong, what went right, and then how it—and you were there for a long time, you took a break, and then you came back. But there seem to have been some critical mistakes that we can learn from, so let’s just embrace it and go right into it. Tremendous success in an American company coming back now, I think, reasonably. But when we look back on it and do our postmortem, what were the mistakes, and what would we change in terms of the direction of that company?

Having spent so much of my life there, I view it—I joined when I was 18. I went through puberty at Intel, right? That’s my joke. I was so early. Grove, Noyce, Barrett—they were the people I grew up with. They were my mentors, the people I adored, and they were deeply technical.

Andy Grove.

Andy Grove, Gordon Moore, Bob Noyce, the co-inventor—you know, these were deeply technical leaders. I remember when I joined the executive staff for the first time. There were probably 15 of the 20 people in the room who were PhDs. It was just that technical. One of the things that went off the rails was when it started to be run by business people instead of technical people.

Yeah, bean counters, the finance people.

Yeah. When I became CEO in 2001, I was the first technical leader in essentially 15 years. If you have a business leader, who does he promote? Business leaders. I think one of the fundamental things is that, as you look at the great technology companies today, they’re deeply technical.

And founder-led, typically.

Even if they’re not—Satya is not a founder.

No.

Sundar is not a founder as well, but they’re deeply technical individuals. When you’re making these hardcore technical decisions that affect billions of dollars, you don’t do that through a spreadsheet.

Right. That’s a lousy investment.

Right, unless the technology trends make it the right investment. I think that’s one of the fundamental things. In the 5 or 6 years before I came back, Intel gave $100 billion to shareholders.

Oh, the dividends and stock buybacks.

Yeah, $100 billion. What I wouldn’t have done for another $100 billion on the—

Well, I mean, what would you have done? You probably would have made chips for the iPhone, which Intel passed on, yeah?

Yeah. But it hadn’t built a new factory in a decade when I got there. How can you not be building? How could you not buy EUV machines? There are all of these things that you would only do as a technologist because the economics behind them by themselves were not good.

Getting back to the core of technology, to me, was the fundamental thing. You make good decisions and bad decisions as leaders. Every business does that as they go along. But fundamentally, this is a technology business, and you need technologists running technology, who then hire technologists onto the staff, who then hire the best technologists.

And take big swings at categories that could matter in the future, like skating to where the puck’s going. If you look at Apple, they’ve done the same thing for the past 15 years: buying back stock and paying tremendous amounts of dividends. They’re the largest holder of capital of any company, I believe, to this date.

What companies do they buy? They buy little, tiny acquisitions on the margins. I think the largest one was Beats, because they wanted to get inroads into certain demographic segments, like in the Android space, that they couldn’t get into. My God, what a colossal waste of time. Like you said, they could have done so many amazing things.

Tell me about Steve Jobs in 2008 or 2009 deciding, “I think we’re going to make our own silicon,” and the impact of that. Was that a covert product project? Did you guys know he was doing that? Did he inform you? That seemed to be another one of those forks in the road, yeah?

Steve was an incredible leader. He was also a ruthless leader, right? Very difficult. Read Walter Isaacson’s book, Steve Jobs, as well. I had many conversations with Steve over the years.

When they moved to Intel with the Centrino chip, it was a big deal. They were putting extraordinary demands on Intel: make the chip smaller, drive lower power. They were a demanding customer. When he was no longer convinced that we could continue to do that, he started the project.

If you remember, what was it? PA Semi. They acquired some small company and started to build some competency. They did a few little chips internally. It wasn’t a big deal, and then the little chips got a little bit bigger. Steve was a master of this—starting small efforts to build core competence inside the company.

I remember when we had the first conversation with Steve about porting the operating system to the Intel chip from the PowerPC chip that they were running on before they moved to Intel. We were quite proud of the silicon and software competencies that we had in compilers and operating systems. “Steve, we’ll help you port the operating system to x86.”

I remember that Steve said, “I’ve been working on that for the last 4 releases.”

He had been preparing the core technologies inside Apple for something that might happen in the future. I remember, I was just shocked: “I ported the last 4 releases to x86. I think we’ve got this.”

