Andrew NG on The Biggest Bottlenecks in AI

Thank you, Harry. I watched a bunch of shows. I really enjoyed your recent one with my friend Martin Martin Casado as well. That was very memorable, so I’m actually thrilled to be here.

I would say the two biggest bottlenecks right now. I think electricity is one of them. So in The US, I am honestly worried that, many data center operators were stuck in kind of permitting and, you know, and and I know that local community support is important, and some people don’t want a data center there. But, once we build roads and railways as the infrastructure for a certain generation, Data centers are the critical infrastructure for building the digital economy, and so lack of electricity in America and in number of Western countries is a problem.

And in contrast, I see China building power plants left and right, including nuclear. So that will be an interesting dynamic. And then semiconductors is another bottleneck. But AI is so complicated. I think we also need more data. We also need more better algorithms. All of it is worth working on. But in the short term, some constraints with electricity and and and semiconductors.

First, in my career working in AI, I have yet to meet a single AI person that ever felt like they had enough compute. You know, get us any amount of compute, we will use it all up and say we still don’t have enough. So this is a constraint for the last twenty years or so. But what I’m seeing is with the rise of GenAI, there are very valuable workloads. For example, AI assisted coding, you know, is fantastic. It’s making us so much more productive. But if you use cloud code enough, sometimes you get really limited, and I find that many companies really have excess demand, which is a very rare problem to have.

But so many people want more LLM inference, want more tokens generated, and we just don’t have the semiconductors and data centers and electricity to meet the demand. But, you know, there’s a lot we could do with AI token generation, and it’s frustrating when we can’t supply enough to people that want it on the demand side. You know, you you get very limited if you if you use too much.

It is true that token generation is getting more efficient and cheaper. If you look at OpenAI’s Open Wake model, they they actually release models that are very efficient to run. So I think they did a good job. Was it, like, a 120,000,000,000 parameters or something with, I think, 5,700,000,000 active? So it’s a very efficient model to run. But despite the costs of token generation falling, our demand for it is insatiable. One of same thing that’s happened in AI is if we look at where are the buckets of value, one of the big buckets of value is AI assisted coding.

I think this harkens back to an earlier era. In a previous generation, I think Google came to dominate, you know, horizontal information discovery, like web search, but there’s room for lots of verticals when the Internet was being built. So we wound up with Travelocity and Expedia, out for travel, ChatGPT seems to be the dominant player in the new gen horizontal information discovery, although I think Gemini with its channel advantage through control of Android and Chrome, you know, is is a is a serious player as well.

But if that’s where horizontal information turns out to be, then there’s still plenty of room for lots of verticals to be built out. And one of the clear buckets of really valuable verticals is AI coding assistance, where Cloud Code is you know, I use that every day. Love it. OpenAI Codecs has a lot of momentum as well, but it’s clearly making developers so much more productive and efficient that the demand is just through the roof for less less use more and more of this.

One thing I find exciting is I often look at AI coding assistance as a harbinger for what might happen to other job functions as well as the AI marketing tools become more efficient, as the AI recruiting tools become more efficient, the AI finance tools become more efficient. So I often look at AI coding assistance as a maybe a foreshadowing of what may happen as well to other sectors as the tools get better for them too.

I don’t know. I I think it’s further along. I think in 2016, image generation wasn’t super valuable. I I think today, AI coding assistance is really actually, at at AI Fund, my head of engineering received I I would say, hey. Let’s think about standardizing on tools, and, you know, basically, he said, I need these tools and you have to pry them out of my code dead hands. I think our developers feel really strongly. I I myself, I don’t ever wanna have to code again with our AI coding assistant.

So I think the tools are really working well, but still with a lot of headroom for how much better you can get.

Over the last few years, the US federal government has done some good things and some less helpful things. I feel like clearing out unnecessary regulations has been a very good move. Even last year, the, bipartisan Schumer AI Insight Forum, I I I think there are a lot of people lobbying the US government to pass stifling regulations. You know? There are a lot of hyped up AI safety narratives saying AI could lead to human extinction, which is kind of a ridiculous statement. So they try to get stifling anticompetitive regulations passed often to try to shut down open source open weight.

Fortunately, beat back a lot of that, but I think the bipartisan, the shoot me inside forum did a really good job digging the truth and concluding that America should be investing in AI rather than, you know, passing unnecessary regulations to slow it down. I think Trump and then and his whole team, David Saxon, Christian, and so on, did a good job clearing out unnecessary regulations. On the flip side, one of America’s huge competitive advantages has been this ability to attract talent, including high school talent as well as young talent that may not currently be high school but could be high school in the future.

