Ben Horowitz

Ben Horowitz says the thing founders do not have is knowing how to do the CEO job. The founders he works with who have not already run giant companies do not know it, and the skill is complicated. They still have to hire for jobs they have never done. He had never been a CFO. He did not know what a control structure was or how to talk to Wall Street. He compares that hire to choosing a Japanese interpreter when you do not speak Japanese: they all sound good. People also have their own career ambitions, and Mark Zuckerberg may be offering the team a billion dollars to leave. Founders who fail at this often hit a crisis of confidence. They know they do not know what they are doing, they know i

Ben Horowitz says the venture industry was basically the same from 1970 to 2008 and 2009, because firms were built not to change. a16z started in 2009. The internet was the first big break: companies got easier to build. A computer company used to need manufacturing, professional services, and a direct sales force before the first product sold. Twitter was five people with laptops and AWS. AI is the next step, so the venture landscape is almost certainly going to change again.

Older bets were expensive and very likely to fail. The hit rate for promising companies went up. What stays is momentum. Two companies can have the same idea and two strong teams, and the one that gets more momentum w

Ben Horowitz says the software application may be the thing that fades the most. A person may just ask an AI to do the work, with no application in between. The arrangement he describes is to organize the data and then have agents do things against that data. He thinks that is how both business applications and personal applications would work.

He does not treat the shift as settled. Software can also get so easy to build that companies ship copies the way fast fashion gets copied, or that each person gets a slightly different version generated at runtime and personalized to them. He calls that second version science fiction that might arrive, and he notes that people still wear similar clo

Ben Horowitz tells Andrew Chen that a very small team can reach a giant audience and a ton of money, and he is sure about that part. Instagram had 13 people and sold for a billion dollars. He thinks Instagram would probably have ended up with a large staff even if Meta had not bought it, and Meta now has a ridiculous number of employees. Ray Kurzweil predicted that everybody would be a solo entrepreneur, and that outcome could still arrive.

The harder question is whether the tiny team can sustain the lead or gets taken out by someone doing more things. People always want more, and the standard keeps rising. John Maynard Keynes wrote in 1970 that once people had shelter, food, and clothes th

Ben Horowitz says the 2007 iPhone was still a breakthrough because a new product only has to be unbelievable at one thing. Reviews called it a bad phone, a way worse computer than a normal computer, and a keyboard much worse than a BlackBerry. It put a camera and a GPS in a pocket, and you cannot build Uber on a PC even now. It did not have to be better than the last thing at everything.

The same test shows when a technology looks either unstoppable or disqualified by today's flaws. Everybody thought 3D printers were the future and would print anything, but 3D printers do not follow Moore's law. More compute power did not make them better, because progress depended on complicated material s

Ben Horowitz says there are no five steps for being a CEO. A mentor worth having is someone who can hold a real conversation, not someone who hands out instructions. He asks CEOs what is on their mind and where they are stuck, then uses mistakes he has already made.

One CEO asked how to think about a CTO who was an asshole. The CTO was executing better than anyone before, and the CEO was not going to fire him, so the real question was how to talk to him without driving him away. The CTO had been talking to a young woman in finance and made her cry. Horowitz would sit him down, call him a great director of engineering who ships products on time, and say he is not effective as a CTO until he

Ben Horowitz went public on $2 million in trailing 12-month revenue, 18 months after founding, on a forecast from $2 million to $75 million. He calls the offering clearly a bad idea. After talking to literally every investor in the private markets, the other choice was bankruptcy, and bankruptcy was worse.

The magic was making that bad choice without hesitating. Many companies went bankrupt at that point because they did not go public. The workable move is to do the thing that can be done, even when it looks horrible, and not hesitate.

Founders get into the same trouble when they know the product has to be re-architected, or the head of sales has to be fired, then stall because raising mon

Ben Horowitz is pretty sure AI video and entertainment is a new medium. AI film will be as different from film as film was from plays.

In the beginning of the movie industry, a fixed camera recorded a play. Film then changed dramatically over the following decades, and the same kind of change is ahead. At first, AI will generate certain scenes and movies. A whole new set of creators will use the new medium to do things. Social media is storytelling where the main character is always you, which carves out people who are less comfortable or less narcissistic and do not want to be the main character all over social media. With AI, the characters can be someone other than the storyteller. AI fi

Ben Horowitz says ChatGPT is only marginally better than Grok if someone were starting from scratch. The gap is not a massive differentiator, and it is not a network effect yet.

Possession is nine-tenths of the law when competing on software. The interface is simple, since a person is just talking to it, so switching is not complicated. Even so, the product someone already uses is very likely the one they keep using. Getting them off it takes something dramatically better. An equally good product, or a slightly better one, does not take out a competitor. It takes a product about 10 times better.

Taking the market is probably the most obvious and most significant advantage. With AI, taking

Ben Horowitz says AI feels like the early internet because every day brings some very new awesome thing. That daily surprise is what is really similar. Everything else is kind of different. AI is the new computer, a probabilistic computer, where earlier computers were deterministic. The internet is a network, so this wave is not another network.