That’s how they got into semiconductors and started doing their own semiconductor design. Steve was thinking, “I’m not sure I can rely on Intel to be that much ahead of the industry. I can start optimizing the system design with the silicon design, as opposed to relying on one that’s been somewhat optimized for a Windows environment versus an iOS environment.”

It was just that kind of thing they used to say: “You failed as a supplier.” No, I can supply myself better.

Yeah, and Jensen decides he’s going to go all in on making these video cards. Talk about just incredible serendipity that these happened to be also very applicable for cryptocurrency and running these AI jobs. Was that luck or skill, or a combination of both?

When you think about that progression, Jensen was just building high-performance computers—throughput machines. When we were at the heights of our strength on CPUs at Intel, we sort of scoffed at his machines.

Yeah.

It was a graphics machine. Who cares? There were some gamers who wanted to use that kind of stuff. It was always the big CPU and those little GPUs. But when they started to build a real software stack with it, it was sort of, “Okay, this CUDA thing and SIMT as a technology, multithreading, and so on,” and it just kept getting a little bit better and a little bit better.

It was a little bit Jobs-like in that way. They were just making it better every release, and it was becoming more robust. All of a sudden, the crazy Japanese HPC guys said, “Hey, we could take those graphics cards and maybe start using them in HPC.”

That was a defining moment where it wasn’t just about doing graphics anymore. This was a more computationally dense platform to start attacking some of the world’s most interesting workloads. I think Jensen would agree that was a defining moment, saying, “These aren’t just graphics cards anymore. These are general-purpose computing devices that can start applying to these other workloads.”

AI had gone through what—its 5th nuclear winter by that point? It was, “Man, this is never going to matter. We’re never going to get the breakthroughs.” But the community around it was continuing to develop.

Yeah.

The CUDA software kept getting better, generation by generation. I had a project at Intel, Larrabee, where we were trying to take the x86 and essentially do the same thing. In my first departure from Intel, the project was killed a week after I left.

Huh.

And the world would have been so much different.

I think it's illustrative of continuous innovation, taking some risks, doing that fundamental research, and the compounding power of technology. I think it was William Gibson who said, “The street finds its own use for technology.”

Nvidia did not predict that this Bitcoin project would take over and that this would be the best way to do those computations. Nor did they anticipate that AI would take off. But because it was the best solution, the hacker community could—

Yeah.

—figure that out.

Well, as we wrap up the Intel portion of your career, Apple silicon is one. Then you have Nvidia. And then you have this Taiwanese company that starts becoming really great at fabricating these chips. Intel missed that as well, right?

Maybe you could talk a little bit about TSMC and its surge, and we can even get into some of the politics of it now. Then we'll get into some of these AI chips and venture investing.

The thing with TSMC was that they started with a vision of a foundry.

Mhm.

They were going to become the factory for the industry. Again, these factories are so expensive—$20 billion, $30 billion—and there is so much engineering and continuous investment required to do it. It was a stunning vision at that point in time.

Intel was an IDM, as we called it: an integrated device manufacturer. We never worked to make our process and our factories available for third parties.

Mhm.

It was always this thing: “Hey, we do enough CPUs ourselves. We reuse it for chipsets and some of the other things that we're doing.” But it was never standardized in a way that it could be made available for a broad ecosystem using PDKs and all the design tools.

We did a lot of our own EDA tools ourselves. One of the projects that I started early in my career was the foundation of EDA as well: the first place-and-route, the first standard cells, and the first hardware description language. It was so proprietary.

TSMC basically cut that in half and said, “I don't care whose chip it is. I don't care what you're designing. I'll be your manufacturing partner.”

Yeah.

At the time, that was such a trivial piece of the business that Intel didn't even care.

Mhm.

Over steady progress over a long period of time, with Apple as a customer driving them to become good, it became really meaningful. Obviously, the world changed.

Mhm.

When I came back to Intel in 2001, TSMC was producing 5× the wafers of Intel.

Wow.

Not 10% more—5×. All of a sudden, that foundry model became the model of the semiconductor industry, with 2 exceptions: Intel and memory.

