AI is democratizing startups. Don't let it dumb you down.

Patrick Collison argues that despite AI's growing capabilities, fundamental cognitive skills remain essential. He uses a cache analogy: neuronal lookups (internal knowledge) are far faster than querying an AI. Engineers should still learn deeply; compilers didn't eliminate the need to understand memory layouts.

Collison reveals Stripe data showing new business creation is up 2x year-over-year, with median revenue also rising. This counters fears that AI will centralize wealth. Instead, it suggests a decentralized boom where more startups succeed faster due to enterprise fear of being left behind.

The conversation challenges the lean startup dogma. Collison suggests that in the AI era, founders should be more ambitious from day one. He points out that many top companies (e.g., Stripe, Anduril) were anti-lean, taking years to launch. AI reduces the cost of exploring divergent ideas.

A key tradeoff: while AI agents reduce the need to write boilerplate code, Collison warns against outsourcing critical thinking. He personally avoids AI-written prose, finding it uncompelling. The real risk isn't failure—it's succeeding at something you don't enjoy for decades.