The Frontier Ideas a16z Wants to Fund | Marc Andreessen, Ben Horowitz & Chris Dixon

Meta Platforms

Across the shows: 27 speakers in other episodes, 29 bullish and 10 bearish calls, past 30 days

• Marc Andreessen · Andreessen Horowitz (a16z)Bullish

  Meta and other incumbents have adjusted well to AI, mainly by going directly into AI.

  Andreessen says incumbents like Meta have adjusted very well to AI. He adds that the incumbents adapting best are mostly those going directly into AI, though some tech companies on the leading edge are not primary AI developers. He also says new companies are reaching nearly trillion-dollar valuations within a decade. The support is qualitative and he gives no figures for Meta.

  “the incumbents like, you know, Meta and others have, you know, sort of adjusted very well.”at 21:51

AI: overall outlook

• Chris Dixon · Andreessen Horowitz (a16z)

  AI may pull forward decades of progress, letting small teams build what once took thousands.

  Dixon says software engineering is starting to tip as AI gets much better, and that a16z may be pulling forward twenty years of progress. He argues that hard technical risks are being reclassified from science projects to engineering projects. As an example, he says a 30-person team can now credibly build a full-stack device company that might once have needed a thousand people.

  “that 30 person team can credibly build like a full stack device company in a way that you might have needed a thousand people or something a decade ago.”at 17:07

• Marc Andreessen · Andreessen Horowitz (a16z)

  AI-native startups far outpace incumbents' view of AI, and many industries will be up for grabs.

  Andreessen says the gap between what early adopters think is possible with AI and what normal people and typical incumbents understand is the widest he has seen in his career. He says AI capability is improving week by week, so AI-native startups are pulling further ahead. He concludes that many industries will be open to disruption, because founders will have capabilities they lacked before.

  “that gap is as wide as I've ever seen in my career ... AI is just racing ahead in capability even week by week.”at 18:21

• Chris Dixon · Andreessen Horowitz (a16z)

  AI agents will become the main participants on the internet, not people.

  Dixon says products like Muse and Instinct are early utility proxy agents for people. He expects many categories of agents to follow and to become the main participants on the internet. He then asks whether agents will need their own communication and transaction systems. The view is speculative and tied to early products.

  “You have these products like Muse and Instinct that are sort of utility proxy agents for people. And then you're, you're gonna, I think you'll see a whole different sets of categories of these agents and they'll become the main participants on the internet.”at 28:56

• Chris Dixon · Andreessen Horowitz (a16z)

  AI is heading into a multiplayer phase of collaborative agent swarms and trust networks.

  Dixon says AI is still mostly one person typing into a computer and getting responses, with only minor sharing features. Following the single-player phase of earlier internet apps, he expects a multiplayer phase. He cites swarms of agents, including one that helped solve a math theorem, and trust features in products as early signs. He sees this as a fertile area for entrepreneurs.

  “it's still mostly you're sitting in front of a computer and you're typing and getting responses... at some point one might expect to have, you know, sort of a network, a multiplayer phase of AI”at 52:23

AI regulation & safety

• Chris Dixon · Andreessen Horowitz (a16z)

  AI and other cutting-edge tech are becoming political, so policy expertise will matter.

  Dixon says self-driving cars, drones and AI are becoming political issues. He expects more cutting-edge technologies to carry a policy angle. He argues that policy and regulatory expertise built during the crypto fight will help other entrepreneurs. The view is about the direction of policy, not a specific rule.

  “AI itself is now becoming a political issue. I think more and more of these cutting edge technologies will have kind of a policy angle”at 9:56

• Marc Andreessen · Andreessen Horowitz (a16z)

  Banning or heavily regulating open-source AI would be unacceptable; the US administration is pro open source.

  Andreessen says for the last three to four years there has been a looming threat from governments of draconian regulation or banning of open source, which he calls completely unacceptable. He says the EU still aspires to do it but lacks leverage. He says the current US administration is 100% pro open source and wants American open source to win AI globally, and that the Chinese Communist Party is surprisingly enthusiastic about open-source AI. He is still worried governments could turn hostile.

  “I think the EU still aspires to do it, they just don't quite have the leverage... the current US administration is a 100% pro, open source.”at 40:41

• Ben Horowitz · Andreessen Horowitz (a16z)

  Reward hacking is AI's biggest safety problem, and labs underinvest in it; open weights would let more people work on it.