To the extent that America is not investing as much in attracting talent, I think that would be an unforced error. And then I think lastly, investments in science. Right? I think, helping our institutions of higher education have the resources to train our grad students to invest in science and technology, I think that’s really precious. And so anything that damages that, I think, would also be very unfortunate.

America is fortunate to have a lot of very smart people wanting to come here to do really challenging, really tough problems. Many of our Nobel laureates are immigrants. Einstein, for example, was an immigrant. I think continuing to cultivate America as a place to attract great talent, to work together in a democratic nation that that respects the rule of law, I think that would help us move ahead. I think that securing the semiconductor supply chain would be very valuable as well. A lot of friends in Taiwan.

I love Taiwan. And also America’s dependency on TSMC, is concerning in case anything happens. And and and frankly, there’s one very funny thing that happened in society. There was recently a Pew report showing, I think, how much Americans think AI would be good for them, enthusiastic versus non enthusiastic. And even though a lot of AI technologies were invented in America, a lot of people don’t trust or don’t like AI.

In the case of software engineering, it is accelerating the writing of codes. There are so many projects that used to take six engineers half a year to build that today, I or one of my engineers can build in a weekend. I hope that we never have to go back to coding with our AI systems again because the acceleration, the productivity boost is incredible. I mean, for example, one weekend I thought I wanted flashcards for my daughter to practice modification. And she wants to practice modification. She wanted flashcards.

So I thought I could either drive to the store and buy a bunch of flashcards for her, or I could just, you know, use AI to write code for me to generate and print out a bunch of flashcards. And so I did the latter. And so this is a very low economic value task. But with AI assisted coding, if we get that done very quickly

I think we need all of the above. I’ve had mixed dealings about the term vibe coding, but nitpicking terminology aside, think I everyone should learn to code. What I’m seeing is for a lot of job roles that aren’t just software engineering, people that can code can get more done than people that can’t code. For example, I think my marketer wanted to run a user survey. She wanted, you know, something for people to give live feedback. And she looked at the App Store, couldn’t find anything, so she said, you know what?

I’m gonna spend two days to code that. It did take take her two days, but my marketer then built a little mobile app where users could swipe left or right to give feedback on some marketing messages we wanted to use to test. And because of that, we’re able to run user experiments, get feedback, and so it helped her do her job better as a marketer. Whereas in contrast, a marketer that couldn’t encode the blowout app to, you know, let people swipe around and get feedback, they would just not have been able to do this, would not have gotten the feedback, would not have been able to move forward.

Today, my best recruiters, not only do they screen resumes by hand, they are writing prompts to get AI to help them screen resumes.

I think there’s a small subset of jobs that, frankly, are in trouble. But I think for the vast majority of knowledge workers, AI is amazing. There’s a lot of stuff it can’t do. So this phantom AGI someday with AI, they can do everything a human can do, I think we’re very far away from that. I would say, you know, like decades away, maybe even longer. And the trick is if AI could do 30% of a recruiter’s job, maybe 50%, although that feels a little bit high.

There’s another, like, 50 to 70% of stuff that we still need a human to do. But it’s also clear that if you use AI and someone doesn’t, that’s actually a huge difference in what you can accomplish. So much better off using AI. But because AI can’t do everything, there’s still plenty of work that we still need humans to do for a lot of job roles.

I don’t think it’s as dire as that. I think there is a big problem, but I don’t think it’s exactly that problem. So let me tell you what I’m seeing software engineering. The most productive engineers I know, they’re not fresh college grads. They are people of, you know, ten, twenty years of experience or whatever and really on top of AI and know the AI tools and understand AI code. So those people, experienced and on top of AI, move faster than anything the world has seen even one or two years ago.

One tier down is actually a fresh college grads that are really on top of AI. So I’ve hired quite a few people, fresh college grads, that for whatever reason, through their social network community, really learn the AI tools, and they move really fast. But they’re not as good as people experience. One tier down from the fresh college grads is the people with ten years of coding experience, but who had a comfortable job and for whatever reason is still coding like it’s 2022 before ChatGPT. I just don’t hire people like that anymore.

But there are people that, you know, did a comfortable job, they kept coding the old way, and they just did not learn AI. I think those people may get into trouble at some point. But there’s one other one, which is the the tier that is in trouble, which is, the fresh college grads that don’t know AI. One unfortunate thing is university curricular is slow to change. I actually feel pretty bad that even today, there are, you know, universities graduating CS undergrads that have not made a single call to a single API on the Internet.

Imagine graduating a CS undergrad that has never heard of cloud computing. It’s like, what is a cloud? Oh, I don’t need to just run things? That that’s that’s weird. You just can’t be a CS major and not know how to do things in the cloud. And I’m feel like I’m getting to a point where I I feel like we’ve gotta not train CS majors without also making sure they know how to use AI to help them with coding, without also making them know the AI building blocks.

