--:-- Host: Nothing seems fundamentally so hard that it couldn't be solved by the smartest people in the world working incredibly hard for the next five years. Humanity went through the agricultural revolution and industrial revolution. We're going through another revolution. We will not be able to call it something. It's a future people will call it something. But we are going through something. The number of solo entrepreneurs that this technology is going to enable is vastly increased what a single person can do. For the first time, opportunity is massively available for everyone. Just the ability for more people to be able to become entrepreneurs is massive.
--:-- Host: The age of solo entrepreneurship powered by AI is here, but the path to full automation is messier than the hype suggests. Today, you'll hear from Adam D'Angelo, founder of Quora and CEO of Poe, and Amjad Masad, founder and CEO of Replit, on why we're in a brute force era of AI rather than true intelligence and what that means for the future of work. We discussed the expert data paradox, how automating entry-level jobs creates a crisis in training the next generation of experts, why managing tens of agents in parallel will define the next wave of productivity, and how the sovereign individual framework might be the best lens for understanding AI's economic and political impact. Plus, Adam makes the case for why vibe coding is radically underrated, and Amjad explains what Claude 4.5's strange new self-awareness might signal about the path ahead. Let's get into it.
--:-- Adam D'Angelo: The world was very different. Judging on how much progress we've made in the last year with things like reasoning models, the improvement in code generation ability, the improvements in video gen, it seems like things are going faster than ever. I don't really understand where the bearishness is coming from.
--:-- Host: Well, I think there's some sense that we hoped they would be able to replace all tasks or all jobs. Maybe there's some sense that it's middle to middle, but not end to end. Maybe labor won't be automated in the same way that we thought it would on the same timeline.
--:-- Adam D'Angelo: I don't know what the previous timelines people were thinking were, but if you go five years out from now, we're in a very different world. A lot of what's holding back the models these days is not actually intelligence. It's getting the right context into the model so that it can use its intelligence. There are some things like computer use that are still not quite there, but I think we'll almost definitely get there in the next year or two. When you have that, I think we're going to be able to automate a large portion of what people do. I don't know how we call that AGI, but I think it's going to satisfy a lot of the critiques that people are making right now. I think they won't be valid in a year or two.
--:-- Host: What is your definition of AGI?
--:-- Adam D'Angelo: I don't know. Everyone thinks it's something different. One definition I like is if you say that you have a remote worker, a human, any job that can be done by someone whose job can be done remotely, that's AGI. You can then say it does have to be better than the best person in the world at every single job. Some people call that ASI. It does have to be better than teams of people. You can argue with those different definitions. But once we get to be better than a typical remote worker at the job they're doing, we're living in a very different world. That's a very useful anchor point for these definitions.
--:-- Host: So you're not sensing the same limitations of LLMs that other people are saying? You think there's a lot more room that LLMs can go from here? We don't need a brand new architecture or other breakthrough?
--:-- Adam D'Angelo: I don't think so. There are certain things like memory and learning, continuous learning, that are not very easy with the current architectures. Even those you can sort of fake and maybe we're going to be able to get them to work well enough. But we just don't seem to be hitting any kind of limits. The progress in reasoning models is incredible. The progress in pre-training is also going pretty quickly, maybe not as quickly as people had expected, but certainly fast enough that you can expect a lot of progress over the next few years.
--:-- Host: Amjad, what's your reaction here in all this?
--:-- Amjad Masad: I think I've been pretty consistent and consistently right, perhaps. Dare I say. Consistent with how things are unfolding. I started being a bit of a more public doubter of things around the time when the AI safety discussion was reaching its height back in maybe 2022, 2023. I thought it was important for us to be realistic about the progress because otherwise we're going to scare politicians, we're going to scare everyone. DC will descend on Silicon Valley, they'll shut everything down. My criticism of the idea of AGI 2027, that paper I think it's called, Alexander someone else wrote, and the situational awareness and all these hype papers that are not really science, they're just vibe.
--:-- Amjad Masad: Here's what I think will happen. The whole economy will get automated. Jobs are going to disappear. All of that stuff is unrealistic. It is not following the kind of progress that we're seeing, and it is going to lead to bad policy. My view is LLMs are amazing machines. I don't think they are exactly human intelligence equivalent. You can still trick LLMs with things like they might have solved the strawberry one, but you can still trick it with single sentence questions like how many R's are in this sentence? I think I tweeted about it the other day. Three out of the four models didn't get it. GPT-5 with high thinking had to think for 15 seconds to get a question like that. LLMs are a different kind of intelligence than what humans are. They have clear limitations, and we're papering over the limitations and working around them in all sorts of ways, whether in the LLM itself, the training data, the infrastructure around it, everything we're doing to make them work. That makes me less optimistic that we've cracked intelligence.