Memory companies design and manufacture, which is uniquely different. Obviously, we're seeing $3 trillion memory companies—just extraordinary—and a trillion-dollar foundry company in TSMC.

The industry has said, “I want a lot of wafers. I want a lot of innovation across different designs. I want a layer of standardization and EDA tools.” The world changed. As I came back to Intel, that was one of the core theses of the new strategy: We must become a foundry as well.

Five to 1, and now it's more like 7 to 1 in terms of wafers, TSMC to Intel.

Are we going to be able to onshore that? Obviously, we had the CHIPS Act. Give us the broad strokes of what you think is going to happen here. Taiwan is obviously in play.

Some people in the administration believe it's going to happen the year after Trump is out, unless he takes his 3rd term. Other people believe it could happen as early as '27 or maybe going into '28.

Are we going to be able to replicate that here in America in a reasonable amount of time, or could this truly be a cataclysmic event if, God forbid, China decides, “Hey, we're going to blockade Taiwan,” and then the Taiwanese decide, “Yeah, we're going to burn the fabs and fly out all of the engineers and ship them to America?”

Mhm.

Well, there's a lot in that question. Do we have an hour to talk about this question?

Well, I mean, we have 6 minutes.

Oh, okay.

Yeah, do the best you can.

Okay.

I also want to talk about the AI bubble.

Three things about this, super quick. The first is that the CHIPS Act is having a benefit.

Yeah.

When we started the CHIPS Act, and when I came back, the U.S. was building about 12% of leading-edge chips. Today, that number is more like 18%.

Mhm.

We're making progress.

It's not 50%.

Mhm. We have a long way to go. Intel is starting to be a real foundry.

Okay, that's real progress.

TSMC's factories are up and operating at scale. We have Samsung as well, but I'd say the Intel and TSMC progress is meaningful.

Now, let's make it ugly for a second. The island of Taiwan has less than 3 weeks of energy reserves. There was a big article in The Wall Street Journal 2 weeks ago on this. That should just put a chill in everybody's spine.

Wow.

Because after 3 weeks of blockade, the island browns out. When you turn off a fab, it doesn't come back on for 90 days. The economic impact of a brownout in Taiwan is greater than the Great Depression in the world.

You never need to do anything or fire a shot. You just need to say, “Great, no energy for 3 weeks.”

No oil. Yes.

Right, right. No LNG. That's how the island runs. That is scary to me. We need more resilient supply chains associated with it.

I don't think this is an alternative for the world because, if it really does become a risk—and I don't sit in the situation room and get all the data and so on—let's remind each other that I think China has blockaded the Taiwan Strait 7 times over the last 4 years.

Yep.

This isn't a theory.

No, no, they're running exercises. They're being pernicious and—

Right.

—pretty provocative in terms—

Is that 2027? Is that 2030? Is that 2035? Their intentions have been clear over a sustained period of time. We need more resilient supply chains.

That's something I put a lot of my time and energy into. We're making progress, but we need to go faster and make it more meaningful.

Yeah. Let's talk a little bit about the AI build-out. You watched the PC revolution, servers, and the internet. These were all extraordinary build-outs, and now this is the build-out to end all build-outs: the amount of data centers, the amount of chips, and the amount of inference needed.

Do you think it's a bubble? I think I've heard you say that it's obviously a bubble, but what's the risk factor here? That we build too much, or that the technology doesn't solve enough problems and we're swimming in tokens?

What worries you about what you're seeing now? The valuations of these companies have gotten quite extraordinary. If they build too much and spend too much money and don't make enough money, based on your experience with running a public company, that's a lot of tension on it. When you don't make as much money as you're spending, people tend to fall out of love with these stocks, right?

I do think there is a silver lining here that guarantees we don't get too far ahead of ourselves in terms of a bubble, and that is energy capacity.

Right.

Energy capacity in the world is expanding at 5%. In the U.S., we had a decade at 1%. It's just hideous what we did to our energy grid over about a decade and a half. But now that's getting built out.

Essentially, nobody's going to build and buy GPUs or build data centers if they don't have energy. So you have an upper bound on how aggressive, hyped, and bubbled we get. I take a lot of solace in that.