That’s a cohort of students that enter the job market that’s really struggling. But if the fresh college grads know AI, we can’t find enough of them. So many businesses love to hire those fresh college grads.

It is very hard to say. I I know a number of people that have gotten really huge pay package. I’m actually very happy for them. I think it’s great, the funding going into pay AI people really well.

I have a lot of Silicon Valley friends that, you know, for whatever reason, have made a little bit of money. Many of them just keep working really, really hard. But equally before and after, you know, they wound up making a little bit of money. So I I find that a lot of the tech culture, we do stuff because it’s fun, because it lets us, you know, hopefully help other people. It’s a way to change the world. I I I find that wealth mixed people become lazy much less than one might guess.

I hope we can get much closer to five, six, or more percent GDP growth. When looking to the future, it turns out one of the most expensive things in today’s world is intelligence. This is why it’s so expensive, at least in The US, to hire a highly skilled doctor to advise us on the medical condition or hire a highly skilled tutor to, you know, patiently teach our kids, because that intelligence training up that wise doctor, wise teacher, wise adviser is very expensive. But with AI, we finally have a path to make intelligence cheap.

And so in the future, if everyone can be assisted by an army of smart, well informed staff on all of these topics under the sun that currently only the relatively wealthy in society can afford to hire people for, then individuals will be so much more empowered and able to get so much more done. Lives will be so different, the GDP growth will be will be massive.

It’s still very dynamic. So for a lot of American companies, the leading frontier model is often kept closed, and then the one tier down model, not quite as good as releases open. I think it’s much better than nothing. I’m actually grateful for all the teams that are releasing open source, weight models. And then the other dynamic is China, especially, has been really taking the lead or well, taking a lead or getting up there in terms of releasing tons of really good open way models. I would say it is kind of not whatever predicted a decade ago that China AI would end up being more open than America AI.

Why do you think

It turns out that openness is great for a country’s development. So it turns out that when a team releases open source software, circulation of knowledge is much faster to the close by community. And so what I see is when a team in China releases an open way model, then, yes, of course, American can take advantage of it, but the China economy benefits even more from it. Because once something is open, it’s easier for teams to call each other and say, hey, buddy. How does this really work?

I’m having trouble with this final model. It’s just that circulation of knowledge is really valuable for innovation. And when The US has more closed models and when, you know, teams are trying to pay these $100,000,000 salaries to extract talent, then that circulation of knowledge becomes very slow, and it slows down the rate of American and and European innovation.

I think in addition to increase innovation and circulation of knowledge, which the open way models helps with, I think that open way models is a tremendous source of geopolitical influence. For example, if someday, you know, some kid in some developing nation ask a question about a publicly sensitive topic or ask, hey, where are the national borders in this case, or what’s the history of this event or that event? The country of origin of the model they end up using will be delivering some answer. Whether the answer skews towards one nation’s values or another nation’s values is actually a tremendous source of influence and soft power.

Like it or not, open way models are a key part of the AI supply chain. China releasing low cost or free models into that key part supply chain means it’s really starting to build up a lead, right, and and build up a commanding user base. And this is why I think nations with a strong media and entertainment industry. It turns out South Korea has vastly disproportionate influence because of their leading entertainment industry. So people listen to whatever, you know, k pop or whatever, and that buys the nation a lot of influence.

Hollywood was a tremendous source of soft power for America. It paints a certain vision of the American dream, talks about the values of freedom and democracy. This is another frontier of communications and soft power.

I think there’s a lot of room for cooperation and then also some places that will be competitive. So first, while people sometimes, even me, talk about the AI race, there’s no single efficient finish line. It’s not one race. It’s AI is a general purpose technology, and you could be better aware at coding, better aware at answering questions, better aware at helping with, you know, mockers and finance and so on. So AI has many different capabilities, and there’s no one finish line. And even though it’s one capability, we’re gonna keep on improving for a long time.

So I I I feel like because of PR goals, AGI has been hyped up as it was a finish line, but I don’t think it’s a finish line. It’s just we’ll have continually improving capabilities for decades to come. Having said that, nations with stronger AI capabilities are gonna be more powerful. The citizens will be more prosperous. The economies will grow faster. So to the extent that different nations’ incentives are not aligned, nations with more powerful AI capabilities will be able to do more.

Just like, you know, if if a country has a fantastic electricity grid and another country, you know, has power outages and so on, well, one country can just use the electricity grid to do more manufacturing, more industrial work, just do a lot more that way.