--:-- Amjad Masad: Once we truly crack intelligence, it'll feel a lot more scalable. The idea behind the bitter lesson will actually be true, that you can just pour more power, more resources, more compute into them and they'll just scale more naturally. Right now there's a lot of manual work going into making these models better. In the true pre-training scaling era, GPT-2, 3, 3.5, maybe up to 4, it felt like you could just put more internet data in there and it just got better. Now it feels like there's a lot of labeling work, a lot of contracting work, a lot of these contrived RL environments are getting created in order to make LLMs good at coding and becoming coding agents. The news from OpenAI is that they're going to do that for investment banking. I try to coin this term I call functional AGI, which is the idea that you can automate a lot of aspects of a lot of jobs by just going in and collecting as much data and creating these RL environments. It's going to take enormous effort and money and data to do that.
--:-- Amjad Masad: I agree with Adam that things are going to get better, 100%. Over the next three months, six months, Claude 4.5 was a huge jump. I don't think it's appreciated how much of a jump it was over 4. There are really amazing things about Claude 4.5. There is progress, we're going to continue to see progress. I don't think LLMs, as they can understand, are on the way to AGI. My definition for AGI is the old school RL definition: a machine that can go into any environment and learn efficiently, in the same way that a human could go into a pool game and within two hours they can play pool and be able to do it. Right now, there's no way for us to have machines learn skills like that on the fly. Everything requires enormous amounts of data and computing time and effort. More importantly, it requires human expertise, which is the non-bitter lesson idea. Human expertise is not scalable. Today, we are in a human expertise regime.
--:-- Adam D'Angelo: Humans are certainly better at learning a new skill from a limited amount of data in a new environment than the current models are. On the other hand, human intelligence is the product of evolution, which used a massive amount of effective computation. This is a different kind of intelligence. Because it didn't have this massive equivalent of evolution, it just has pre-training for that, which is not as good. You then need more data to learn every new skill. But in terms of functional consequence, when will the job landscape change? When will the economic growth hit? That's going to be more a function of when we can produce something that is as good as human intelligence, even if it takes a lot more compute, a lot more energy, a lot more training data. We could just put in all that energy and still get to software that's as good as the average person at doing a typical job. I don't disagree with that. It feels like we're in a brute force type of regime, but maybe that's fine.
--:-- Host: So where's the disagreement then? I guess there's agreement on that. Where is the divergence perhaps?
--:-- Amjad Masad: I don't think that we'll get to the singularity or the next level of human civilization until we crack the true nature of intelligence. Until we understand and have algorithms that are actually not brute force.
--:-- Host: And you think those will take a long time to come?
--:-- Amjad Masad: I'm sort of agnostic on that. It does feel like the LLMs in a way are distracting from that because all the talent is going there. There's less talent trying to do basic research on intelligence.
--:-- Adam D'Angelo: At the same time, a huge portion of talent is going into AI research that previously wouldn't have gone into AI at all. You have this massive industry, massive funding, funding compute, but also funding human employees. Nothing seems fundamentally so hard that it couldn't be solved by the smartest people in the world working incredibly hard for the next five years on it.
--:-- Amjad Masad: Basic research is different. Trying to get into the fundamentals as opposed to industry research, like how do we make these things more useful to generate profit? That's different. Thomas Kuhn, this philosopher of science, talks a lot about how these research programs end up becoming like a bubble, sucking all the attention and ideas. Think about physics and how there are these industries of string theory that pull everything in. It's a black hole of progress.
--:-- Adam D'Angelo: One of his things was, you got to wait until the current people retire to have a chance at changing the paradigm. It's very pessimistic about paradigms. But I feel like the current paradigm is pretty good. We're nowhere near diminishing returns of continuing to push on it. I would bet that you can keep doing different innovations within the paradigm to get there.
--:-- Host: Let's say we continue to brute force it. We're able to automate a bunch of labor. Do you estimate that GDP is something like 4% or 5% a year, or are we going up to 10% plus? What does it do to the economy?