The history of the industry has always been networks and computers. Integrated circuits, mainframes, SNA, PCs, LANs, smartphones, and the internet were very different technologically and had very different adoption curves, but they enhanced each other. AI takes the computer role now. The move from a deterministic computer to a probabilistic computer

Andy Jassy tells Ben Horowitz and Erik Torenberg that compared with 2021, when he took over as CEO, Amazon's overall revenue is about 80 percent higher, AWS is almost three times bigger, and ads revenue is about two and a half times bigger.

On the retail side, he says Amazon doubled a fulfillment network that had taken 25 years to build, and did it in about 18 months, then built a last-mile transportation network the size of UPS.

Chris Dixon says the internet is headed for a world in which the most numerous participants are agents, not people. The shift is still early, and he calls it somewhat inevitable. Muse and Instinct are utility proxy agents that act for people. Further categories of agents come next, and those agents become the main participants on the internet.

Agents will need their own ways to communicate, transact, and share knowledge, so a run of new products has to be built for them. They have already built their own version of a public forum inside a Hugging Face app. The agent ecology online is going to be spectacular, and people will plug into it in all kinds of ways.

Chris Dixon says founders should ask whether an idea is ambitious enough, not whether it is too ambitious. What is possible is spectacular, and a moderate idea probably will not be differentiated enough to matter.

Some incumbents, including technology companies in Silicon Valley, will be able to adapt, and Andreessen Horowitz wants to arm them with the new tools. He would tell any of them listening that the clock is ticking. He is full-time on the work and describes himself as drowning in how fast development is moving. A super-empowered 24-year-old with AI, in his view, will achieve amazing things in the next decade.

Marc Andreessen says AI-native startups already see what they can do with AI differently from a typical incumbent in almost any industry, and he calls the gap as wide as any he has seen in his career. The gap is still opening, because the capability races ahead week by week. Those founders will not stay inside one category. They will not think of themselves as only a software company, only a hardware company, or only an AI company. He already sees them choosing how many areas to touch and how to combine those areas.

He sets the change against Netscape in 1994. If he and Ben Horowitz had today's tools then, he says they would have built the whole internet, gotten crypto done then, had stream

Chris Dixon says a16z Frontier chooses to be too early because every good idea the firm has seen that involved technology has eventually worked, or he believes it will work if it has not worked yet. The qualifier is a good idea. Bad ideas exist. Almost everything that worked had a precursor that failed and was maybe 10 years too early. Timing is the hardest thing to predict, because the technology, the culture, the business environment, and the political environment all have to line up.

The fund accepts category risk. Some categories may not work. Oculus was bought, and 13 years later he still believes strongly in virtual reality, but it is not as mainstream as he wanted, because more hard

Chris Dixon says Douglas Adams, author of The Hitchhiker's Guide to the Galaxy, described how people receive any new technology in three different ways, split by age.

Below 15, the technology is just how the world works. He tells of introducing his own child, at seven, to a computer that answers questions. The child treated it as obvious: what good is a computer that does not answer questions? From 15 to 35, the technology is new, exciting, and a place to build a career, so people want to learn it. Above 35, it looks unholy and evil, sure to overturn existing structures and destroy society. Refusing to answer the telephone fits that pattern, and an old man yelling at a cloud keeps showing u

Chris Dixon says humans just figured out how to turn sand into intelligence, a framing he credits to Marc Andreessen. The way to learn it is math and computer science, and maybe economics, though he is not sure economics belongs on the list.

He feels fortunate that a lot of his career, with Ben Horowitz and Marc Andreessen, happened during the growth of the internet, something bequeathed by people before him. The AI work now is going to be the dominant narrative of the next 20, 30 years. Being in the mix, at the cutting edge, and using the tools is the obvious thing to do, and the work does not have to be solely technical.

A lot of other fields are going to be revolutionized. Classics is b

Chris Dixon says the internet business model is breaking, and he puts the start of the break maybe in the last decade while calling the break especially sharp now. He names the old arrangement the internet's covenant, a deal between the distribution side, Google and Facebook, and the content side, news sites, travel sites, and other publishers. A site would give Google a snippet, let Google index the page, and Google would send clicks back. A Google search now returns an answer, and social networks try to keep people on the platform instead of linking out.

Websites that make money on ads and conversions can no longer do that. Stack Overflow and some other websites saw traffic drop precipito

Marc Andreessen says open models are lapping closed models by six months to 12 months, depending on how you count, and open source is racing ahead.

The U.S. and China, the two great tech superpowers, both have active open source worlds alongside companies building proprietary systems. For the last three or four years, governments have threatened draconian regulation or a ban on open source. The EU still aspires to do it and does not quite have the leverage to threaten fines until the work stops. The current U.S. administration is 100% pro-open source and 100% wants American open source to happen and to win AI globally. The Chinese Communist Party is, at the moment, enthusiastic about open s

Marc Andreessen calls the pattern product manager brain: product managers design products for product managers. People build products for themselves, and product managers still tell a story about building for other people. The use cases stay in the same circle, such as booking a restaurant in San Francisco for dinner with friends who are also product managers, or booking a flight to a product management conference to spend time with more product manager friends.

The same narrowness shows up in an assistant pitched as happy, harmless, and honest, which becomes the most boring possible product and refuses a question it is not allowed to answer. Early companies held a blinkered view of human a