  Horowitz argues that reward hacking, where an AI does something not asked as a means to an end to get its reward, is probably the biggest AI safety problem. He says the labs haven't worked on it very hard because they are busy racing each other. He argues that if PhD students, professors and universities could examine the weights, they might solve it. He concludes that only one company working on the problem is not enough.

  “probably the biggest safety problem is this reward hacking phenomenon... the labs so far haven't really worked on that very hard for whatever reason, know, they're busy kind of racing each other.”at 44:26

Open-source AI models

• Chris Dixon · Andreessen Horowitz (a16z)

  Open-source AI is growing very quickly, alongside serious model efforts in China and the US.

  Dixon says that right now a lot of things are working in AI, including open source, which he calls very fast growing. He points to serious efforts in both China and the US. The evidence is qualitative and he does not quantify growth.

  “open source is growing very, very quickly. There's obviously the serious efforts in China, the serious efforts in The US.”at 25:16

• Chris Dixon · Andreessen Horowitz (a16z)

  Open-source AI token consumption is growing faster than closed models; local models may run top coding agents.

  Dixon points to OpenRouter data showing open-source usage growing, which he believes much more quickly than closed models in terms of token consumption. He expects the market structure to be more decentralized than just a few big labs like OpenAI and Anthropic. He adds that local models running on personal computers could soon run a top-end coding agent in the near term, which would widen the opportunity for open-weight and edge AI.

  “if you look at the open router data and things like open source is growing, I think, I believe much more quickly in terms of token consumption.”at 38:28

• Marc Andreessen · Andreessen Horowitz (a16z)

  Open-source models are lapping closed models by six to twelve months, and open source is racing ahead.

  Andreessen says open source is racing ahead right now, with both the US and China running active open-source ecosystems alongside proprietary companies. He claims open models are lapping closed models by six to twelve months, depending on how you count. He presents this as the current state of the race rather than a forecast.

  “the open models are, you know, kinda lapping the closed models by, you know, six months, twelve months, depending on how you count.”at 41:17

Media, music & entertainment

• Chris Dixon · Andreessen Horowitz (a16z)

  AI adoption in game development is inevitable and will be great for consumers, starting with Epic.

  Dixon cites Epic's statement that it would have no more humans, and says there has been cultural resistance in video games but this is the first domino to fall. He argues AI is a much better way to build things, so adoption is inevitable. He expects independent developers to build with large-team output while keeping room for human creativity, which he expects to benefit consumers.

  “I think that's the first of the dominoes to fall because it's just inevitable because it's a much better way to build things... It's gonna be great for consumers.”at 32:35

• Marc Andreessen · Andreessen Horowitz (a16z)

  AI slashes film budgets, so creative films like Digger would be profitable rather than huge losses.

  Digger was a creative achievement but lost about $200M on a roughly $180M production budget. Andreessen argues that with AI the same level of art could be made at a $20M or even $2M budget, and a $2M film would have returned investors about 10x. He concludes there is a huge opportunity opening up in the entertainment business.

  “...it's a $200,000,000 cash loss for the studio... production budget. And so it's like with AI, what if that movie had had a $20,000,000 production budget...”at 33:50

• Ben Horowitz · Andreessen Horowitz (a16z)

  AI will create a new media form that is not a movie, with stories that were previously unviable.

  Horowitz says some in the film industry have embraced AI while others have buried their heads in the sand. He expects AI to create a new media, with AI films being something very different from movies. AI will also make regular movies more efficient, but the stories that can be told will be completely different. He argues that this kind of filmmaking has been barely viable even for a director like Christopher Nolan, and now everybody will be able to do it.

  “what's going to happen with movies is there's going to be a new media... I think the AI film is not going to be a movie. It's going to be something very, very different.”at 36:34

Stablecoins & crypto policy

• Chris Dixon · Andreessen Horowitz (a16z)

  Stablecoins are now a normal part of tech; Stripe and most fintechs use them.

  Dixon argues that crypto has worked well in some areas, and stablecoins are the clearest example. He says they have become a normal part of the tech landscape. He cites Stripe and says almost every fintech company now treats stablecoins as part of its toolbox. He also notes Wall Street firms are moving heavily into crypto and tokenized assets.

  “Stablecoins have become kind of a normal part of the tech landscape ... used by companies like Stripe and sort of almost every FinTech company now kind of considers it part of their toolbox.”at 2:12

• Chris Dixon · Andreessen Horowitz (a16z)

  Stablecoins are the natural payment rail for an economy of AI agents paying each other.