--:-- Adam D'Angelo: It depends a lot on exactly where we get to and what AGI means. Let's say you have LLMs that with an amount of energy that costs $1 an hour, they could do the job of any human. You're going to get to much more than 4% to 5% GDP growth in that world. The issue is you may not get there. It may be that the LLMs that can do everything a human can do actually cost more than humans do currently, or they can do 80% of what humans can do and then there's this other 20%. At some point you get to LLMs that can do every single thing a human can do for cheaper. I don't see a reason why we don't eventually get there. That may take 5, 10, 15 years. Until you get there, we're going to get bottlenecked on the things the LLM still can't do, or building enough power plants to supply the energy, or other bottlenecks in the supply chain.
--:-- Amjad Masad: One thing I worry about is the deleterious effect of LLMs in the economy. LLMs effectively automate the entry level job, but not the expert's job. Take QA, quality assurance. LLMs are so good, but there are still all these long events that they don't handle. You have a lot of really good QA people now managing hundreds of agents. You effectively increase productivity a lot, but they're not hiring new people because the agents are better than new people. That feels like a weird equilibrium to be in. I don't think many people are thinking about it.
--:-- Adam D'Angelo: It's happening with CS majors graduating from college. There are just not as many jobs as there used to be. LLMs are a little more substitutable for what they previously would have done. That's contributing to it. It means you're going to have fewer people going up that ramp that companies paid a lot of money to employ and train. It's a real problem. That problem also creates an economic incentive to solve the problem. There may be more opportunities for companies that can train people or use AI to teach people these things. But for sure, that's an issue right now.
--:-- Amjad Masad: Another related problem is that since we're dependent on expert data to train the LLMs, and the LLMs start to substitute those workers, at some point there are no more experts because they're all out of jobs and they're equivalent to the LLMs. But if the LLM is truly dependent on labeling data and expert RL environments, how would they improve beyond that? That's something for an economist to really sit down to think about. Once you get the first tick of automation, there are some challenges there. How do you go to the next part?
--:-- Adam D'Angelo: A lot of it is going to depend on how good RL environments can be created. On the one extreme you have something like AlphaGo where it's just a perfect environment and you can blast past expert level. But a lot of jobs have limited data that anyone can train from. It'll be interesting to see how easy it is for research efforts to overcome that bottleneck.
--:-- Host: If you had to make a guess on what job category is going to be introduced or explode in the future, some people say everyone's an influencer or in some sort of caring field, or everyone's employed by the government as some sort of bureaucrat, or maybe training the AI in some way. As more things start to get automated, what is your guess as to what more and more people start to do?
--:-- Adam D'Angelo: Doing art and poetry. At some point you have everything automated, and then I think people will do art and poetry. There's a data point that the number of people playing chess is up since computers got better than humans at chess. I don't think that's a bad world if people are all just free to pursue their hobbies, as long as you have some way to distribute wealth so people can afford to live. In the near term, that's a while away. 10, 15 years out.
--:-- Amjad Masad: In the near term, the job categories that are going to explode are the jobs that can really leverage AI. People who are good at using AI to accomplish their jobs, especially to accomplish things that the AI couldn't have done by itself. There's massive demand for that. I don't think we're going to get to a point where you automate every job, definitely not in the current paradigm. I'm not certain it would ever happen, but definitely not in the current paradigm. A lot of jobs are about servicing other humans. You need to be fundamentally human to understand what other people want. You need to have the human experience. Unless AI is actually embodied in human experience, humans will always be the generators of ideas in the economy.
--:-- Host: Adam, we respond to Amjad's point around the human part because you created one of the best wisdom of the crowds platforms in the universe, and now you've gone all in with Poe. What are your thoughts on to what extent will we be relying on humans versus trusting AIs to be our therapists, our caretakers in other ways?
--:-- Adam D'Angelo: Humans have a lot of knowledge collectively. Even one individual person who's an expert and has lived a whole life and had a whole career and seen a lot of things, they often know a lot of things that are not written down anywhere. Tacit knowledge. Also what they're capable of writing down if you did ask them a question. There's still an important role for people to play in the world by sharing their knowledge, especially when they have knowledge that just wasn't otherwise in an LLM's training set. Whether they will be able to make a full-time living doing that, I don't know. If that becomes a bottleneck, then for sure all the economic pressure goes to that.
--:-- Adam D'Angelo: In terms of having to be human to know what humans want, I don't know about that. Recommender systems, the system that ranks your Facebook or Instagram or Quora feed, those recommender systems are already superhuman at predicting what you're going to be interested in reading. If I gave you a task that was make me a feed that I'm going to read, there's no way no matter how much you knew about me that you could compete with these algorithms. They have so much data about everything I've ever clicked on, everything everyone else has ever clicked on, what all the similarities are between all those data sets. As a human you can simulate being a human and that makes it easier to test out ideas. Composers and artists, chefs, they produce something and taste it, and it's important that they can taste it. But they have very little data compared to what AI can be trained on. I don't know how that's going to shake out.