Yeah. To be fair, I think US, Europe, China all have problems as well. But having said that, I think the work ethic, the velocity, when China’s government makes a all nation commitment, it’s all industrial commitment, that’s actually a very powerful force with kind of state level investments in semiconductors, in this education system. So, you know, like k 12 kids being trained to use AI. Businesses are asked to use AI, share knowledge, and then sometimes, you know, build this up and also sell internationally with state apparatus that may or may not be the whole control over rare earth elements.

So I think that whole of economy, whole of country efforts is actually a very powerful force that I wouldn’t underestimate.

I think the export control on chips has largely backfired. The way The US first put restrictions on Huawei and then later on, you know, export of NVIDIA and AMD and and other semiconductors, that really incentivized China. So before the export controls, semiconductor development in China, it was not frankly, it wasn’t moving that fast. You know, it was a nice area. There was some investment. But when America did that, then China really accelerated its semiconductor development. And so America incentivized China to do this, and it is paying off for China.

I think, you know, a number of Chinese companies are building offerings that individual chips are less powerful, but maybe have much larger number of chips, trying to build offerings competitive with certainly the last generation of NVIDIA, maybe increasingly the current generation. If I were to analyze just purely, you know, US national self interest, I think that caused China to accelerate the semiconductor industry in a way that may not be helpful to The US long term.

If I had one wish for the European regulators, spoken quite a few European regulators, I I was hearing things like, we wanna be leaders in regulating AI, and that’s a competitive advantage. And with all due respect, that’s not a competitive advantage. So my one wish for Europe is, stop regulating so much and just focus on investing and building. The thing is, it’s still early in the days of AI. It’s still early in in the game. And Europe has plenty of smart people. Let people work hard.

Don’t force them to not work hard. Let people that want to work hard, work hard, and stop overregulating and just go and invest and build stuff.

There’s tons of capital going into data centers and infra. We can debate, is there a bubble or not? We definitely need a lot of investments. Are we getting to the point where people are using such esoteric financial instruments to find cash for it that there’ll be a bubble? We could debate that. Right? So we definitely need a lot of investments, but when does it become overinvestment? That that’s a that’s an interesting question. The other place that I think we need to invest in a lot is not just the infra data center foundation model layer, but the application layer.

It turns out that because of others having spent, you know, billions of dollars to train these AI models, we can now access them for, you know, hundreds of dollars or thousands of dollars or whatever for all tens of dollars. It’s wonderful to build tons of applications that just were not possible before. Now from a VC investment perspective, I’ve heard from multiple VCs, is, bizarrely, the cost of trying something out is so low that there are fewer ideas. It’s not quite sure where to put massive amounts of capital to work at the application layer.

In fact, if you look a lot of the application layer investments, sometimes it feels like, you know, firms are putting in a $100,000,000 so that they can pay OpenAnthropic, so that OpenAnthropic can pay NVIDIA, which is where all the money is is ending up. Having said that, there’s so many valuable bets to be placed at the application layer to just build stuff. But the dilemma is you could do it in a very capital efficient way. So if someone wants to say, I wanna put $10,000,000,000 to work.

You know, yes, you could build $10,000,000,000 worth of data centers. We know how to spend that money. But how do you spend $10,000,000,000 in building applications? The problem is almost, only cost me a million dollars to try an idea out, so how do I spend $10,000,000,000? It’s kind of a problem and also not a problem, but I think we

I think it still varies. I’m seeing a lot of green shoots of software applications that were not that expensive to build. And if your LLM token usage is not the majority of your expense

So the dynamic that I’m excited about is I think that as token costs continue to come down, we’ll see how the economics change. Right? Right now, token is just expensive, but hopefully that will change and the value created is really large. Actually, I I remember an earlier era in early days of food delivery, for example, I saw this in both The US and China. There’s lots VC subsidized eating. Right? You know, it was great. We could we could eat food delivered. It was basically VC subsidized.

I think we’re seeing that right now with a lot of VC subsidized AI coding. The laws of physics or the laws of finance says that at some point, right, this can’t go on forever. But where it settles down, I think there will be some very valuable businesses that are not perpetually VC subsidized, but navigating this crazy VC subsidy world to get to a good outcome takes a lot of skill. But having said that, I still wanna say, there are a lot of, smaller applications that are not yet doing these, you know, hundreds of millions of dollars.

Maybe they’re doing millions of dollars or tens of millions of dollars of revenue that haven’t been quite expensive to build and to operate and that I think we’ll we’ll see continue to grow.

