YC Discusses Abandoning Stable Jobs for Entrepreneurship at Twenty, Degrees No Longer Automatically Convert to Wealth, Admissions Now Focus More on GitHub
Garry Tan, President and CEO of Y Combinator, along with managing partners Harj Taggar, Diana Hu, and Jared Friedman, discussed at the AI Startup School's Lightcone event about "how to navigate your twenties in the age of artificial intelligence." The core judgment is that the traditional path of going to school, joining large companies, and climbing the corporate ladder is failing. The unemployment rate for new computer science graduates has reached twice that of art history graduates, and degrees no longer automatically convert to wealth. Meanwhile, small, focused teams can achieve eight-figure revenues within months.
The four are betting on proactivity rather than credentials. AI excels at following instructions, while humans must compete on deciding what to do and accumulating judgment in their fields. Harj mentioned a "frontline engineer" path, diving into real work environments with tools. Diana emphasized that starting points must be narrow: Airbnb initially offered air mattresses in living rooms, and Paul Graham advised Brian Chesky to find ten true fans of a product rather than a hundred casual users; AI allows previously niche endeavors to charge for "selling work" instead of "selling software," making market boundaries less clear, necessitating a deep focus on one thing before expanding.
Speed has been redefined. Previously, graduating from university and securing Series A funding within a year was a milestone; now, some teams achieve billion-dollar valuations within a few years post-graduation, with new benchmarks being around $10 million to $12 million in revenue within twelve months. In the Winter 2025 batch, about a quarter of companies have 95% of their codebase generated by AI; Emergent from Summer 2024 reached nine-figure revenue in about eight months after public release, with annualized revenue of $15 million and around 15 employees; Winter 2024's Retell AI has an annualized revenue of about $60 million with around 40 employees. Garry later stated that his output in 2026 would be about 400 times that of when he wrote Bookface in 2013, and admissions now focus more on GitHub than on Harvard, Stanford, or Google resumes.
They also outlined the negatives. Don't wait for permission and gradual guidelines; your twenties are a window for experimentation and mistakes. Don't just aim to grow a company; first, solve real human problems. Avoid casting a wide net; focus on excelling at one thing early on. Don't maximize entrepreneurship courses and credentials as a measure of growth. The decision to drop out or when to leave a job is highlighted: the window for co-founders may not come again, but the startup project itself could be labeled a trap. The judgment standard can be summed up as: the world looks at what you have done, not what you plan to do.
In market mechanisms, what is being bought are young people who can leverage models to produce products and enter vertical fields; what is being sold are entry-level white-collar jobs that can be instructed. Event-driven models have driven prototype costs close to zero, while quality thresholds and distribution have actually increased. Funding and opportunities flow to those with work, domain judgment, and the ability to deliver weekly updates. Accelerators and small teams that can quickly validate their ideas benefit; those who treat a computer science degree as a safety net and leave side projects in demo warehouses face pressure. The advice itself is a talent supply narrative: YC wants the next batch of founders, not to supply large companies with cogs.
Source: Public Information
ABAB AI Insight
This discussion is not a policy released in 2026, but a recording from the AI Startup School in July 2025, later summarized into seven principles and circulated in the Chinese community. The backgrounds of the four speakers shape their perspectives. Garry Tan transitioned from a designer at Palantir to Posterous, which was acquired by Twitter, then founded Initialized Capital, and returned to YC as president in 2023. Harj Taggar worked on Auctomatic and Triplebyte, advising over a thousand companies and five public ones. Jared Friedman is a co-founder of Scribd. Diana Hu worked at Escher Reality, which was sold to Niantic, and managed an augmented reality platform. They describe the "disappearance of stable jobs" as a structural fact while framing the exit as their funnel: joining or founding startups.
The capital path is a classic closed loop of accelerators. First, declare the large company path ineffective, then showcase the million-dollar per capita output in batches, and finally write proactivity, niche, and tool leverage into admission standards. The 95% code generated in Winter 2025 and Emergent's nine-figure revenue in eight months serve as poster numbers for applicants. Garry mentioned at SXSW that he cares less about schools and previous employers, focusing instead on repositories, effectively shifting the selection from diploma markets to verifiable outputs. YC does not pay twenty-somethings a salary; it takes equity from the next batch instead; the more the advice resembles life philosophy, the cheaper the recruited founders.
Comparisons need to be specific. In the 2010s, YC's narrative was about "creating what people want" and weekly growth; in 2013, a computer science degree was the safest ticket. In 2025 to 2026, the same institution uses the unemployment rate of computer science majors as a cautionary tale, aligning with the Wall Street Journal and New York Fed's descriptions of young graduates' dilemmas, but with opposite prescriptions: while media data stops at unemployment, YC's prescription is to start a business immediately. Paul Graham's advice to Airbnb about finding "ten true love users" has been directly applied to the AI niche. The industry is in a phase of "prototype devaluation and distribution appreciation": anyone can create a chatbot, and pure software categories like Calendly have been pointed out by Diana as dead.
Structural judgments belong to the reconstruction of the industrial chain after technological substitution. The mechanism is that models consume jobs that can be executed by instructions, leaving the remaining premium for topics, domains, and customer sites. Accelerators use "don't wait for permission" to shift risks from institutional balance sheets to individuals in their twenties while using batch examples to prove that risks can be realized. Deepening one line is necessary because casting a wide net in a market where everyone can prototype equates to having no signal. The world values delivered products, not career planning slides—this applies to both founders and employees, but YC only recruits the former.
ABAB News · Cognitive Laws
When stable paths fail, institutions will rename risks as opportunities.
After prototypes are free, what is expensive is judgment and customers.
The world settles on what has been made, not on business plans.