Key Moments

Patrick Collison: "What If You Succeed?"

Y CombinatorY Combinator
Science & Technology4 min read31 min video
Jul 31, 2026|7,937 views|296|15
Save to Pod
TL;DR

Founders should ask "what if we succeed?" because the cost of failure is often overestimated, and success can bring unforeseen challenges and rewards.

Key Insights

1

Cognitive L1 cache, or knowledge held internally, is significantly faster than retrieving information via AI agents, suggesting continued value in deep learning.

2

Patrick Collison dropped out of MIT twice to start companies, highlighting that returning to education is possible and the perceived risks of dropping out are often minimal.

3

Stripe took nearly two years for its public launch due to the complexities of security, payments infrastructure, and partnerships, but maintained production users from early on.

4

The "lean startup" doctrine may be evolving, with a potential shift towards more ambitious, "anti-lean" company launches in the AI era, as seen with companies like OpenAI and Anthropic.

5

Stripe data shows a significant increase in new business formation (nearly 2x year-over-year as of July 2026), with median businesses performing better and reaching revenue thresholds faster.

6

Despite fears of AI leading to economic centralization, Stripe data suggests a trend towards a more decentralized world with more broad-based prosperity and thousands of winners.

The enduring value of internal knowledge versus AI augmentation

Patrick Collison emphasizes the continued importance of internalizing knowledge, likening it to 'cognitive L1 cache.' He argues that information retrieved from one's own mind is vastly faster than relying on AI agents. While acknowledging the power of AI for complex computations, Collison suggests that the speed and efficiency of 'neuronal lookups' remain a significant advantage. This perspective is supported by observed trends in companies like Stripe, where there's still a strong premium on cognitive ability. He advises against prematurely abandoning the pursuit of deep understanding before its benefits are fully saturated, even with advanced AI tools.

The non-linear path of education and entrepreneurship

Collison shares his personal experience of dropping out of MIT twice to pursue entrepreneurship. This offers a crucial perspective for students considering similar paths: dropping out is not a permanent, irreversible step, and returning to education is a viable option. He suggests that while enjoying college and completing a degree is perfectly fine, if it's not a passion, the perceived risks of leaving early are often overstated. Collison recounts feeling an urgency to act due to a perceived ephemeral nature of startup opportunities, a notion he now views as a 'poor intuition' given Silicon Valley's consistent abundance of opportunities.

Stripe's deliberate launch strategy in a complex domain

Unlike the typical Y Combinator 'launch early, launch quickly' mantra, Stripe took nearly two years from initial work to public launch. This extended period was driven by the inherent complexities of building a financial services business, requiring significant groundwork in security, payment infrastructure, reliability, and crucial banking partnerships. Despite the long lead-up, Stripe maintained active production users from very early on. This 'just-in-time' development, fueled by continuous feedback from early clients like Ross Buché, allowed them to iterate on core functionalities such as charging, refunding, and payouts, grounding their development in real-world needs rather than theoretical assumptions.

Rethinking 'lean startup' in the age of AI

Collison posits that the traditional 'lean startup' doctrine, which emphasizes identifying small niches and iterating, may need re-evaluation in the era of advanced AI. The internet's growth and AI's capabilities could make it harder to find and sustain such niches due to increased competition. He suggests that companies might benefit from more aggressive and ambitious 'anti-lean' approaches from the outset, citing examples like OpenAI and Anthropic. The ease with which AI can enable diverse capabilities might allow founders to start with broader, more divergent visions.

The 'what if you succeed?' question for founders

Beyond the fear of failure, Collison stresses the critical importance of asking 'what if you succeed?' He highlights that the reality of a successful, scaled business – with employees, investors, and customers – might not align with a founder's long-term desires or interests. Using Larry Ellison's Oracle tenure as an example of long-term commitment, Collison emphasizes that founders should consider if they will genuinely enjoy working on their creation for decades. For him, Stripe's appeal lies in working with innovative companies and finding the 'totality of Stripe' intellectually stimulating, proving that a business can be the opposite of a 'schle blindness' scenario.

AI's impact on business creation and economic decentralization

Stripe's data indicates a significant surge in new business creation, with a nearly twofold year-over-year increase in new businesses starting on the platform as of July 2026. Crucially, this growth is not just in quantity; the median business is performing better, and the probability of reaching revenue milestones is increasing. Collison attributes this to businesses being 'spring-loaded' to adapt and fearful of being left behind by archaic methods. This dynamic, coupled with enterprises' increased willingness to adopt new solutions from startups, suggests a more favorable environment for new ventures and a potential trend towards a decentralized economy with broad-based prosperity, challenging fears of AI-driven economic centralization.

New Business Growth on Stripe

Data extracted from this episode

Time PeriodChange in New Businesses StartedMedian Business PerformanceRevenue Threshold Probability
Last year vs. This yearApprox. 2x year-over-year (largest relative jump)Doing better than a year agoIncreasing for $1M, $5M, $10M thresholds
2019 to 2020 (during COVID)Approx. 50% year-over-yearN/AN/A

Common Questions

While AI can compute and look up information, knowing it directly in 'cognitive L1 cache' is significantly faster for complex problem-solving and system building. There's still a premium on cognitive ability, and abandoning it prematurely due to AI advancements would be unwise.

Topics

Mentioned in this video

More from Y Combinator

View all 616 summaries

Ask anything from this episode.

Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.

Get Started Free