What, then, is the incremental value of a token? If it's a measure of intelligence, it's somewhat infinite. If I have more intelligence, I will have a better supply chain. I will do better finance. I will have more efficient logistics—all of those things.

To me, the potential value that we unleash in a token economic world is somewhat infinite, particularly with the labor shortages that we see in developed countries. I am an optimist that we're in a couple-decade build-out.

Wow.

Not a couple of years—a couple of decades. One of the big objectives I've said is that I have to make AI 10,000× better.

Mhm.

Right? It’s way too expensive today. We want to drop by 5 orders of magnitude the cost per token and the energy per token, so that we really do have Jevons’ paradox, where we just explode access to AI in much more economical ways.

It does seem like Jevons’ paradox has been at play over the last year. These tokens are so cheap and the tools are getting so good. I’m just going to start using these tools all day long until the bill comes in, and you’re like, “Okay, yeah, maybe I need to get some ROI out of this.” But you do have these incredible companies—Cerebras, Groq, et cetera—making inference.

And silicon and so on. If we accomplish these orders-of-magnitude improvements in token economics and availability, along with a reduction in the energy costs associated with it, we just have a fantastic couple of decades in front of us.

There has not been a time in human history when it’s been better to be a technologist than the one we’re in right now. We will solve chemistry. We will solve language. We will invent new materials and new forms of interaction, kill cancer, and lift people out of poverty. There is not a better time to be alive than the one we’re in right now. As technologists, we get to sit in the driver’s seat of it.

Pretty amazing. You’re investing now, and that’s your passion. What do you think of these valuations? It seems quite different if you lived through the dot-com bubble. We did see a disconnect there. These companies are slightly different. We just had ElevenLabs at $600 million in revenue. Lovable, I think they’re at $500 or $600 million. So that’s quite different from the dot-com speculation, yeah?

Yeah. Fundamentally, we have real revenues and real margins coming out of these businesses as well. That said, anytime the multiples get too high, we get some corrections. To me, periodic corrections that keep the earnings multiples and so on at reasonable levels are good, because this will not be a smooth curve.

I’m predicting 2 decades of goodness, and there are going to be lots of disruptions along the way. It’s not going to be a smooth curve. Every time we have one of those corrections, say thank you. We’re not letting the bubble get ahead of itself.

We had the SaaS apocalypse. There are going to be other apocalypses on that journey when industries get impacted by the capabilities that will be unleashed. That’s even before it gets exciting with what I call the trinity of computing: classical computing, AI computing, and quantum computing. When those 3 come together, that’s when things get really exciting.

Hey, quantum has been about 5 years away for 25 years. When is it actually going to do anything?

This decade.

This decade, so by 2030?

Yep.

It’ll be meaningful. What should we expect in terms of its impact in 2030?

You’re going to be able to start doing things that cannot be computed today. Chemistry and biology—there will be things that can’t be computed today. Some of the easy things will be logistics, where I will compute the best answer to get this thing to you.

Traveling salesman problem?

Right. All of a sudden, all of those problems. Obviously, it’s probably going to be 2032 or 2033 when we solve things like encryption. You’ll have the fundamental Q-day implications, but this decade we will see quantum-supremacy results across multiple industries.

We know how to build qubits. We know how to error-correct qubits. We now have algorithms for quantum. Now it’s just about engineering scale.

Who’s going to win?

Well, obviously, I’m a PsiQuantum guy, since that’s one of our portfolio companies. But what you’re seeing is that you now have 4, 5, 6 modalities of quantum demonstrating pretty good results across trapped ions, photonic approaches, and spin approaches.

You can now say modality is not an issue. Error correction has been proven across them. I think the race will be on, and my prediction is meaningful results before 2030.

Wow. You realize that’s about 40 months from now. Yeah, okay. Meaningful results. Thanks so much, Pat, for sharing all this incredible information and knowledge. Great to see you.

Very good.

Anton Osika is one of my favorite founders. He’s the founder of Lovable. Why do I love this founder? Well, he built a product that people are addicted to—primarily, Anton, the people who work for me.