--:-- Amjad Masad: That's a good point. Recommender systems are aggregating all the different tastes and finding where you sit in the multidimensional taste vector space and getting you the best content there. I think that's more narrow than we think. Yes, it's true in recommender systems, but I'm not entirely sure it's true of everything.
--:-- Amjad Masad: The best prediction for where the world is headed, and this is not an endorsement, but I think the sovereign individual can change to be a really good set of predictions for the future. It's not a scientific book, it's a very polemic book. In the late 80s, early 90s, two people out of the UK wrote this book about trying to predict what happens when computer technology matures. They said humanity went through the agriculture revolution and the industrial revolution. We're going through another revolution. Information revolution, now we call it intelligence revolution. I think we will not be able to call it something. It's a future people will call it something, but we are going through something. They try to predict what happens from here. They arrive at the idea that you're going to have large swaths of people that are potentially unemployed or economically not contributing, but the entrepreneur capitalists are going to be so highly leveraged because they can spin up these companies with AI agents very quickly. They have interesting ideas, they're human, they have interesting ideas about what other people want. They can create these companies very quickly and these products and services and organize the economy in certain ways.
--:-- Amjad Masad: The politics will change because today's politics is based on every human being economically productive. When you have massive automation and only a few entrepreneurs and very intelligent, generative people are actually able to be productive, the political structures also change. They talk about how the nation state subsides and you go back to an era where states are competing over people, over wealthy people. As a sovereign individual, you can negotiate your tax rate with your favorite state. It starts to sound like biology a little bit. I don't think it is far from where it might be headed. It's not a value judgment or desire, but I do think it's worth thinking about. When people are not the unit of economic productivity, things have to change, including culture and politics.
--:-- Adam D'Angelo: There's a question with that book and this conversation more broadly of when does the technology reward the defender versus the aggregator? When does it incentivize more decentralization versus centralization? Peter Thiel had this quip a decade ago of crypto is libertarian, it is more decentralizing, AI is communist or more centralizing. It's not obvious to me that that's entirely accurate on either side. AI does seem to empower a bunch of individuals, as you were saying. Crypto turns out is like fintech or stable, it does empower nation states. There's an open question as to which technology empowers the edges or the center. If it empowers the edges, it seems like the sovereign individuals. Maybe there's a barbell where the incumbents just get much, much bigger and there are these edges.
--:-- Adam D'Angelo: I'm very excited for the number of solo entrepreneurs that this technology is going to enable. It's vastly increased what a single person can do. There are so many ideas that never got explored because it's a lot of work to get a team of people together, raise the funding, get the right kind of people with all the different skills you need. Now that one person can bring these things into existence, we're going to see a lot of really amazing stuff.
--:-- Amjad Masad: I get these tweets all the time about people who quit their jobs because they started making so much money using tools like Replit. It's really exciting. For the first time, opportunity is massively available for everyone. That is the most exciting thing about this technology, other than all the other stuff we're talking about. The ability for more people to be able to become entrepreneurs is massive.
--:-- Host: That trend is obviously going to happen. As we look out at the next decade or two, do you think that AI is more likely to be sustaining or disruptive in the Christensen sense? Do you think that most of the value capture is going to come from companies that were scaled pre-OpenAI starting? So Replit still counts as the latter and Quora to some degree. Or do you think most of the value is going to be captured by companies that started after, let's say, 2015, 2016?
--:-- Host: So there's a related question, which is how much of the value is going to go to the hyperscalers versus everyone else. I actually think we're in a pretty good balance. There's enough competition among the hyperscalers that as an application-level company, you have choice and alternatives, and the prices are coming down incredibly quickly. But there's also not so much competition that the hyperscalers and the labs like Anthropic and OpenAI are unable to raise money and make these long-term investments. So I actually think we're in a pretty good balance, and we're going to have a lot of new companies and a lot of growth among the hyperscalers.
--:-- Amjad Masad: I think that's about right. The terminology of sustaining versus disruptive comes from the innovator's dilemma. It's this idea that whenever there's a new technology trend, there's a power curve. It starts as a toy that doesn't really work or captures the lower end of the market. But as it evolves, it goes up the power curve and eventually disrupts even the incumbents. Originally, the incumbents don't pay attention to it because it looks like a toy, and then eventually it disrupts everything and eats the entire market. That was true of PCs. When PCs came along, the big mainframe manufacturers didn't pay attention. Initially it was like, "Yeah, it's for kids." But now even data centers are running on PCs. So PCs were a hugely disruptive force.
