Has AGI already arrived? And what comes after?

What was discussed

Nam Sung-pil, CEO of AB180, opened the evening. His subject was the AI-native organization: not a company that bolts AI onto what it already has, but one designed around AI from the very beginning. A single internal automation system called “A-Bot” had shipped 800 pull requests, one example that made the idea click immediately.

Kim Sung-sik studied CS at Northwestern and began his career at Uber before moving through Google DeepMind (Bard, better known internally as the Code Red team) and Cruise, then joining xAI. Day-to-day life at Google and at xAI could hardly have been more different. Google meant a 20-to-50-hour week and a three-week onboarding; at xAI, he spent thirty minutes setting up his machine on day one and was in the deep end straight after. There were no real working hours, and yes, the tent photos that circulated on Twitter around the Grok 3 launch were real.

Elon met with every team at least once a week. Not executive briefings: he would single out individual engineers, ask what they were working on right now, and drill down at least three levels. He was just as direct about how he felt, and what he disliked most was a direction that did not hold up under first principles. His insistence that self-driving had to be solvable with cameras alone was the classic example.

The project that drew the most interest was Macro Hard. The premise: Microsoft runs almost entirely on software, with no hardware to speak of, so if AI could replace a company like that with no human in the loop, that would be AGI, and the name was a nod to running in the opposite direction. Kim Sung-sik was tapped for the project, but Elon and the team clashed repeatedly over direction, and a restructuring eventually broke the team up. Two mentors he had followed left the company, and he had reached his own limit: he was still writing code ten minutes before his wedding. He left xAI.

Thinking Machines Lab, where he works now, has a different character. Where most labs build technology that replaces people, TML is after a relationship in which people and AI work together. Two things were shared publicly here: multimodality and full duplex. Instead of today's “email-style” exchange, where you paste a screenshot and describe it in text, the goal is an assistant that sees and hears and can be shown things directly, and communication that is not turn-based but JARVIS-like, talking, interrupting, listening, and thinking at once. RSI, where models train models, came up as well.

Every lab defines AGI differently. For OpenAI it is matching the average expert across every profession; for Elon it is a single company running on AI alone; Demis Hassabis applies yet another bar. Kim Sung-sik's own view was that knowledge and reasoning can already outstrip humans, but the generalist knack for picking up something entirely new, and continual learning, are still missing. He pointed to how fast Korea has adopted all of this, first or second in the world in token usage per capita, with terms like “DX” and “AX” now part of everyday speech, and predicted that it will take longer than people expect for every job to disappear. In a post-AGI world, he argued, the human role will actually grow in areas like entertainment, where the job is “setting the standard for what is good and what is not.”