Garry Tan on YC Founders, Personal AI Agents and Open Markets

Walter Isaacson asks how Y Combinator selects founders who will succeed. Garry Tan says YC receives roughly 80,000 to 90,000 applications a year for about 800 places, and spends only ten minutes with founders at the final interview. The short interview is a deliberate constraint.

Tan wants founders to explain a real user’s problem, how their technology solves it and why the result matters. If time remains, the interviewers ask why this particular founder should build it. In his view, the decision should be a clear yes or no.

Asked whether YC would recognize a young Elon Musk, Tan says that is a question the organization uses to judge itself. If it could not say yes to people with the potential of founders such as Bill Gates, Steve Jobs or Brian Chesky, he says, it would need to rethink its process.

Tan thinks his own company, Posterous, would have passed the test. It offered simple blogging by email and had a strong demonstration. When he and his cofounder showed it to YC’s founders, he says, they understood it immediately.

His experience building a company also informs his investing. Tan describes venture capital as a business of offering money in a market where more capital wants to enter technology than there are strong opportunities. He recalls that, when YC began, access to investors often depended heavily on personal connections.

Isaacson questions whether entrepreneurship still requires a financial safety net. Tan agrees that YC cannot claim to be a perfect meritocracy, though he says it should aspire to one. He points to YC’s $500,000 investment as a way to let people participate without relying entirely on wealthy parents.

Tan recalls being an early Palantir employee and learning from Alex Karp. On one trip to demonstrate software to senior customers, Karp chose an expensive hotel for the young engineers. His point, Tan says, was that even casual details of presentation could shape customers’ perception of the company and its product.

The conversation turns to antitrust. Tan argues that the US government’s case against Microsoft helped create room for Google to grow. He now sees self-preferencing by large platforms as a threat to new entrants and argues that open markets give startups a chance to compete.

Tan praises aspects of Europe’s Digital Markets Act, including provisions he associates with access, portability and interoperability. He compares a dominant platform to a monopoly railroad: control over a route to market can influence many businesses that depend on it.

Isaacson challenges the analogy, noting that newer AI companies have grown despite the power of established platforms. Tan agrees they have emerged, but calls leading AI companies a new form of big tech and worries that access to frontier models could become restricted.

Tan describes YC as an advocate for startups, with some union-like characteristics even though it is not a union. He says the organization wants a voice on issues from antitrust to taxes, and alleges that some groups claiming to represent small companies have instead promoted large-company interests.

He says YC has a small Washington policy operation but can amplify founders’ voices through its network. Tan acknowledges that YC itself is powerful; his stated goal is to help startups obtain access to markets and participate in policy debates.

Isaacson asks about immigration and Tan’s family. Tan says immigrants contribute significantly to successful startups and describes his mother’s family arriving as refugees from Burma and his father coming from Singapore. He sees the ability to attract people from around the world as an American strength.

Tan also discusses growing up with an unpredictable father who sometimes struggled with alcohol. He says the experience made him attentive to other people’s emotional states, a skill he now uses when meeting founders. Isaacson connects that upbringing to his leadership.

The discussion shifts to AI agents. Tan argues that AI should be treated as more than a chat feature and describes YC’s internal agent, Straylight, as having access to organizational context about founders, applications and communications.

He emphasizes the amount of information an agent can work with, but Isaacson presses him for something beyond retrieval. Tan points to more autonomous assistants that can use tools and carry out work, while acknowledging a tension between repeatability, safety and the experience he wants.

Tan says his personal agent has access to extensive email, notes and files. To prepare for this interview, he says, it gathered and transcribed several of Isaacson’s podcasts and extracted themes. He sees the value in an assistant that can perform such preparation on a person’s behalf.

He predicts personal agents will become common within two or three years, acting as a persistent assistant that knows a user’s goals and offers suggestions. He also notes that his own setup is expensive: he estimates spending $100,000 to $200,000 a year on tokens, a privilege few users currently have.

Asked whether AI will create or destroy jobs, Tan says he hopes it creates more but does not claim certainty. He acknowledges that some jobs may disappear in the short term and that retraining will be necessary. He favors addressing misuse rather than stopping the technology altogether.

Tan describes GStack as an open-source effort to help founders use coding agents more effectively. He hopes tools and training can let someone with domain expertise, such as medical billing, work with a technical partner or learn to build a company around a problem they know well.

The opportunity, he says, begins with asking what a person knows from their own work and life that others may not. He also argues that leaders with resources should look beyond cutting costs: if the gains from AI accrue only to capital, he warns, the result could be deeply unequal.

Isaacson asks whether YC considers social consequences when funding a company. Tan says it does, and recalls a company involving ADHD medication for children that YC declined to fund. He says the decision reflected uncertainty about whether the proposed expansion met a real need, not merely whether it could make money.

Isaacson asks whether social media could strengthen rather than divide society. Tan imagines a personal agent that filters a social feed according to the user’s own goals, rather than leaving every decision to a platform’s algorithm. He presents this as a possible future, not a system already available to everyone.

Isaacson counters that personalization has existed before while platform business models have rewarded divisive engagement. Tan argues that user-controlled personal AI could change who controls the filter, comparing the possibility to the shift from centralized computing to personal computers.

Tan expects competition among AI agents to resemble a new browser or operating-system battle. Isaacson asks whether consolidation into a few providers would be harmful. Tan says concentration is already a concern and wants an accessible, inexpensive open-source alternative—the equivalent of Linux for personal AI.

YC’s response, he says, combines policy advocacy with funding new talent and supporting open markets. Asked about regulation, Tan argues for rules that allow applications and services to interoperate with major platforms rather than requiring users to depend solely on the platform owner.

Tan also argues that US and European open-source AI developers should have a stronger opportunity to compete. He points to model distillation and proposes examining rules around it. His broader concern is that concentrated control and self-preferencing can prevent the next generation of companies from emerging.

Isaacson compares the debate to earlier conflicts over railroads and industrial monopolies. Tan says history offers lessons both about opening markets and preserving the ability to create businesses and wealth. He opposes both concentrated technology power and proposals he regards as excessive asset seizure.

Tan closes by arguing for more successful companies, consumer choice, jobs and prosperity. He cites OpenAI’s connection to YC as an example of an organization that grew from a small group into a force changing technology. His preferred path, he says, lies between big-tech oligarchy and broad confiscation.