I love talking to you because, as the founder, you have a North Star. You’re incredibly laser-focused on enabling anyone to build great software. That’s the mission of the company. I’m paraphrasing here, but essentially that’s the mission of Lovable.

The mission I talk about is empowering humans.

Empowering humans.

The first gap is to build a product. The second gap is to build a business around that product.

Right.

Now, at Lovable, everyone at Lovable is working on both of these 2 gaps. The first one, we’ve gotten very far. We’re seeing 1 million new projects built every single week on the platform.

On the second one, we’re investing a lot in making it easier to run your business, to get people to care, and to help people discover what you build. For the entire business of whatever you’re doing, as a small business or as a large business, we’re also getting a lot of traction.

As proof of that, we’re seeing more than 700 million visits to the applications every month. Every month, there is extreme growth in the surface area of the entire platform, with more than 50 million apps built on the platform to date.

How many years has Lovable been in the market, or how many months now?

20 months since we launched, yeah. Again, we’re seeing people who are first-time founders. We’re seeing enterprise leaders move much faster together with their teams on this platform, which has a lot of opinionated pieces in how you should create software, how to operate that software, and how the different applications in your company connect to each other over time.

That’s why we’re seeing so much growth on the enterprise side, where we’re actually growing fastest right now.

This is really interesting, because 10 years ago people were doing WYSIWYG software. What was the name for it before vibe coding? No-code, low-code. Yes.

When I saw that 10 years ago in my incubator, every 20th company, somebody would come in who was an MBA or not a developer, and they had no-coded something. They were using these different software platforms, and the software didn’t look good. It didn’t work perfectly well. It was slow, but the promise was there.

I guess it took LLMs and this new intelligence to make actually good software. Maybe you could talk a little bit about who the customer is. Do developers use Lovable, or is it the other 95% of society that are your customers? How do you think about who your ideal customer profile is?

Yeah. We’re seeing people use Lovable with both technical and nontechnical backgrounds. About 20% are technical or some type of engineer, and they love that we’re quite opinionated. We put all the best practices into how the software is architected, and we make it seamless to set up payments from the get-go in a very secure way.

We do things like run security scans after every change and monitor the projects in the background. It’s actually quite appreciated by the engineers in the technical community, also because it’s a great bridge to the nontechnical people, who are 4 out of 5 users.

They’re often building first to figure out what is the right thing to build, which is where Lovable has always been exceptional. Now what we’re seeing is that people are running businesses making more than $1 million in revenue on the platform.

So it’s building for everyone. It’s this entire spectrum. What’s exciting to see is that often, if someone discovers Lovable from their colleagues at a large company, they go out and run a side hustle. Some of those side hustles really work. They make hundreds of thousands of dollars, and then they become a founder after that. So this is cross-pollination from both—

Yeah, and this is the really interesting thing about vibe coding. If we were sitting here last year, people would look at it and say, “It’s a great way to make a mock-up.” Like you said, “A great way to think about product and maybe create wireframes or a workable prototype.”

All of that’s out the window now. The whole concept of building wireframes and building a mock-up—well, you can just go right to building the product in a day or 2 days.

What people, I think, don’t appreciate about what you’re doing at Lovable is that after you’ve made a product that you’re proud of and that has some product-market fit, there are many more steps that are required.

You mentioned payments and security, making sure that the data isn't lost or leaked. That's changed dramatically over the last 12 months, yeah?

Very much so. Many engineers don't look at the code. They don't write code anymore. That means that you don't need to be an engineer to create software, right? But what Lovable does for anyone, including nontechnical people, is create a structure for the architecture of the software that you build and make sure that you don't go off a cliff.

Things like setting up payments and emails, getting discovered by other AI chat engines, and getting discovered by Google Search are taken care of. You don't have to know how all these things work in detail. You can trust the platform to take care of data security and connect to other tools that you might be using in a secure way.

That's really where being opinionated from day one and being focused on making this for the 99%—it's a vast market, right?

Right.

Being focused on the 99% from day one is what made us very successful.

