Y CombinatorPaul Buchheit: Why Evals, Not Code, Are the Real AI Moat
Predicting the next token dissolved the paperclip maximizer fear; now eval sets, not codebases, are the moat, as Jerry shows with 50% growth post-GPT-4.
Episode Details
EPISODE INFO
- Released
- January 24, 2025
- Duration
- 39m
- Channel
- Y Combinator
- Watch on YouTube
- ▶ Open ↗
EPISODE DESCRIPTION
In this special episode of Lightcone, we’re joined by YC partner and creator of Gmail Paul Buchheit to dig into some of the latest trends in the world of AI startups. We recorded our conversation at a recent retreat where 300 of the top AI founders in the world gathered to share expertise and make predictions about how this technology will shape our future. In the discussion we cover a wide range of topics including: The future of work, the power of agency and taste in an AI world and why this is the absolute best time to be building a startup. Learn more about the YC startups mentioned in this episode: Tavus: https://www.tavus.io Jerry: https://getjerry.com Apply to Y Combinator: https://ycombinator.com/apply Chapters (Powered by https://bit.ly/chapterme-yc) - 00:00 - Intro 01:17 - Retreat. 05:04 - Demand for AI. 08:00 - Evals. 10:00 - Product iteration. 14:37 - Balance. 16:16 - Automation. 19:31 - Predictive models. 22:21 - OpenAI. 24:58 - AI tools. 26:23 - Cursor. 29:30 - Scaling. 33:16 - ROI. 36:37 - Startup success. 38:30 - Outro
SPEAKERS
Garry Tan
hostPaul Buchheit
guestHarj Taggar
hostJared Friedman
hostDiana Hu
host
EPISODE SUMMARY
In this episode of Y Combinator, featuring Garry Tan and Paul Buchheit, Paul Buchheit: Why Evals, Not Code, Are the Real AI Moat explores aI Agents Explode Startup Growth, Redefining Work, Wealth, And Software The episode covers how AI, especially AI agents for businesses, is radically accelerating startup growth, with many YC companies hitting revenue milestones in months that once took years. The hosts describe unprecedented enterprise demand for AI, where buyers already want solutions and technical founders win by simply delivering products that actually work. They argue that the true moats are now eval datasets, prompting expertise, rapid iteration, and willingness to constantly rebuild on the latest models. The conversation broadens into how AI can increase human agency, reshape labor and wealth (machine money vs human money), and why we may be on a “good timeline” for AI development rather than a dystopian one.
RELATED EPISODES
