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

Expanding Beyond Crypto: Why a16z Frontier Exists

The firm is widening the mandate of a16z crypto into a16z Frontier, a home for founders building at the edges of what is technically possible. The reasoning is twofold. First, crypto has matured into ordinary infrastructure: stablecoins are now a standard part of the fintech toolbox, Wall Street is moving into tokenized assets, and companies sitting at the intersection of fintech, traditional finance, and crypto no longer fit neatly into a single category. Second, the firm’s existing verticals are deep and busy, which leaves a gap for the “outer planet” opportunities that do not yet have a natural home. The term comes from Martin Casado’s framing, borrowed from Firefly, where inner planets are core markets and outer planets are the stranger, earlier bets.

Chris Dixon describes the thesis as a continuation of his long-running interest in second-order effects. Smartphones went mainstream around 2013, and the first-order wave was apps. The more interesting wave came from the components themselves: cheap sensors and chips that could be recombined into drones, VR headsets like Oculus, and consumer robotics. Crypto, in his telling, was a downstream effect of open source software turning into services and of social media enabling communities to form outside mainstream acceptance around blockchains.

Ben Horowitz frames the move as filling an architectural hole. The vertical strategy has worked well, but novel categories that fall between verticals have been handled in an undefined way. Dedicating a motion to these areas could eventually become a full fund, or Frontier could itself become that structure. Marc Andreessen adds that the firm has built deep expertise in policy, which will matter as self-driving cars, drones, and AI become political issues.

The Early-Adopter Gap and the Cost of Being Early

The conversation returns repeatedly to a gap between what early adopters believe is possible and what everyone else understands. Chris describes this as the widest he has seen in his career, and says it is widening week by week as AI capability keeps advancing. AI-native startups operate with a fundamentally different view of what they can build than incumbents do in almost any industry.

On timing risk, the partners agree that nearly every technology idea that was good has eventually worked, but often after a precursor failed a decade early. Oculus is the cautionary example: sold, and thirteen years later VR still has not reached mainstream adoption because the hard technical problems took much longer than expected. The firm’s stated position is that Frontier should err on the side of being too early and accept category risk. Chris argues that AI may be pulling two decades of progress forward, with software engineering improving faster than math and hardware now within reach of small teams. A thirty-person team can credibly build a full-stack device company that once required a thousand people.

Marc points to a related bottleneck: the individual adopts AI far faster than the institution. Employees are already using these tools to draft, fact-check, and write code, while their companies lag. This asymmetry, in his view, will drive a sharp rise in startup formation, which he says is already happening.

Incumbents, Disruption, and the Shifting Frame

The group questions whether AI will sustain incumbents or favor startups. Their current read is that both are happening. Large labs and established companies are adapting, but a new wave of AI-native companies is compounding quickly. Chris argues that the Christensen disruption frame, which assumes competition within a contained tech world, is probably the wrong lens now. The more accurate picture is technologically modern companies entering other industries and reshaping the broader economy. Full-stack companies, such as the food service business Adams that the firm recently backed, illustrate this pattern.

Marc suggests a broader pattern in technology adoption. Every new platform first gets used for existing behaviors: catalogs moved onto websites, documents moved into apps. The truly transformative shifts come when people invent new behaviors. YouTube’s early bet was partly on a cultural change, that influencers would emerge, which required more risk than simply digitizing something known to work. The current AI phase, in his view, is still dominated by re-doing old tasks such as coding, math, knowledge work, and search, but the radical, native uses are beginning to appear.

Agents, Social Networks, and Product Manager Brain

Chris expects the most numerous participants on the internet to become AI agents rather than people. Early products like Muse and Instinct act as utility proxies for their users, and new categories of agents will need their own ways of communicating, transacting, and sharing knowledge. Moltbook and experiments in which agents coordinate among themselves hint at the ecosystem to come.

Marc is more skeptical of the current generation of consumer products, which he calls “product manager brain”: products designed by product managers for product managers, with use cases like booking restaurants for dinner with other product managers. He also criticizes the cautious, overly restrained persona of today’s assistants. The next wave, he argues, will open up entertainment, creator economies, new art forms, and income opportunities that current designers have not imagined.

Games, Film, and Hardware as Frontier Domains

Video games are a leading indicator for media because they are a technical culture that iterates quickly. Recent examples include mashups of two games into a combined one and the ability to disassemble and port games with AI, letting some run faster on Unix than on Windows. The group sees cultural resistance in the industry, including a recent statement from Epic, as likely to erode quickly because AI is a much better way to build and will benefit consumers.

Film presents a different economic case. Marc cites Digger as a creatively ambitious movie with a production budget around $180 million that lost money, and argues that an AI-enabled production at a fraction of that cost could be highly profitable and still deliver real artistry. Ben Horowitz goes further: the AI film will not look like a movie at all, but a new media form with stories that were previously unviable outside a few filmmakers like Christopher Nolan.