It’ll be all of the above. We we will have large models and mid sized models and tiny small models. And the reason I’m confident about that is because, the nature of intelligence is diverse. Sometimes we do intellectually really easy tasks. Like, if someone asked me, like, alright. Yesterday, my daughter, she misspelled the word butterfly, so I need to tell her how to spell butterfly. Is it low you know, it’s an easy intellectual task. And sometimes I’m sitting down thinking for hours about some complex technical problem that’s really hard.

And so intelligence has a range of things we want to do, and so the set of things we want AI to do too has a huge range. If you want AI to do basic grammar checking and spell checking, you don’t need a trillion parameter model. Use a tiny model, maybe running locally, just do that. But if wanted to do complex reasoning to write a piece of code, then yes, having a powerful model is gonna do better. I’m actually very confident we’ll end up with a huge range of models, small and large, to do the huge range of tasks.

Just just like we have humans do a range of tasks with difficulty, same with AI.

I disagree with that. I think we’re seeing useful agentic workflows right now. So AI Fund, our team has built so many agentic workflows, for so many tasks where, you know, we just could not even do the task but for agentic workflow. Can you give me an example? I’m fascinated. Sure. So for example, over a year ago, we thought that, after one of the Biden Trump debates, for better or worse, we thought that tariff compliance may become an issue. Unfortunately, we turned out to be right. But so last year, I think it was around August, we started building started exploring building technology to help with tariff compliance.

And by the way, I don’t know if you’ve seen these tariff compliance docs, but frankly, when I look at what it takes to follow these these paperwork, it’s like it it makes me wonder, oh my god. What is this? So, you know, you you say import a bicycle, you know, then you look at the specs of the bicycle, how much does it cost, the size of the wheels, there are all these rules and regulations to, import a bicycle. It just makes me go, oh my god, are humans really doing this?

So we built agentic workflows to read the tariff compliance documents carefully, get the spec for what someone wants to import carefully, try to match, make suggestions. And, so this is now one of our portfolio companies called Gaia Gaia Dynamics that, you know, because of the increased complexity in tariff compliance, has been doing pretty well. We just could not have done this without agentic workflows. We have, medical assistant operating in India, AI assistant, Callitus, helping process legal documents. Many of these workloads, we just could not do.

So I I I find that they’re useful AI agentic workflows already today, and the large businesses too, not just our startups. When we look at the hyperscalers and I chat to friends and some of the large businesses, there’s a bunch of internal workflows that we just could not be doing without these AI agents.

At some point, the laws of physics or the laws of finance or something, margins do matter. But one of the tricky things about AI is we know the technology is gonna change, so we don’t build assuming the technology will be stagnant. We do build assuming the technology will evolve. So one one obvious one, LLM token prices have been rapidly falling, depending on who you believe, falling 80% year on year or whatever. Kind of frankly, when we build prototypes, we routinely just not worry about token cost because the first most important thing is just build a product that users love, and then what we find is after we build some this actually happened to me a few times now.

We’ll build something and not worry about the cost, and then, you know, users start to use it, and then our API build starts climbing. And then it is really, like, kind of you look at this every few weeks and go, woah. This is getting really expensive. This is costing me salary of 1 engineer. It costs me more than two engineers. Costs me more than a whole bunch of engineers. It’s alright. Just but fortunately, when that has happened, almost every time so far, we’ve been able to use techniques to bend the cost curve back down even faster than the rate at which token prices are falling in the market.

And so I find that absolute margins are important, but when you have a view for where the technology is going, then it lets you not build for the margins today, but what you can forecast them being in the future, and I think that’s an important distinction. But we don’t take a blind utopian AGI, blah blah blah view either. I think that’s also overly simplistic.

Motes are changing. I find that moats tend to be a function of the industry rather than a function of the technology. So AI as a technology doesn’t really offer an answer to the moat for most businesses. So if you’re building AI, you know, for drones or legal or for whatever, the moat is more of a function of that industry. But one thing that is changing with regard to moats is previously, software used to be a mode. Right? If you have, you know, invested ten years to build a software, it’s really hard to replicate that.

That one mode is much weaker than before, but other modes, like, are you trying to use AI to accelerate to build a two sided marketplace, which can be very defensible, defensible? Or are you building for a consumer more focused for an enterprise? Are there brand and reputational effects, right, that can help you build defensibility there? So I find that the software mode has changed, but other modes tend to be analysis based on the industry.

I think the biggest barrier in most large enterprises is is actually, people and change management. Not data? It’s not data. I think it’s definitely not data. Not that data is not important, but that’s definitely not the bottleneck. The interesting thing about AI hype is there’s almost always a gem of truth in the hype. It’s just that it’s been hyped up 10 times more than the reality. And maybe actually, let me give you one example that will come back to data. There’s been this buzz about, oh, with AI, we’ll have unicorns with one employee.