Yeah, and I can tell you that internally, I gave my team all the different tools they could possibly want to use. Somebody had started with Lovable. I think I told you the story when you were on This Week in Startups a year ago. They made some interesting websites and were trying to make an intranet, but they couldn't quite get it done.

Then I had some people who started using Cursor or Claude Code. They started vibe-coding stuff, but they couldn't finish the product. Then people tried to solve some problems with Cowork. I really like Perplexity Computer.

Then my team came to me and, for one of our projects—I was talking to you about Founder University, our pre-accelerator—they wanted to make an intranet. This is something I would have never okayed because it would have cost $500,000 10 years ago to make, and we don't have that kind of budget. We'd rather put that toward the founders in the program and getting more people into the program.

In 4 to 8 hours, they made the whole intranet, and they made a bunch of things I hadn't asked for. It was the person running Founder University who made it. She did it on her own, without permission, in Lovable. I said, “Whoa, how did you build this?” She said, “Lovable.” I was like, “Oh, we still have Lovable?” She said, “I just put it on my corporate card.”

To your point, she made it. That software is now driving the program, and the reason people do the program in their country—we have it in Saudi Arabia and Japan—is because it has economic impact.

Yeah.

So I said, “Hey, I have an idea. Can you make for me an economic impact of the 50 companies that are in the program?” She asked Lovable to do it. I gave her some prompting, human prompting, boss-to-employee prompting.

Now it has the economic impact in there, and with our prompting, it considered, “Well, how many people work at each company? What are they paying in taxes? How much do they rent their home for? What is their average salary?” It built something that I would have never been able to afford to build.

Lovable is $50 a month, I think. I don't know how much you charge, but it's far too little.

That's if you're on a business plan. I start at $25.

Yeah. The economic impact of what you're building—I would equate what you built for us to something that would have cost me $500,000 2 years ago. It was built in 4 hours by an employee, which, if you just put employees at $50, $60, whatever, $70 an hour, plus the cost of your software, it got made for less than $2,000 in a year. It's extraordinary.

I'd love to hear more about the progress of the intranet. Anything that you asked for that you want to forward directly to me?

Well, right now, my concern was security and making sure that data didn't leak. They talked to your team, they went through it, and it's secure, so we feel good about it.

Well, look, I'm now asking people who do penetration testing to say, “I want you to compare all the tools and make sure that all the work we're doing that's not visible on security and trust is accounted for.” There's a lot of other things where we invest and spend money, too. Also, free users get a lot of security scanning running in the background that actually translates to something that security experts can see.

A year ago, we were at mockups. Now we're at functionality that's secure and super-viable for deployment. Where will you be in a year?

What we're seeing is that there's a gap in being able to build the product, right? You built an entire intranet on the platform. That's great. What we've done since then is introduce a new product line, basically the hosting part, which includes both the AI and all the normal hosting. That product line has been growing faster than the building thing I mentioned.

AWS competitor.

Let's say it's not. You run all your software, and we're working with companies like AWS and Red Hat as well.

What you also want to have is to use Lovable—we're seeing this with our customers—as an AI co-founder, a partner that you talk to about everything in your business. If you're running your apps and tools on the platform, then just talking to Lovable means it has access to all the data that you might want to know about your company and how it's doing.

We're working with some of our customers in pre-release to give them access to a co-founder that works for you even when you're sleeping and comes back to you in the morning and says, “Here are some strategic directions you could go. Here are some optimizations you could make in terms of growing your business faster and serving your customers better, faster.”

That's the evolution toward operation and intelligence, toward driving outcomes for your business.

So you come and build the software, but you stay to build the business.

Yes, to operate your business. What we're already doing—and I've been doing this for a very long time—is compound everything we're learning. Every time Lovable makes a mistake, it goes through our agentic system, with our engineers improving it. That compounding intelligence is, of course, applicable to our customers and users running their businesses on our platform as well.

Is software going to become 100% bespoke, even the internal tools? I was looking at Slack, and our bill for Slack, even on the highest version, is maybe $10,000 a year. It's not a lot of money. It's well worth it. But I was starting to think, well, maybe I should vibe-code my own Slack so it's integrated into everything we do at a deeper level.

