How AI Agents Are Creating 12X Productivity Gains
Joe Lonsdale interviews Scott Wu and Russell Kaplan, the founders of Cognition, about how AI agents are revolutionizing software engineering and delivering 6-12x productivity gains. The episode explores the emerging era of "software abundance," where engineers no longer write code manually but instead direct AI agents using plain English, and discusses the profound implications for businesses, government, and society.
Key takeaways
- • AI agents like Devon are enabling 6-12x productivity gains in software engineering by automating the implementation work while engineers focus on high-level problem-solving and architecture decisions.
- • Engineers at Cognition no longer type code; instead they describe what they want built in English, fundamentally shifting the medium of programming away from code itself.
- • The best engineers using AI agents adopt an abundance mindset, running multiple solution approaches in parallel and testing them simultaneously rather than sequentially deliberating on a single approach.
- • Agency and reasoning are now the critical skills for leveraging AI agents effectively—the ability to think clearly about problems and take initiative matters far more than low-level coding ability.
- • AI agents are moving from reactive to proactive engineering, with companies automating security vulnerability remediation (70% automatic remediation rates) and triggering engineering work based on system events rather than human requests.
- • Software is entering a hyperdelationary cycle where building becomes so cheap and fast that demand explodes—creating an explosion of small businesses empowered to build sophisticated products solo.
- • Government agencies are adopting AI agents to modernize legacy systems (especially COBOL), dramatically accelerating permit processing, security fixes, and IT modernization despite regulatory constraints.
- • The MER metric shows AI capability is doubling every 2-3 months in terms of uninterrupted work tasks, with current models handling 10-20 hours of human-equivalent work before requiring feedback.
Recommendations (3)
"When I was at autopilot, I was a machine learning scientist working on the visional network and Elon had this phrase that he really drilled into us, which is everyone is chief engineer."
Russell Kaplan · ▶ 6:55
Mentioned (8)
More from these creators
How 200mph Flying Taxis Will End Traffic Jams
Jake Paul on Endorsing Trump & Investing in Attention
NASA's Warning: We Are 1 Year From Losing the Moon
6 Levels of AI Investing Everyone Should Know
Exposing a Billion Dollar Fraud Scandal (Using Palantir)
Former Crown Prince of Iran: Why This Regime Must Be Taken Down