What It Takes to Win as a Founder | Ben Horowitz

Ben Horowitz says founders are driven by a need to build something they believe should exist, and that AI may make starting a company dramatically easier without making the founder’s job easier. The work still means betting on yourself amid uncertainty. Self-doubt is normal, not evidence that you should stop: Horowitz recalls Mark Zuckerberg wondering at 22 whether he would succeed. He argues that founders should be judged by the magnitude of a distinctive strength, not by an absence of weaknesses; skills that once mattered can lose relevance, making it necessary to become world-class at something new.

AI differs from the early internet: Horowitz calls it a probabilistic computer, whereas the internet is a network. AI products can reach large audiences quickly because internet access is already widespread, but building an application still requires more than wrapping a model. Applications can combine sophisticated models—he said Cursor used 14—with specialized user data that helps them serve a particular audience and potentially improve as they gain users. Familiarity also protects incumbents: a rival may need to be 10 times better to dislodge a product customers already know. Horowitz expects products initially to compete on whether they work, with distinctive experiences and network effects potentially emerging later; personalized software generated at runtime remains speculative.

The pace and direction of AI progress are uncertain. Horowitz said language models had plateaued “quite a bit,” while reinforcement learning was scaling; code was improving substantially, but chatbots were changing less. A breakthrough or plateau could affect applications and infrastructure alike. He cautions investors against projecting a technology’s current trajectory forward: 3D printers did not improve simply with more computing power because progress depended on materials science. A technology also need not beat its predecessor at everything. The iPhone was criticized in 2007 for its phone, computer, and keyboard, but its camera and GPS enabled uses such as Uber that a PC could not support. AI hallucinations, he adds, may be manageable and sometimes useful rather than disqualifying.

As software has become easier to build, companies have shifted away from complex hardware-era operations toward smaller teams, and Horowitz sees AI accelerating that trend. Twitter, he said, was built by five people with laptops and AWS; Instagram had 13 employees when it sold for $1 billion. Yet whether software companies can remain small as they grow is unresolved, and a small team may struggle to hold its position against larger competitors. Easier company creation is also likely to change venture capital. Horowitz says capital can help a company build momentum against a similarly strong rival by attracting engineers, improving the product, winning customers, and reaching its next financing round sooner.

Horowitz says a founder’s choice of seed investors should reflect practical help with hiring and reaching the next round, as well as whether the founder can work with them while learning to be a CEO. He argues that venture firms should offer more than a smart board member and that person’s network. As founders gain expertise, board-level product advice may become less useful; practical assistance such as customer introductions and evaluating executive candidates can matter more. But investors can also give confident, generic advice without having served as CEOs, so founders should value situation-specific guidance over instructions presented as universal rules.

The CEO’s hardest decisions are often psychologically difficult rather than hard to identify. Delaying a product rearchitecture or replacing a sales leader can slow a company and foster politics; in a crisis, Horowitz says, the CEO should identify the best available action and take it. He illustrates the stakes with his own company, which went public 18 months after founding with $2 million in trailing-12-month revenue and a forecast of $75 million. Facing a choice between going public and bankruptcy, he chose to go public without hesitation. There is no standard five-step CEO training plan, he says, because the right response depends on the company and people involved. A useful mentor offers informed, specific advice—for example, giving a strong CTO a chance to improve when poor cross-functional behavior is undermining the role.