Hardware has long been treacherous terrain for software people. Marc describes a recent open source tool that uses an AI model to reverse engineer any piece of hardware and produce source code that can then be reprogrammed. A thermostat, a home automation system, or a device with a poor interface can now be taken over and rewritten. Home automation, historically a nightmare for customers, might become an AI agent that controls all of a home’s hardware. The broader point is that anyone who understands software may be able to do many things in hardware that were previously inconceivable.

Chris extends the same logic to architecture and manufacturing, describing an architectural design steered by AI and models that can produce design-for-manufacturing specifications, traditionally one of the hardest steps.

Centralization, Open Source, and Who Gets to Learn

The group revisits Peter Thiel’s line that crypto is libertarian and AI is communist. Chris argues that the market structure of AI is still open. Large labs like OpenAI and Anthropic will remain, but open source is growing rapidly by token consumption, and local models running on personal hardware could decentralize much of the value. Crypto, in his view, could provide a base layer of programmable money, with stablecoins as the natural settlement layer for autonomous agents paying each other, on which decentralized lending and other primitives could be built.

Marc is optimistic about the present state of open source, noting that open-weight models trail closed ones by six to twelve months and that the current administration is supportive. He is wary of regulatory threats, including a looming EU push to restrict open source, and he argues that the deeper stakes are educational and civic. Every kid, he says, should be able to learn how these systems work, and AI itself is one of the best teachers of AI, though some proprietary models refuse to explain their own internals. Open source also enables defensive uses, from vulnerability scanners to offline models for wilderness survival where connectivity is unavailable.

Ben adds that centralization is historically a source of harm, and that safety problems such as reward hacking may need many researchers working in parallel rather than a handful of labs. Open source is the mechanism that makes that broad participation possible.

The Internet’s Broken Covenant

Chris describes an informal bargain that has underpinned the web: distribution platforms like Google and Facebook sent traffic to content creators in exchange for indexing and content. That covenant is breaking. Google increasingly answers queries directly, and social networks keep users on platform rather than linking out. Websites dependent on ads and conversions are suffering, and Stack Overflow is an early example of traffic dropping sharply once AI alternatives appeared. Transactions will still happen, perhaps through agents that visit travel sites on a user’s behalf, but the flow of attention and economics will be reorganized. He expects the next decade to bring a deep reconfiguration of how attention flows, how knowledge is shared, and how organizations interact online, and describes this as more dramatic than what mobile did.

What Comes After the iPhone

Wearables are an obvious area, now unlocked by AI and voice. Chris has long believed in VR, arguing that the form factor and ergonomics still need work, but that AI may unlock the content that makes it compelling, including recent community ports of games like Titanfall. He also points to brain-computer interfaces, drones, and self-driving cars as parts of the next interface landscape.

He anticipates a resurgence of homebrew hardware. Tools like Claude can design 3D-printed enclosures, cheap boards can be paired with custom firmware, and the maker movement that stalled over a decade ago may finally become practical. Ben notes that AI-native use cases in devices and networks may depend on multiplayer rather than single-player design. Early iPhone apps were single-player utilities; social and commerce networks came later. AI today is still largely one person at a keyboard. Swarms of collaborating agents solving hard problems, from building design to scientific research, may be the next multiplayer phase, and trust networks like those emerging in Muse and Instinct point toward it.

What to Study at Eighteen

Asked what they would study at eighteen, the group converges on a shared answer: math, computer science, and possibly economics, because the central fact of the moment is that humans have learned to turn sand into intelligence. Chris argues that AI will be the dominant narrative of the next two decades, and that being at the cutting edge, even without being purely technical, is the obvious move. Fields across the spectrum are being transformed, from classics, where AI is helping decipher the Herculaneum scrolls, to economics and biology, where design and discovery are accelerating.

Douglas Adams’s generational framework captures the reception of new technology: children treat it as the natural order, people from fifteen to thirty-five see an exciting career opportunity, and anyone over thirty-five regards it with alarm. Chris describes teaching his seven-year-old that a computer that answers questions is simply normal. Schools banning AI may inadvertently make it more appealing to students, who will treat it as a secret superpower. Marc adds that the adults worrying about what children should do may be the ones who are obsolete, because the next generation will enter fields that were previously closed to them.

The conversation closes on the idea of definite optimism, tempered by humility. The firm’s bet is on great entrepreneurs, and ideas outside its core verticals remain welcome. Ben and Marc emphasize that the firm has been wrong many times, which is why it wants to back people with their own conviction that the future is coming and will reward those who build it.

• Tags:
• AI Agents
• Open-Source AI
• Stablecoins
• A16z Frontier
• AI-Native Startups
• Hardware Reverse Engineering
• AI and Entertainment
• Venture Capital Strategy
• Internet Business Models
• AI and Education