Paul Graham on Startup Ambition, AI and What Makes Founders Formidable

At YC’s original Mountain View office, Vivian Shen talks with Paul Graham about startups, ambition and AI. Graham says he rewrites his talk for each batch, even though many startup principles recur.

Graham argues that some current YC startups pursue more ambitious problems than earlier companies. He points to intercontinental cargo and cancer research as examples of ideas he considers unusually serious.

Discussing cancer, he suggests that progress may come through many different approaches rather than one universal cure. He describes a startup offering research support to individuals; these are examples of ambition, not evidence of a validated treatment.

Graham revisits his essay on frighteningly ambitious ideas. He argues that a major incumbent may be challenged when technology changes the underlying way people meet a need, citing his own shift from web search toward asking AI for information.

He says founding a startup is grueling, so ability alone is not enough: founders need a strong drive to get through obstacles. He calls that drive ambition.

On a daily basis, he says, founders may be motivated less by imagining wealth than by preventing immediate failure. A crashing server demands attention because the product they care about is in trouble.

Graham thinks some founders already have ambition but have learned to suppress it in environments that reward obedience. This is his interpretation of personality, not a settled rule about how ambition develops.

He warns against starting a company merely as a credential. A successful startup can become years of one’s life work, while the process is difficult even when the team is accepted into a prestigious program.

For Graham, a formidable founder is someone who finds a way to achieve what they set out to do. He says investors look for that quality because their outcomes are tied to the founder’s.

Asked whether lean startups are obsolete, Graham says he has not read the book and asks for a concrete definition. When the question becomes whether a focused company can start with little money, he argues that many still can.

He expects inference costs to fall, while acknowledging that tokens were expensive at the time. His projected rate is a forecast, not a guaranteed price trajectory.

Even capital-intensive projects can begin by matching the next milestone to available resources, he suggests. A rocket team might start with a design or simulation, seek expert feedback and use that evidence to approach investors.

Graham notes that credibility differs: an established expert can persuade backers with evidence that may not suffice for an unknown founder. Funding is not assured merely because a milestone has been reached.

Looking back on AI research in the 1980s, he describes expecting systems to master small, animal-like capabilities before reaching human-level breadth. Instead, language models surprised him by producing broad, plausible but often unreliable answers early on.

He and Shen call this uneven capability a jagged frontier: a system can appear impressive on one task and fail a seemingly simple question in another. Graham uses an everyday example of unreliable restaurant hours.

Asked to define AGI, Graham references the Turing test and says the imagined finish line now looks more like a broad, uneven region. This is his metaphor for uncertain boundaries, not a definitive test or claim of achievement.

Shen asks whether AI changes his belief that shipping quickly predicts startup success. Graham says speed still matters, but teams must also come up with ideas; more powerful tools do not remove differences in execution.

One visible change, he says, is that some companies now have large AI and compute bills alongside salaries. His cited costs describe particular examples, not the budget every startup needs.

Graham describes the benefits of a YC batch: founders gain peers who can share advice on similar problems, and other startups may become early customers willing to try a new product.

The original idea for YC, he recalls, was an early-stage investment firm making small investments with standardized paperwork. Funding a group at once began as a way to learn investing and offer students an alternative to summer jobs.

The batch format proved useful, and the founders and investors learned quickly. Graham says YC has grown in scale while organizing companies into smaller groups that maintain peer connections.

Asked about the next giant company, Graham returns to founders rather than naming a specific sector. He believes determined founders and promising ideas will continue to matter, while acknowledging that ideas can change.