What do you think the future will look like in terms of some of these foundational pieces of software that every startup and every enterprise uses—Salesforce, HubSpot, Slack, the Google Suite, Microsoft Office? Will bespoke software start to replace those? Do you believe?

I like this question. Let me answer you, but I'll just give you a story about someone I recently heard who's going on this journey. They're quite the brand.

Nenad works at a pretty large company in the US, Nursa. He came to our platform because he wanted to build out the new product lines—Nursa Academy, for educating more nurses. He built out all the admin tools for the program, the scheduling for the nurses, getting their licenses, and their certification management.

He was able to build that into a product and take it to market because they had all that access to nurses wanting their certification. What he also did was take it into the back office internally, and they've now replaced more than 10 tools they had with bespoke applications.

I think, in terms of your question, you can do that for multiple reasons. In their case, they're saving more than $1 million per year.

Right.

That's huge, right? But it's also the case that in some cases, you have specific requirements where the tools that you've been using to date aren't suited for those requirements exactly. In those cases, I think, yes, you will have more bespoke solutions.

Yeah.

But I also expect us to see that Lovable continues to interoperate with all of those tools. I'm not sure if you tried this, but if you ask for a connection to anything in the Google Suite, anything in the Microsoft Suite, or Slack, Lovable guides you through all the steps to do that in a way where you can get a very good overview of exactly how the data flows. That's of course very important—you don't want to give the wrong person access to the wrong data.

You can continue to use Salesforce, HubSpot, and all the tools that you like to use under the hood, but with a bespoke interface on top of them.

How have these new frontier models—They're in some ways competitive, but in some ways you can use them to power Lovable. How do you think about the competition with them, open source, and the future of Lovable?

People have announced that Lovable's dead every 6 months since you started, and then every 6 months you go from $100 million to $200 million to $300 million. I think you're at $400 million in revenue, something crazy.

We reached $500 million in May.

Okay.

Yeah, growth is phenomenal.

So you're dying again by another $100 million in annual revenue. But underneath the hood, you're using some of these—

Yeah, let me explain.

Yeah. So, we've always had the strategy that we do whatever is best for our customers. In terms of intelligence, that means we're using multiple models. If you ask Lovable now, it's routed to the model that's most suitable to whatever you want to do, including commercial frontier models from multiple vendors and, increasingly, open-weight models. When our team gets routed to our own model, that model becomes more intelligent for our agent harness.

Yeah.

From multiple vendors.

Yeah.

Especially on the mistakes that it might be making in some cases: which tool to call, which integration to create, and how to guide you through success for your business.

Right. So, you're all in on open source. You believe that's the future of Lovable.

Well, I'm leaning into it. We have multiple partnerships, and we're investing heavily to be close with those partners. They're the big labs, and it's also about making sure that we get the fastest performance at the lowest cost for our customers when we know we can do that with our own models.

Right.

We have a really strong research team up in Stockholm working on what's called post-training. We're applying all the best practices to do that and scaling up that team quite significantly. We also believe it's an important part of the European ecosystem to have that capability in Europe specifically.

Are you doing or using any of the data-labeling or data-training companies to help you understand the most common businesses and build that proprietary data?

What we're doing is looking at the mistakes that any of the models make right now.

Ah.

Then we prioritize them by what drives the most impact for our customers. We create datasets where we do something called reinforcement learning, specifically for the problems where the frontier models are making mistakes for us right now. We have this enormous token distribution from a million new projects being built every single week.

You're burning a lot of tokens.

We are, yes. That's a lot of signal for making the system better: both the agent harness and what we've been refining over the last 2 years, which are the skills that we have—the internal-type skills that tell the agent when to remember facts, based on feedback from our software engineers who know how to build really good software. We're modifying both of those every single week.

It makes total sense. Somebody told me some companies are doing token dumping. They're selling $100 worth of tokens for $50. Basically, they become token resellers in some ways, and they're money-losing businesses. You're money—you’re profitable, I believe, now or close to it.

We always monitor our margins, but again, we're doing what's best for our customers. That often means more intelligence, so we're not looking at, “Oh, let's use a cheaper model here,” if it's measurably worse for our customers. We can measure what's best for our customers.

