Marc Andreessen and Amjad Masad on AI coding and AGI
Marc Andreessen speaks with Replit founder Amjad Masad about AI agents that build and test software from plain-language prompts. Masad explains why coding has become a leading domain for reinforcement learning and verifiable results, how Replit extends agents through testing loops, and why useful AI may advance faster than true AGI. The conversation closes with Masad’s path from Jordan to building Replit.
Passages
22:34 “So we’re kind of getting into here kind of the holy grail of AI, which is sort of generalized reasoning, you know, by the machine. So you mentioned this a couple times with this idea of verification.” Share
39:13 “And usually AGI is defined as able to do everything better than a person can. And it’s like, well, okay, so if doing everything better than a person can, it’s like, if a person can’t do any transfer learning at all, doing even a little bit of marginal bit might actually be better.” Share
Search what Andreessen said from Claude, ChatGPT or your own code. See plans