--:-- Amjad Masad: But there are technologies that come along and really benefit the incumbents and don't really benefit the new players or startups. I think Adam's right. It's both. Maybe for the first time, it's both in a huge technology trend, because the internet was hugely disruptive. But this time, it feels like it's an obvious supercharge for the incumbents, for the hyperscalers, for the large internet companies. But it also enables new business models that are perhaps counter-positioned against the existing ones.
--:-- Amjad Masad: Although, I think what happened is everyone read that book and everyone learned how to not be disrupted. For example, ChatGPT was fundamentally counter-positioned against Google because Google had a business that was actually working. ChatGPT was seen as this technology that hallucinates a lot and creates bad information, and Google wanted to be trusted. Google had ChatGPT internally but didn't release Gemini until like two years after ChatGPT, and ChatGPT had already won the brand recognition. So in a way, OpenAI came out as a disruptive technology, but now Google realizes it's a disruptive technology and responds to it. At the same time, it was always obvious that AI is getting better with Google. At minimum, search overview has gone a lot better. Their workspace suite is getting a lot better with Gemini. Their mobile phones, everything gets better. So it seems like it's both.
--:-- Adam D'Angelo: Yeah, I really agree. Everyone read the book, and that changes what the theory even means. All the public market investors have read that book, and they now punish companies for not adapting and reward them for adapting, even if it means they have to make long-term investments. I think all the management leadership of the companies have read the book, and they're on top of their game. I think also the people running these companies are smarter than the companies from the generation that book was built on. They're at the top of their game, and a lot of them are founder-controlled, so it's easier for them to take a hit and make these investments. If you had an environment more like we had in the 90s, I think this would actually be more disruptive than the hyper-competitive world we're in now.
--:-- Host: One mistake that we as a firm have reflected on over the past few years is this idea that we've passed on companies because they weren't going to be the market leader or the category winner. Learning the lessons from Web2, you have to invest in the category winner. That's where things consolidate, and value accrues over time. So why do the next foundation model company if the first one already has a head start? But it seems like the market has gotten so much bigger that in foundation models, but also in applications, there are just multiple winners. They're fragmenting and taking parts of the market that are all venture scale. I'm curious if this is a durable phenomenon, but that seems like one difference from the Web2 era: more winners across more categories.
--:-- Adam D'Angelo: I think network effects are playing much less of a role now than they did in the Web 2 era. That makes it easier for competitors to get started. There's still a scale advantage because if you have more users, you can get more data and raise more capital. But that advantage doesn't make it absolutely impossible for a competitor of smaller scale. It makes it hard, but there's definitely room for more winners than there was before.
--:-- Adam D'Angelo: I think another difference is that people are seeing the value so strongly that they're willing to pay early on. The question with Web2 companies was, "How are they going to make money?" With Facebook and Google, it was, "Oh, how are they going to monetize?" The companies here are monetizing from the get-go. Your companies included.
--:-- Amjad Masad: Yeah, yeah. With the earlier generation of companies, monetization depended on scale. You couldn't build a good ad business until you got to millions, tens of millions of users. Now with subscriptions, you can just charge right away. Thanks to things like Stripe that make it easier, that's also made it a lot more friendly to new entrants.
--:-- Host: There's also questions of geopolitics. It seems clear that we're not in this globalized era, and perhaps it's going to get much worse. So investing in the foundation, in the OpenAI of Europe, might be a good idea. Similarly, China is an entirely different world. So there's a geo aspect to it.
--:-- Amjad Masad: Yeah, all of a sudden our geopolitics nerdiness is helpful.
--:-- Host: Adam, we were talking about human knowledge. Did you see yourself with Poe kind of disrupting yourself in a sense? Talk about the bet you made with Poe and the evolution there.
--:-- Adam D'Angelo: I think we saw Poe more as an additional opportunity than as disruption to Quora. The way we got to it was in early 2022, we started experimenting with using GPT-3 to generate answers for Quora. We compared them to the human answers and realized they weren't as good. But what was really unique was that you could instantly get an answer to anything you wanted to ask about. We realized it didn't need to be in public; your preference would be to have it be in private. So we felt like there was a new opportunity to let people chat with AI in private.
