Artificial General Intelligence (AGI)

Today’s AI is narrow: a model can write code, summarize text, or play Go at superhuman level, but each capability is specialized and none transfers freely to arbitrary new problems. AGI names the hypothesized point where one system matches a human across essentially any intellectual task — learning new domains on its own rather than being trained for each.

The term is more contested than it sounds. There’s no agreed definition and no clean test, so “are we close to AGI?” often turns into an argument about definitions. On the show the useful framing is less about the finish line and more about trajectory — what each capability jump actually unlocks for builders now, like the Poolside founders’ “first code, then AGI” bet that mastering software is the path. For day-to-day work, the practical questions are about frontier models and reasoning, not a labeled arrival.