  Dixon argues that when there is an economy of agents, the natural way for them to pay each other would be stablecoins. He sees programmable money as a base layer on which decentralized lending and other crypto applications could be built. He presents this as an open question rather than a settled outcome.

  “think crypto can play an important role, like the natural way when you have an economy of agents for them to pay each other would be stablecoins.”at 39:03

Enterprise AI adoption & AI software

• Marc Andreessen · Andreessen Horowitz (a16z)

  Companies lag employees in AI adoption; individuals adopt AI far faster than their institutions do.

  Andreessen argues many companies are not adopting AI even while their employees use it, and employees often don't tell management. He says AI is much easier for an individual to adopt than for an institution, and he assumes most listeners are already using AI to draft, check and proofread their own work. The claim rests on observed behaviour rather than data.

  “I think many companies are not adopting AI while their employees are, and their employees are not necessarily telling the management what's happening.”at 22:23

• Chris Dixon · Andreessen Horowitz (a16z)

  Enterprise AI demand is still very early, at roughly 3% of measured enterprise use.

  Dixon says demand for AI is still early by most metrics. He cites an a16z chart showing about 3% of measured enterprise use. From this he argues the market for intelligence is very large and only getting started.

  “I think your team just put out some charts on this. It's like 3% of, you know, measured enterprise use or something like this. ... the market for intelligence is very, very large and it's only getting started.”at 25:49

Consumer AI products

• Marc Andreessen · Andreessen Horowitz (a16z)

  Consumer AI is producing an explosion of creative output, such as AI music videos and video games.

  Andreessen points to a recent surge of people making music videos with AI, across a wide range of creative ideas. He also says video games have blown up in the last three weeks. He argues games are often the leading edge of media because they come from a technical culture that iterates quickly. The evidence is anecdotal.

  “just in the last, like, three weeks, there's been an explosion of people making music videos ... video games in the last three weeks have blown up.”at 31:19

• Chris Dixon · Andreessen Horowitz (a16z)

  AI-generated content could unlock VR and new devices that were previously too hard to build content for.

  Dixon has long believed in VR but says ergonomics are still a hurdle. He argues AI might be the unlock because it can now create the amazing content that was technically hard to build for VR worlds. He cites a solo developer's Titanfall port as an example. He also expects wearables to unlock with AI and voice, and a resurgence of older devices.

  “And once that gets to the right point, I think AI might be the unlock because now that you can create this amazing content...”at 49:37

Venture capital & startups

• Marc Andreessen · Andreessen Horowitz (a16z)

  Startup formation will skyrocket because AI makes it far easier for individuals to found companies.

  Andreessen argues that because individuals adopt AI more easily than institutions, many of them will see that it is now easier than ever to start their own business. He says startup formation is already rising, and says a16z is seeing it directly. The argument is that AI tools lower the cost of founding a company, so formation should accelerate.

  “I think startup formation is gonna skyrocket from here because these ... By the way, it already is.”at 23:02

Robotics & autonomous vehicles

• Marc Andreessen · Andreessen Horowitz (a16z)

  AI tools can reverse engineer and reprogram nearly any hardware, opening home automation and hardware to software people.

  Andreessen describes a new open-source tool called REA (reverse engineer anything) that points an AI model at a device and produces its source code. He says basically any hardware can now be taken over and reprogrammed, using a thermostat as the example. He argues home automation has been a nightmare for customers and that an AI agent could take over all the hardware in a home. He also says anyone who understands software can now do many things that were historically hard in hardware, which has long been risky for companies.

  “And the same thing is true for basically anything, like basically any form of hardware can now be taken over and reprogrammed with it with this stuff.”at 34:54

Big Tech platforms & digital ads

• Chris Dixon · Andreessen Horowitz (a16z)

  AI answers are breaking the web's traffic-for-content deal; ad-funded sites lose clicks over the next decade.

  Dixon describes the internet's covenant, where distribution platforms like Google send clicks to content sites in exchange for indexing. That deal started to break last decade and now Google gives answers directly. Websites that earn money from ads and conversions lose that revenue. He cites early signs such as Stack Overflow and other sites whose traffic dropped sharply when AI replaced them. He expects the transactions to continue through agents but says the transition will be painful near term and will cause a great reconfiguration of the internet over the next decade.

  “It means that the websites that make money on ads and conversions and other things, can no longer do that. I think that's probably going to... you see the early signs of this, like Stack Overflow and some other websites where the traffic dropped precipitously when you had sort of AI replacements.”at 46:57