Patrick Collison: Is AI Breaking the Lean Startup Playbook?

Predictions and Expectations:

• Patrick Collison expects that for a long time to come, human "neuronal lookups" (cognitive knowledge) will be significantly faster than AI agent lookups because the brain can handle many more "round trips" than external tools.
• He believes renouncing human cognitive effort and first-principles reasoning is premature given the current "enormous premium on cognitive ability" seen in companies like Stripe and Anthropic.
• Collison doubts the prevailing "millenarian model" that this specific moment is the final window to escape a permanent underclass, suggesting it is "hard to predict anything, especially in the future."
• He predicts that if organizations do not adapt to AI quickly, they face a high risk of being left behind with "archaic and antiquated ways of operating," making the "status quo" extremely dangerous.
• Patrick Collison expects the future to trend toward a more decentralized world with "many thousands of winners" rather than a hegemonic centralization by a few AI labs.
• He believes the fear of big model providers (the "labs") completely obliterating independent startups is overstated due to the organizational complexity required to prosecute hundreds of priorities simultaneously.
• Collison predicts the trend of rapid new business formation will continue, citing that the number of new businesses starting on Stripe is currently up around "2x year over year."

Timelines and Milestones:

• Collison predicts opportunities will not disappear in "three or four years," citing a historical surfeit of opportunities in Silicon Valley over "many decades."
• He notes that for Stripe, the time from first lines of code (fall 2009) to public launch was "almost two years."
• He mentions that the relative growth rate of new businesses starting on Stripe in the current period is the largest observed in any given year since "2019 to 2020."
• He observes that the time to revenue for new companies incorporated via Atlas is declining, implying a shorter timeframe to market validation compared to the past.
• Collison references a timeframe of "five years" as a speculative horizon for how the world might change.

Technology and Product Direction:

• He suggests that in the era of AI, companies may need to "more aggressively de-correlate" their starting points to avoid competing in "niche" markets that have become "more aggressively tilled."
• Collison expects agentic capabilities to "obviate a bunch of specific verticals or tasks" depending on forecasts of model capabilities.
• He predicts that companies will increasingly start "much more aggressive and ambitious things up front" rather than following the traditional "lean startup" method of starting narrow and iterating slowly.
• He anticipates that enterprise customers will be "spring-loaded" to adopt new products and services, allowing startups to achieve meaningful scale "right out of the gate."

Market and Industry Outlook:

• Collison believes the internet is now a much bigger place than "20 years ago," making it harder to find unoccupied niches and requiring more divergent starting points.
• He projects that "median business" performance is better this year than a year ago, and the probability of businesses reaching revenue thresholds of "$1 million, $5 million, $10 million" is increasing.
• He expects "new contracts" with enterprises to become a primary driver of growth for YC companies, a dynamic he describes as "the new thing."
• He predicts that consumers, despite complex views on data centers, will remain "beguiled" by AI products and show a strong "predisposition and openness to experimenting with the new."
• Collison forecasts a future where the "hunger and intensity" of companies retooling for AI will lead to "broad-based prosperity."

Company Plans:

• Stripe plans to give "free Atlas incorporation" to all attendees who email "startup school at stripe.com" after the event.
• The speaker notes that Stripe has been increasing the number of private beta customers "every single month" leading up to public launches.
• Collison expects to continue working with "the world's most interesting and innovative companies" through partnerships and feedback loops for the "next 10 years, 17 years, for 30 years."

Financial Guidance:

• Patrick Collison states that the number of new businesses starting on Stripe is "up around it's a bit under but around 2x year over year."
• He mentions that from "2019 to 2020," the growth rate in new businesses inflected to "maybe 50% or thereabouts" year over year.
• He observes that the "time to revenue" for new companies is declining, though no specific monetary percentage is provided for this metric.

Risks and Caveats:

• Collison admits he is "not sure" if AI capabilities will specifically obliterate verticals or if it will be the model capabilities themselves, noting this is "contingent on one's forecast."
• He warns that if you are not in the "speed running" mode of starting a company, you might worry about missing the window, though he believes this fear is largely a "poor intuition."
• He cautions that while the current data suggests it is "the best time to start a business," he does not know what the world will look like in "five years" and things "can change."
• Collison notes that the "cost of failure" for high-stakes projects is high, which is a reason to take longer to build in certain domains, contrasting with the "launch early" advice.

Confidence and Disagreement:

• Patrick Collison expresses a strong belief that "it is never been a better time" to start a business based on current Stripe data, though he frames this as a trend observation rather than a guarantee.
• He is skeptical of the "perpetual underclass" narrative, predicting it will be an "under" to think the current couple of years will trigger a permanent transformation of society.
• He disagrees with the necessity of dropping out of college, stating the cost is "de minimis" and that "nobody has ever cared" if you return later.
• He expresses uncertainty about whether AI will make it easier or harder for future generations to build software, saying "I don't know" but suggesting AI makes spinning up organizations easier.