--:-- Adam D'Angelo: It was also a bet on diversity of model companies, which took a while to play out. But I think now we're getting to the point where there are a lot of models and a lot of companies, especially when you go across modalities. You think about image models, video models, audio models. The reasoning research models are diverging. Agents are starting to be their own source of diversity. So we're lucky to now be getting into this world where there's enough diversity for a general interface aggregator to make sense. It's surprising that even not particularly technical consumers actually use multiple AIs. I didn't expect that. People only use Google. They never looked at Google and then Yahoo. But now you talk to average people, and they'll say, "Yeah, I use ChatGPT most of the time, but Gemini is much better at these types of questions." The sophistication of consumers has gone up. People say that different AIs have different personalities, and they resonate with Claude more or whatever.
--:-- Host: I want to return to this point we said earlier, Adam, about dark matter, about how we're going to brute force a lot of knowledge that people have that isn't categorized yet. It's not just task knowledge; it's actually knowledge that you could ask them about and they could describe it. One question people have with LLMs is, "We've already trained the whole internet. How much more knowledge is there? Is it 10x? Is it 1,000x?" What is your intuitive sense of if we do brute force it and build this whole machine that gets all the knowledge out of humans onto a dataset that we can then implement? How do we think about the upside from there?
--:-- Adam D'Angelo: I think it's very hard to quantify, but there's a massive industry developing around getting human knowledge into the form where AI can use it. This includes things like Scale AI, Surge, Mercor, and there's a massive long tail of other companies just getting started. As intelligence gets cheaper and more powerful, the bottleneck is increasingly going to be on the data. What do you need to create that intelligence? That's going to cause more and more of this to happen. It might be that people can make more money by training AI, or that more companies get started. The economy is going to naturally value whatever the AI can't do.
--:-- Host: What is the framework for what it can't do? What is the mental model?
--:-- Adam D'Angelo: You could ask an AI researcher and they might have a better answer, but to me, there's just information that's not in the training set. That is inherently something that the AI can't do. The AI will get very smart. It can do a lot of reasoning. It could prove every math theorem at some point if it starts from axioms you give it. But if it doesn't know how this particular company solved this problem 20 years ago, if that wasn't in the training set, then only a human who knows that is going to be able to answer that question.
--:-- Host: Over time, how do you see Quora interfacing? How are you running these in parallel? How do you think about this?
--:-- Adam D'Angelo: Quora's focus is on human knowledge and letting people share their knowledge. That knowledge may be helpful for other humans, and it's also helpful for AI to learn from. We have relationships with some of the AI labs. Quora will play the role it is meant to play in this ecosystem, which is as a source of human knowledge. At the same time, AI is making Quora a lot better. We've been able to make major improvements in moderation quality, in ranking answers, and in improving the product experience. So it's gotten a lot better by applying AI to it.
--:-- Host: I'm going to talk about your future as well. Obviously, you had this business for a long time focused on developers. At one point, you were targeting the end tech market. I believe you had $2 or $3 million in revenue reported. And then recently, TechCrunch reported something like $150 million. I know it's higher. So you've had incredible growth as you've shifted the business model and customer segment. How do you think about the future of Replit?
--:-- Amjad Masad: Karpathy recently said that it's going to be the decade of agents, and I think that's absolutely right. Prior modalities of AI in coding were autocomplete with Copilot, then chat with ChatGPT, then Cursor innovated on the composer modality, which is editing large chunks of files. But that's it. I think what Replit innovated is the agent and the idea of not only editing code, but provisioning infrastructure like databases, doing migrations, connecting to the cloud, deploying, having the entire debug loop, executing the code, running tests. The entire development lifecycle loop happens inside an agent. That's going to take a long time to mature.
--:-- Amjad Masad: Our agent in beta came in September 2024. It was the first of its kind to do both code and infrastructure, but it was fairly janky. It didn't work very well. Then Agent V1 came around December, and it took another generation of models. You go from Claude 3.5 to 3.7. 3.7 was the first model that really knew how to use a computer, a virtual machine. Unsurprisingly, it was also the first computer use model. These things have been moving together. With every generation of models, we find new capabilities. Agent V2 improved on autonomy a lot. Agent V1 could run for like two minutes. Agent V2 ran for 20 minutes. Agent 3, we advertised it as running for 200 minutes. It just felt like it should be symmetrical, but it actually runs kind of indefinitely. We've had users running it for 20-plus hours.
