Marc Andreessen, Chris Dixon and Ben Horowitz on AI and the Frontier for Founders

Chris Dixon announces that a16z is expanding its crypto fund into a16z Frontier. One reason is that crypto has become part of the broader technology landscape: stablecoins are entering fintech toolkits, and financial firms are exploring tokenized assets. As these uses spread, the boundaries between crypto, fintech and other industries have become less clear.

Dixon says the firm also saw room to invest in what his partner Martin calls the ‘outer planets’: unconventional opportunities that do not fit neatly inside established investment groups. Those groups already have many compelling opportunities in their core areas, leaving a possible gap in coverage for ideas further afield.

For Dixon, the new mandate continues work he began after joining the firm in 2013. He looks for second-order effects of technological shifts. Smartphones, for example, enabled apps, but they also created a supply chain of better sensors and chips that entrepreneurs could recombine into products such as Oculus and Skydio.

Dixon also describes crypto as connected to the evolution of open-source software and the communities made possible by social media. Frontier, in his telling, builds on the firm's long history of backing such technologies rather than representing a sudden change in direction.

Ben Horowitz says the firm has historically developed dedicated verticals after identifying promising new areas. That strategy has worked, but new ideas can still fall outside those categories, and the process for turning them into a practice has not always been defined.

Frontier gives the firm a way to back founders in those emerging areas. If a category becomes large enough, it could eventually support a separate fund, or Frontier itself might evolve around it. Horowitz emphasizes that the firm invests in entrepreneurs as much as in categories or products, and some of the most interesting founders work beyond established boundaries.

Asked what drew them to Dixon, Horowitz recalls his original ideas and writing, as well as his experience as an entrepreneur. Dixon understood what the founder's journey felt like, an experience Horowitz says mattered when the firm was establishing its culture.

One influential Dixon framework was to look for good ideas that initially look like bad ideas, where the entrepreneur knows something others do not. Horowitz says this helped the firm distinguish potentially unique investments with substantial upside from opportunities whose profits might be competed away.

Marc Andreessen adds that Dixon's team combines deep technical expertise with business and market judgment. Crypto investing, he argues, requires understanding complex software, cryptography and game theory beneath the industry's more visible cultural phenomena. The expanded mandate lets that team apply its experience more broadly.

Dixon points to another transferable capability: policy and regulatory work. Drones, autonomous vehicles and AI all raise public-policy questions, and he expects experience gained through the crypto industry's regulatory battles to help founders in other frontier technologies.

Discussing the firm's history, Dixon says Andreessen and Horowitz have consistently encouraged ambitious ideas. He describes a culture willing to reconsider its structure and bring in new ideas when they can be defended. Having previously been a customer of venture capital as a founder, he valued the possibility of a better service for entrepreneurs.

Horowitz recalls a phrase attributed to J. P. Morgan about doing first-class business in a first-class way. He connects it to the importance of being principled and clear about one's identity in venture capital, so that people know where the firm stands and can trust it.

The group notes that companies that look successful from the outside usually have difficult periods too. Erik Torenberg then asks how investors should think about being too early, a familiar risk in venture investing.

Dixon says many technologies that eventually worked had earlier versions that failed, sometimes by a decade. Timing depends on technology, culture, business conditions and politics coming together. Frontier intends to err on the side of being early and accept that some emerging categories may not work.

Virtual reality illustrates the uncertainty. Oculus found a buyer, but more than a decade later VR has not become as mainstream as Dixon hoped, in part because many hard technical problems remain. He still believes in the category and would like to invest in it.

Dixon argues that AI could force investors to reassess technical risk. As capabilities improve in software engineering and mathematics, he sees a possibility that progress in other fields could be pulled forward. A project once dismissed as a science problem rather than an engineering problem may need a fresh evaluation.

He suggests that a small team might now credibly attempt a full-stack device company that once required far more people. Founders must still think about defensibility: a product sitting just ahead of rapidly improving AI capabilities could be overtaken. But he sees many unexplored ways to use those capabilities.

Andreessen says he sees an unusually wide gap between what early adopters understand AI can do and what most individuals and incumbent companies believe is possible. In his view, AI-native startups have a very different sense of their capabilities, and the gap keeps widening as the technology advances.

The implication, Andreessen argues, is that founders may enter industries with much more technically sophisticated product plans and broader ambitions. He jokes that if Netscape had had a tool like today's AI during its early years, he and Horowitz might have built far more of the internet themselves.

Those founders need not define their companies as purely software, hardware or AI businesses, Andreessen says. They may combine several domains. Horowitz adds that the question for an investor is shifting from whether an idea is too ambitious to whether it is ambitious enough to stand out.

Andreessen expects some incumbent companies to adapt, but warns that change is moving quickly. He says even someone following developments full time can struggle to keep up, while young founders equipped with AI may accomplish things that once required much larger teams.

Torenberg asks whether AI will primarily strengthen incumbents or disrupt them. Andreessen observes that individuals can adopt the tools faster than institutions: employees may use AI for their work without their employers fully understanding the change. Some of those individuals may discover that starting a company has become more feasible.

Dixon suggests that the usual frame of technology companies competing with one another is too narrow. AI-enabled teams can build full-stack businesses in other sectors of the economy. The opportunity is not confined to a contest among established technology firms.

Asked which layer of AI will capture value, Dixon says activity is spread across major labs, open models, hardware and applications. He expects the distribution of value to change over time, but sees demand as early relative to the potential market for intelligence.