Is it unlimited for the $50, or do you have caps now?

We have caps.

Yeah, you have to have caps. Are people starting to hit the existing caps?

Yeah, our customers definitely hit caps, and then you can top up. We have multiple subscriptions here.

What percentage of people need to top up? They're so addicted to it that they're blowing past the limits.

I'm hearing that more and more often: people are willing to pay the overages because they're getting so much value. I think that's the future of the business. People are looking at it and saying, “Well, if I'm paying $600 and if you token-max it to $6,000 a year, but this is a $500,000 piece of software, I don't care. I'm still paying somewhere between 0.1% and 1% of what I would have paid 3 years ago. Who cares? Go for it.”

Yeah, what we're seeing is that everything is about moving fast these days. More AI usually lets you move much faster, so this spend is usually worth it.

Hey, do you see your customers—final question for you—because I'm starting to see this now, where multiple people in the organization try to solve the same software problem and compete with each other? For example, this intranet I'm talking about: we built one for Japan.

Yeah.

But somebody built the U.S. one. So now I have 2 pieces of software. I said to the 2 different people, “Did you guys fork each other's code?” They're like, “No, we just built 2 different Lovable projects.” And I'm like, “Is that the right thing to do?” Because you went faster and I had 2 swings at bat—2 different intelligent, brilliant people making their version of the software.

But you would never have done that in the previous way of building software. You would have had one trunk of software, and you would have been building Frankenstein software, trying to get all the needs into it from the 2 different groups.

Yeah, I'm actually a huge fan of very rapid experimentation. For a while, I worked at a place called CERN, where they do particle physics. It's pretty unique, right? That's where I was introduced to this concept of competition, where they have 2 actually quite isolated teams working on the same particle accelerator, but at different places on it. They don't share the results until they publish, and that way they can, over time, learn what's working best in the different organizations. You don't get stuck in a local minimum.

Free markets work extremely well because of competition, and they do that in academia as well. Now, since engineering is less of the bottleneck, it's more a question of what is the right thing to build. I think it's a great thing to have, if you have sufficiently many humans, to attempt to solve the same problem in different ways.

If you do that on Lovable, what I like to do is bring up a new project—or one of the projects—and say, “Hey, can you check out this other one and take these 3 things that I really like and bring them over here? Maybe even run a split test, run an experiment, to see if it's improving the metrics for the customers we're trying to serve?”

Did you see somebody use Fable to build Fortnite?

I think there are some 3D games, yeah.

Yeah. What is your take on this latest version from Anthropic, Fable? I know they're a partner—or I assume they're a partner. I don't know that.

Yeah, we use Fable as one of the models in Lovable.

So, what do you think of it compared to the last generation? Faster, better, both? Is it a massive step function?

What I've seen is that, on the first attempt, it can create very sophisticated things that look really good. As you're evolving, it's still the same thing: as a human, you have to think, and you often have to plan together with your agent about what is the right thing to do. That's more of the bottleneck.

More intelligence is great on some tasks. It creates really beautiful things—3D games, for example—but figuring out what to build, figuring out the right strategic directions or experiments you should run to improve the outcomes for your business, that's not changing as fast. It's humans knowing how to use the tool and plug in all the right data to be able to make the right decisions to take your product and your business forward.

Listen, I love the product, but even more than I love the product and you as a founder, I love the outcome. The outcome for business is extraordinary. Anybody who's listening, Lovable is absolutely worth your time. Don't wait. Just put it on your corporate card and start building. That's my message. Just start building with Lovable. It's an incredible product, and congratulations on being reborn 6 times, because every 6 months you had $100 million in revenue, it seems. Then everybody says Lovable is dead because the new foundation model is so good, but you keep studying your customer and somehow surviving and thriving. So, congratulations as an entrepreneur.

Thank you so much, Jason. I enjoyed that talk. I hope you enjoy the rest of your stay here in Paris.

It's pretty great, and the Palace of Versailles is so impressive, huh? Someday we'll be building this with Lovable and Optimus robots.

I look forward to it.