It’s like a thing. And and it’s fine. If you wanna build a unicorn startup, billion dollar with one employee, go for it. It’s a good thing to do. Frankly, if you have a billion dollar valuation, you could afford to pay two employees or even 10. So why do you need to hype it all the way up to say, let’s do this with just one employee. Right? So that so it is true that team sizes are shrinking. We can get more done with smaller teams, so that is true.

But the hype is then saying, let’s build a unicorn with one SaaS. I find a lot of AI hype is so hard to disentangle because there’s a gem of truth in it. It’s just been hyped up a lot more. So on data. Data is important, but it turns out that data is very verticalized, and you don’t need as much of it to get started as you think. So for example, you know, I don’t know, LandingAI does a lot of work with financial institutions, health care. A lot of financial institutions have plenty of transaction data.

Take the PDF file, turn it into already markdown text, go process that, find value in that. Like, for example, I don’t know, we could take SEC filings, large complex financial tables, very accurately turn those financial tables into Excel spreadsheets, then go get your analyst or your AI to analyze that and draw conclusions. So often with bit of scrappiness, look in internal data, look at your public data, you can often get some stuff going. And it turns out that, a lot of Internet data is kind of general purpose data.

Most of the world’s data is actually private, and there’s a lot of businesses. It’s actually very valuable transaction data, you know, sales data, product data, manufacturing data, logistics data. All that data with a scrappy team that knows how to use it can actually start to build something to get value out of it. Not to say more data wouldn’t be even better, but you’re not stuck to even take the first few steps or lack of data.

I think we’ll get there. So I find that a lot of enterprises are adopting OMs, ChatGPT and many others. I think today, there are still businesses that are still on prem rather than on the cloud, but we’re making progress. Actually, one thing about AI, this this hype that we have AGI in two years or whatever, I think that’s just ridic that’s just for most reasonable definitions of AGI, that’s just not gonna happen. And just as how long are we now into the cloud era, but we still have an awful lot of on prem jobs, I think that AI adoption, it will be wonderful.

There will be tremendous GDP growth. It’s also gonna take much longer than the hype says it will. I I actually think that a decade from now, we will still be working to identify valuable applications and enterprises and building them. Having said that, we will make a lot of progress over the next one or two years, but we’re not gonna be done even ten years from now.

A big one for me. Even earlier this year, we saw some senior business leaders advise people to not learn to code on the grounds that AI will automate it. We’ll look back on that as some of the worst career advice ever given. Because as coding becomes easier with AI assisting us, a lot more people should learn to code, not fewer. And I’m already seeing I mentioned the market example just now with building an app for feedback swiping. But for a lot of job functions, people that know how to tell computer exactly what they wanted to do so the computer can do it for you, they’ll just be more powerful.

And for the foreseeable future, the language of precisely telling computers what you wanted to do is coding. It doesn’t mean you should write code by hand. Writing code by hand is becoming obsolete. Right? Oh, really? Don’t do that. Not but to get AI to write code for you, and people can do that would be more effective and more powerful and help more fun.

Oh, I think we’ll see plenty of improvement over the next two years, but I think we still won’t be done getting even more improvements ten years from now. One place where it’s super promising is AI assisted coding. So we’re seeing real productivity gains, real returns. It’s really changing the way software is written. It’s really been fantastic. It’s, frankly, so many of my friends is coding is so much more fun with AI to help us out than without. So we are seeing returns, just to be clear.

But we still won’t be done growing this ten years from now.

To me, the question is, is AI mostly for cost savings, or is it for growth? You know, it’s difficult to change workflows. A lot of companies tend to think cost savings, but it maybe here’s here’s the problem. There’s actually one pattern I see. Let’s say I have a word toss that has, you know, like, five steps. Right? And let’s say each step takes 20% of my effort. Like, maybe I’m on underwriting approvals, you know, do I approve this loan or not? Right? So let’s say let’s say for simplicity, the five steps each takes 20% of my effort.

If you can automate one of those steps, it’s a 20% cost savings, which is really nice. You know, it could be great if you’re a low margin business, but it doesn’t feel like a game changer. What I find is that the more valuable users of AI, it actually requires it often requires rethinking that workflow. And the pattern I see is instead of taking the 20% cost savings, which you could do, that’s fine, nothing wrong with that, the two patterns to then getting growth is either do more or do it faster.

So in the case of underwriting, making loans, if instead of saving 20% of my human labor, if I can now rework the workflow to turn around my decision making time. So instead of someone needing to wait, you know, two weeks before a loan officer looks at it, but we can just give you an initial answer in ten minutes, that changes the product and lets you drive growth. So that’s a faster pattern. And then there’s also the more pattern. So another example, there are lot of businesses that could do, high touch, say, customer service only for expensive high end clients.

