Box CEO Aaron Levie argues the key architectural question of the 21st century in business is how to maximize corporate IP — decisions, insights, workflow patterns, best practices — in a world where so much intelligence is packed into AI. Having frontier AI is just the start; leveraging it uniquely is the real competitive edge.

Replit CEO Amjad Masad demoed Vibe Research: fine-tuning a Qwen-8b model to play chess on Replit, running 3 parallel experiment branches. AI capability for doing ML has improved dramatically — someone with good intuition can now do interesting ML work.

Traditional nuclear focuses on reducing odds — make sure it never happens. Valar's approach: reduce consequence. Design assumption: everything in the plant has failed. Even if all systems fail simultaneously, the reactor won't melt down. TRISO fuel + graphite moderation + helium cooling. They'll demo this Friday: shut off all safety systems and let physics handle decay heat for 2 days.

Isaiah has been obsessed since high school with why did we stop building reactors in the 70s? He spent a decade waiting for someone else — watching every nuclear startup, concluding no one would solve it at the required pace and scale. So he stopped waiting and started building.

Valar deliberately hires from outside nuclear — people who built hard things in the real world. They also bring back people who left nuclear frustrated by its slowness, went and built hardware elsewhere, then returned. One filter question: Do you want to do science and write papers, or build reactors and turn them on?