--:-- Amjad Masad: The main idea there was that if we put a verifier in the loop, it could run longer. I remember reading a DeepSeek paper from NVIDIA about how they used DeepSeek to write CUDA kernels, and they were able to run DeepSeek for like 20 minutes if they put a verifier in the loop, like being able to run tests. I thought, "Okay, what kind of verifier can we put in the loop?" Obviously you can put unit tests, but unit tests don't really capture whether the app is working or not. So we started digging into computer use and whether it was going to be able to test apps. Computer use is very expensive and still very buggy. As Adam talked about, that's going to be a big area of improvement that'll unlock a lot of applications. But we ended up building our own framework with a bunch of hacks and some AI research. Replit's computer use testing model is one of the best. Once we put that into the loop, you can put Replit in high autonomy. We have an autonomy scale. You can choose your autonomy level. Then it just writes the code, goes and tests the applications. If there's a bug, it reads the error log and writes the code again. It can go for hours. I've seen people build amazing things by letting it run for a long time.
--:-- Amjad Masad: That needs to continue to get better. It needs to get cheaper and faster. It's not necessarily a point of pride to run for a lot longer; it should be as fast as possible. So we're working on that. Agent 4 has a bunch of ideas coming out. One of the big things is you shouldn't be just waiting for that one feature you requested. You should be able to work on a lot of different features. The idea of parallel agents is very interesting to us. You ask for a login page, but you could also ask for a Stripe checkout and then an admin dashboard. The AI should be able to figure out how to parallelize all these different tasks. Some tasks aren't parallelizable, but it should also be able to merge across the code. Being able to do collaboration across AI agents is very important. That way, the productivity of a single developer goes up by a lot. Even when you're using Claude Code or Cursor and others, there isn't a lot of parallelism going on. I think the next boost in productivity is going to come from sitting in front of a programming environment like Replit and being able to manage tens of agents, maybe at some point hundreds, but at least five, six, seven, eight, nine, ten agents all working in different parts of your product.
--:-- Amjad Masad: I also think UI and UX could use a lot of work. Right now, you're trying to translate your ideas into just a textual representation, like a PRD. That's what product managers do: product descriptions. But product descriptions are really hard to align on exact features because language is fuzzy. I think there's a world in which you're interacting with AI in a more multimodal fashion. You open up a whiteboard and you can draw and diagram with AI, really working with it like you work with a human. The next stage of that is having better memory inside the project but also across the project, and perhaps having different instantiations of a Replit agent. This agent is really good at Python data science because it has all the information, skills, and memories about my company and what it's done in the past. So I'll have a data analysis Replit agent, and I'll have a front-end Replit agent. They have memory over multiple projects and over time and over interactions. Maybe they sit in your Slack like a worker and you can talk to them. I could keep going for another 15 minutes about a roadmap that could span three to four to five years. But this agent phase we're in, there's so much work to do, and it's going to be a lot of fun.
--:-- Host: I was talking to one of our mutual friends, one of the co-founders of one of these big productivity companies. He leads a lot of their R&D, and he said, "Man, during the week these days, I'm not even talking to humans anymore. I'm just using all these agents to build." So living in the future is already in the present to some degree. There's something interesting about that. Are people talking to each other less at companies? Is that a bad thing? I'm starting to think more about these second-order effects. Will it make it awkward for the new grads? I feel so bad for them. If people are not sharing as much knowledge between each other, or it's not culturally easy to go ask for help because you should be able to use AI agents, there are some cultural forces that need to be reckoned with.
--:-- Amjad Masad: Yeah, I think a lot of tough cultural forces for Zoomers these days.
--:-- Host: Gearing towards closing here, obviously you guys are focused on running your companies, but to stay current on the AI ecosystem, you also make angel investments. Where are you most excited? We haven't talked about robotics. Are you bullish on robotics in the near term, or any emerging categories or use cases or spaces that you're looking to make more investments in?
--:-- Amjad Masad: I actually think vibe coding generally is just unbelievably high potential.
--:-- Host: You think it's under-hyped even still?
--:-- Amjad Masad: I think so. Opening up the potential of software to the mainstream of everyone. One reason I think it's under-hyped is that the tools are still very far from what you can do as a professional software engineer. If you imagine that they're going to get there, and I think there's no reason why they wouldn't, it'll take a few years. But then it's like everyone in the world is going to be able to create things that would have taken a team of 100 professional software engineers. That's just going to massively open up opportunities for everyone. I think Replit is a great example of this, but I think there will be other cases beyond just building applications that this creates.
--:-- Host: By the way, just on that note, if you were going to Stanford or Harvard today, 2025, would you major again in computer science or just focus on building something?