But if you can now serve a much larger group of people or let let’s say financial advice. Instead of giving high touch financial advice to a small group of people, if we can now deliver that quality of service to a lot more people, then that again changes the product and lets you drive growth. So instead of cost savings, if AI lets you do something way faster or lets you take a task and do it a thousand times more, instead of serving a small number of people, let’s serve a lot more people because it’s not economic to do so.

These are the two patterns I’ve seen to drive value increases, and I think that will be important for unlocking lot of this GP growth. You

I think this will evolve over time. I’m gonna make an analogy. In the early days of, say, the computing industry, it was a vertical players that won. Because if you want to connect a keyboard to your computer motherboard, which is a CPU, is it okay if your keyboard has a, you know, plus minus five volts and your CPU has some other voltage? Is that okay or not? So we didn’t know where are the API boundaries or if your CPU has, your memory laid out a certain way, a compute and your your math accelerator, they needed to interoperate with each other.

So before we wound up having a clear conception of where to draw lines and whether the API boundaries, the integrated players, IBM back in the day, could solve all the problems and build valuable working products. But as the industry mature, we started to have standards. Like, for for example, now we have a USB standard. Before, there were other standards. So now you make a computer, someone else makes a keyboard, we plug them together, and it all works. So when an industry is immature, it turns out where to draw those boundaries so that different participants do their part and have it still interoperate.

That’s less clear. But then as the industry matures and is, you know, more standards for, if I wanna publish a compressed OM model on the Internet, what’s the file format for that? And you’re assigned to see more standards, then that makes it easier for individual players to do something and still have it fit into the broader ecosystem.

I think clearly OpenAI’s investments have have paid off to date. It is possible to overinvest at some point, but I don’t know if that is the point. And then I think also the financial instruments being, you know, used by many players to shift risk around have been really interesting. I I find that, overly used complex use of financial instruments to shift risk sometimes increases the risk of there being a bubble at some point, so that’s something to watch out for.

It’s something to keep an eye on. I’m not alarmed by them. I think things could be more frothy or less frothy. Things could be more of a bubble and less of a bubble, and these are signs of things dealing a little bit more bubble ish.

You mentioned the article on the $600,000,000,000 problem of AI. I am concerned about that. But it’s interesting. My concern for different layers of the stack is different. So what I’m seeing is for the application layer, there is very clear ROI. I think it’s fantastic. So someone else trained these models who can build applications for, you know, $100,000 or a million dollars and start generating ROI. And then I think it is calibrating to the right level of infrastructure investment that is tricky. But having said that, it is also at the same time very clear that we do need more, like, c, more data centers, and more semiconductors.

That too is very clear. So we should be investing a lot, and I’m glad we are. But what exactly is the right amount to invest? I think that’s the tricky question. It should be a lot, though. Do you get annoyed by the bubble discussion? I don’t get annoyed by the bubble discussion. I do get annoyed by the hype. When regulators are calling me up and saying, hey. We heard AI could lead to human extinction. Thankfully, much less than that now than a couple years ago. And then then instead, the conversation should be, how can we upscale the workforce?

Where can we invest? You know? Not like, how do we slow this thing down? I think the hype has really distorted public perception of AI. Oh, and and and one downside to the hype too is without public support of AI, things slow down. So for example actually, one of my friend friends works a lot with high school students, and he told me that he was talking to a girl, a high school student, that she was talking to her about maybe pursuing a career in AI, and she said, you know what?

I heard AI could have something to do with human extinction. I don’t wanna have anything to do with that. And so this hype turned a high school girl away from working on AI at a time where it’d be so promising for them to leap into AI, and I think this really causes people to make weird decisions both at the individual school student level as well as at the community level. When a community shuts down building out the data center, they could be good for the community and good for the world.

I think that’s also unfortunate.

Embrace it. Update curricula. Teach them as much AI as possible. Students are gonna live in a world where they will be using AI and having a help them. Gotta teach students to do that. I think it’ll be different for different fields, but one thing that is clear is get all your students to learn to code. What’s

I think my favorite tools keep changing. If you ask me every three months over the last year what my favorite coding tool is, my answer would have kept on changing.

Pretty hard to say. OpenAI has a very strong consumer brand, and that’s very defensible. In contrast, developers are more likely to switch coding tools on the dime. So I love Cloud Code. I think it’s fantastic. But I find myself using OpenAI Codex much more over the last month. I think OpenAI Codex has actually gained raw momentum. And then I’m also keeping an eye on Gemini COI, which I think is also getting better, maybe at a faster rate than people have given them credit for. So decoding dev tools and API tools market, the moat there is weaker than having a strong consumer brand.