Dixon compares the early AI market with the early internet. The first wave often uses a new technology to improve familiar activities, like moving a catalog online. Later waves can depend on new behavior: YouTube, for example, required creators and audiences to develop around a platform that began empty.

The participants expect similarly unfamiliar AI-native products. Dixon speculates that agents could become numerous participants online and need ways to communicate, exchange knowledge and transact. The group points to early experiments as hints of a wider agent ecosystem, while acknowledging that its shape remains uncertain.

Andreessen criticizes what he calls ‘product manager brain’: early products often reflect the lives and needs of their makers. Restaurant reservations and travel booking have their place, but he expects AI applications to become more diverse, entertaining and creative as more people experiment with them.

He points to a recent surge of AI-assisted music videos and game experiments. Dixon describes games as a field that often moves quickly because of its technical culture, citing experiments in porting and combining games as examples of new behaviors emerging around the tools.

The participants predict a creative renaissance in games. Dixon argues that smaller independent teams may be able to work at a scale that previously required much larger organizations. He sees the technology as extending human creativity, though the discussion also acknowledges resistance within the industry.

Andreessen applies a similar argument to film: if production costs fall dramatically, projects that are financially difficult at today's budgets could become viable. His example is a movie he considers creatively impressive but commercially expensive; lower costs, he argues, could change the economics of experimentation.

Hardware is another domain Andreessen expects AI to open up. He describes reverse-engineering tools that can help people understand and modify existing devices. His example is home automation, where an AI system might eventually help a homeowner make disparate devices work together.

For Andreessen, the broader possibility is that people with software skills could take on hardware problems that were previously prohibitive. He and Dixon see effects extending beyond technology products into many parts of the economy.

Horowitz cautions that AI-native applications will be hard to predict precisely because they are new. He expects AI to make conventional films more efficient, but also thinks it may enable a different kind of media and stories that were previously impractical to create.

Dixon compares the change to photography. A camera could threaten some existing forms of visual work, but it also helped enable motion pictures. Similarly, AI may alter established media while allowing people to invent forms that are difficult to describe in advance.

Torenberg asks whether AI necessarily centralizes power, citing a contrast Peter Thiel drew between crypto and AI. Dixon sees several possible market structures: large proprietary labs may succeed alongside open models and, eventually, models running locally on people's own devices.

Dixon also imagines crypto infrastructure serving an economy of agents, with stablecoins as one possible means of payment and other financial services built on top. He presents this as an open question, not a settled outcome. His preferred future combines powerful companies with open-source software and open-weight models that give consumers and startups choices.

Andreessen says open models are advancing in both the United States and China, while warning that governments could impose restrictive rules. His argument for openness includes education: students and independent researchers should be able to study how AI works, understand its problems and learn to build it themselves.

The group also discusses practical access to local models. Andreessen imagines someone carrying a model on a phone when there is no internet connection. He contrasts that possibility with dependence on a small number of cloud providers, while treating regulation and future concentration as continuing concerns.

Horowitz argues that concentrated power is dangerous and that open access could also help with safety research. He raises reward hacking, where a system pursues a reward in an unintended way, and asks whether a wider community of researchers with access to model weights might help find solutions.

Dixon describes an unwritten internet bargain: publishers let platforms index or display their content, and platforms send readers back through links. He says that bargain has weakened as search engines answer questions directly and social platforms keep attention on their own services.

The resulting decline in visits can threaten sites that rely on advertising or conversions. Dixon expects people still to book travel and seek information, but perhaps through agents or other intermediaries. He sees near-term pain for existing businesses alongside an opportunity to rebuild online services and their economics.

On devices after the smartphone, Dixon expects renewed interest in wearables as AI and voice interfaces improve. He remains optimistic about virtual reality, while noting that weight and comfort remain obstacles. AI-generated content, he suggests, could make it easier to build immersive experiences.

Dixon also hopes for a revival of homebrew hardware. He describes combining AI-assisted design, 3D-printed cases, inexpensive boards and custom firmware. The group extends the idea to architecture and manufacturing, where AI could help with work that has historically demanded specialized teams.

Dixon argues that AI is still largely in a single-player phase: a person types into a computer and receives a response. Earlier internet and mobile products later developed social and collaborative layers. He expects AI could similarly move toward networks of people and agents working together.

A collaborative fleet of agents might tackle a difficult scientific problem or help design a building. Dixon also mentions early experiments in trust relationships between agents. He treats these as promising signs rather than proof of a finished product category.

Asked what an 18-year-old should study, Dixon borrows Andreessen's phrase ‘modern alchemy’: humans have learned to turn sand into intelligence. He suggests mathematics and computer science, perhaps economics, and believes it will be valuable to work with the tools even outside a purely technical career.

The group expects AI to affect many disciplines, including classics, economics, biology, design and the arts. They discuss examples ranging from reading ancient scrolls to studying animal communication. Their point is that students may find new opportunities within existing fields as well as in AI itself.

Andreessen recalls Douglas Adams's observation that people experience new technology differently depending on their age. Children may regard a computer that answers questions as normal, while older generations experience the same capability as a disruption. He thinks young people will find uses adults have not anticipated.

Torenberg summarizes the conversation as a form of technological optimism: the firm is willing to bet early even though some bets will arrive before their time. Dixon adds a qualification. The specific ideas discussed are hypotheses; the firm's job is to find exceptional entrepreneurs with their own ideas and back them with intellectual humility.