--:-- Adam D'Angelo: I think I would. I went to college starting in 2002, right after the dot-com bubble had burst. There was a lot of pessimism. I remember my roommate's parents told him not to study computer science, even though it was something he really liked. I just did it because I liked it. The job market is worse than it was a few years ago, but at the same time, having these skills to understand the fundamentals of what's possible with algorithms and data structures really helps you in managing agents when you're using them. I'm guessing it will continue to be a valuable skill in the future. The other question is, what else are you going to study? Every single thing you could imagine, there's an argument for why it's going to be automated. So you might as well study what you enjoy. This is as good as anything.
--:-- Amjad Masad: There's a lot to get excited by. One thing that gets me really fired up is seeing mad science experiments, like the DeepSeek OCR that came out the other day. It's wild. Correct me if I'm wrong, but basically you can get a lot more economical with a context window if you have a screenshot of the text instead of the text itself.
--:-- Adam D'Angelo: I'm not the right person to be correcting you on that, but there are definitely some really interesting things.
--:-- Amjad Masad: I saw another thing on Hacker News the other day about text diffusion. Someone made a text diffusion model by taking a single BERT instance and trying to mask different words and predict different tokens. We have a lot of components now. We have base pretrained models, RL reasoning models, encoder-decoder models, diffusion models. There are all these different things. Mix them in different ways. I feel like there isn't a lot of that. It would be great if a new research company came out that wasn't trying to compete with OpenAI but was instead trying to discover how to put these different components together to create a new flavor of these models. In crypto, they talk about composability and mixing primitives together. In AI, maybe there needs to be more experimentation. There's less playing around, I've found.
--:-- Amjad Masad: I remember in the Web 2.0 era, we were playing around with JavaScript and what browsers could do and what web workers could do. There were a lot of really interesting, weird experiments. Replit was born out of that. The original open-source version of Replit, pre-company, was about whether you could compile C to JavaScript. That was a nasty hack that became Wasm through Emscripten. I think we're in an era of Silicon Valley that is very get-rich driven, and that makes me a little sad. That's partially why I moved the company out of SF. The culture in SF has gotten... maybe during the dot-com era, a lot of people talked about how it was get-rich-fast or the crypto thing. I feel like there needs to be a lot more tinkering. I would love to see more of that and more companies getting funded that are trying to do something a little more novel, even if it doesn't mean a fundamentally new model.
--:-- Host: Last question. Amjad, you've been into consciousness for a long time. Are you bullish that we will, via some of this AI work or just some scientific progress elsewhere, make some progress and understand the hard problem? Something happened recently which is interesting. Claude 4.5 seemed to have become more aware of its context length. As it gets closer to the end of the context, it starts becoming more economical with tokens. Its awareness when it's being red-teamed or in a test environment jumped significantly. So there's something happening there that's quite interesting.
--:-- Amjad Masad: In terms of the question of consciousness, it is still fundamentally not a scientific question. We've given up on trying to make it scientific. This is also the problem I talked about with all the energy going into LLMs. No one is trying to really think about the true nature of intelligence or the true nature of consciousness. There are a lot of really core questions. One of my favorites is Roger Penrose's "The Emperor's New Mind." He wrote a book about how everyone in the philosophy of mind space and the larger scientific ecosystem started thinking about the brain in terms of a computer. In that book, he tried to show that it is fundamentally impossible for the brain to be a computer because humans are able to do things that Turing machines cannot do. Turing machines get stuck on things like basic logic puzzles that we can detect, but there's no way to include that in a Turing machine. For example, "This statement is false." It's a complicated argument. If you read that book or many others, there's a core strain of arguments in the theory of mind about how computers are fundamentally different from human intelligence. I haven't really updated my thinking too much about that because I've been very busy. But I think there's a huge field of study there that is not being studied.
--:-- Host: If you were a freshman entering college today, would you study philosophy?
--:-- Amjad Masad: I would do that. I would definitely study philosophy of mind. I would probably go into neuroscience because I think those are the core questions that become very important as AI continues to take more of the jobs and the economy.
--:-- Host: That's a great place to wrap. I'm John. Adam, thanks for coming on the podcast.
--:-- Adam D'Angelo: Thank you.
--:-- Amjad Masad: Thank you.
--:-- Host: Thanks for listening to this episode of the a16z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at a16z and subscribe to our Substack at a16z.substack.com. Thanks again for listening, and I'll see you in the next episode. As a reminder, the content here is for informational purposes only. Should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any a16z fund. Please note that a16z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures.