So I think that’s something that companies have to sort out.

I really appreciated the speed and intensity of Baidu and also of the China ecosystem. I think it’s really unfortunate that in some parts of The United States, advising someone to work hard is viewed as publicly incorrect or something.

Hopefully, the European viewers won’t hate me or hate us both for that. I think, frankly, I wish people could work four hours a week and and be wildly successful, but the practical reality is when people work hard, they get more done. Now I wanna acknowledge that not everyone in every point of their life is in a position to work hard. So, you know, the week after my kids were born, I didn’t work that hard. I took time off, spent time with the kids, right, for a little more than a week.

And I think we need to respect people in all walks of life, including people that for whatever reason are not in a position to work hard at that moment. But if someone wants to work hard, go quote Steve Jobs, make a dent in the universe. Let’s empower them and celebrate that. And if someone, for whatever situation, can’t work hard, let’s also, right, respect that and maybe celebrate that. But I think this is a moment in time where there’s so much stuff we could build. People that work hard to learn a lot and build things will accomplish a lot.

Did you do nine nine six? The term nine nine six wasn’t an explicit term that I use. I find that, right, these days, I really love what I do. Really doesn’t feel like work, but on a lot of my weekends, I’m sitting in a coffee shop coding away because it’s the most fun thing you could do on a Saturday. So I I I actually don’t bother to keep track of my hours. It’s probably a lot.

Oh, one thing about AI Fund, yes, we call ourselves a fund, but frankly, the way we run the fund day to day, we act much more like operators than investors. AI Fund’s a venture studio, and I believe our skill set is actually in building, not just in capital asset allocation or whatever. So we work really hard to screen ideas. We talk to customers. You know, I’m sometimes on customer calls myself, and then we bring in founders who work alongside us. We’re reviewing their product, giving feedback on the product, arguing about pricing.

So we’re my day to day life is much more operator. And, yes, eventually, we have to do the financial diligence since I write a check and do follow on. We do all that, but a lot more

We call ourselves a venture studio or a venture builder. Incubators usually bring in founders that already have an idea. We go earlier than that. We often work with our investors and partners to come with an idea. And only after we have an idea, then we go and try to find the best founder to co build, to cofound the company with us. So

Depends. We end up with some common stock for the sweat equity of building the the company, and then we usually our first check-in is usually, like, a million dollars at a $4,000,000 cap. So kind of 20% ownership or safe.

Yeah. Plus some common for the sweat equity. The reason we do this is because I find it while there are VCs that, you know, do the competitive deal flow thing, they make a lot of money that way, I think my team’s biggest contributions is not, you know, fighting over hard deals. It is finding ideas and creating companies that would not exist, but for the fact that we and a founder got together to cofounder together. So I think we just create more value in the world by creating new companies rather than only discovering hot companies to try to put money into.

I am worried about the difficulty of bringing everyone along with us. In previous ways of economic disruption, like when when when our nations went from mainly agriculture to non agriculture, some of the was a farmer could keep farming until they retire, but the kids had to learn a different trade, maybe with the city or whatever. The change is so fast this time around that we need people that are alive today to learn new skills as opposed to we need their kids to learn new skills. And that’s actually very challenging.

And historically, I don’t think we’ve ever been good at that.

I think media has an important role to play to curate and disseminate knowledge. I think the quality of questions that reporters are asking has been very clearly trending up over time, but there is still the hype element of it that keeps on distorting the information ecosystem. Unfortunately, there are financial incentives and, you know, regulatory capture and legislative benefit types of incentives and certain types of hype. And that’s actually one pattern that I’ve seen. I I won’t name any companies. But I find that, there are companies with something to lose whose statements over time have become more moderated.

So, you know, I I find that as you’re an established company, you just say more sensible things, But there are some companies that I think are at greater existential risk, and I find some of those companies that I don’t wanna name to be the worst sources of hype because you’ve got less to lose. Let’s just say a bunch of random stuff.

No no no comment on individuals, but but I I think that when you have something to lose, you say more sensible things. But when your company faces greater existential risk, you know, you sometimes people say weird things for fundraising and and that’s unfortunate.

Yeah. I’m sorry to hear about your mother. Thank you. What what excites me, I I wanna empower everyone to build AI. I think the distance between you having an idea and building it is now much shorter, and we need not just software engineers to be creating. So in the future, I hope that a lot of people, instead of saying, is there an app for that? They’ll say, I built an app for that. And instead of just being a software user, there’ll be a software creator, and when you get there, then people all around the world will be much more empowered, get more done, have more fun.

No. I really enjoy your show, so it’s a privilege to be here, Harry.