Vibe Coding the Justice System

Barrett and Greg dig into the strange new phase of the AI race: the companies building the most powerful models are increasingly warning that development may be moving too quickly while simultaneously racing one another to stay at the frontier. They debate whether calls to “pace the frontier” reflect genuine concern, financial incentives, regulatory capture, or some combination of all three. If increasingly capable AI systems are inherently unpredictable, can labs ever build a perfect enough safety harness or should we instead design the physical world around the assumption that eventually something goes wrong? That leads to a broader question: how much control should AI ever be given outside the digital world? Greg argues for keeping humans firmly in the loop around physical infrastructure, transportation, weapons, and other high-stakes systems and definitely not putting AI inside a robot that can do backflips and kick watermelons apart. But the harder problem may be that AI becomes too useful to resist. Barrett and Greg explore what happens when we gradually delegate more of society’s complicated systems to machines. Would flawless AI referees improve sports or ruin part of what makes them interesting? Could AI judges make the justice system faster and more consistent? And if every business leader is leaning on AI to make decisions, where does human judgment and competitive advantage still matter? In the second half, Barrett tries Meta’s new proactive AI assistant and comes away unexpectedly impressed. Instead of waiting for a prompt, the agent watches email and calendars, surfaces important events, and acts on the user’s behalf. Greg argues that this is much bigger than another chatbot: the personal AI agent could become the new browser and the primary interface to the internet. People may stop visiting websites, shopping manually, waiting on hold, or even checking email themselves. Instead, they’ll declare intent and let their agent handle the rest. That raises a new set of questions around advertising, e-commerce, agent-to-agent communication, privacy, healthcare data, and whether access to a capable personal agent will eventually become as necessary as owning a smartphone. 00:00 — Catching up and frustrating AI usage limits 02:25 — Should frontier AI development slow down? 05:00 — Have the AI labs seen something we haven’t? 06:50 — Safety concerns or regulatory capture? 09:00 — IPO pressure and the AI arms race 10:55 — Why can’t Anthropic just slow itself down? 12:15 — What should governments actually regulate? 14:35 — Assume the AI goes rogue 15:15 — Please don’t put AI in a superhuman robot 17:30 — Can the AI labs be trusted to regulate themselves? 18:20 — Steve Eisman on the AI “crisis” 20:40 — Do frontier AI labs actually have a moat? 23:00 — Microsoft, Meta and the race to catch up 25:30 — The real problem: AI is non-deterministic 27:40 — Could AI actually kill everyone? 29:15 — Planes, cars and keeping humans in control 32:55 — We’re already delegating things we don’t fully understand 34:50 — Would robot referees make sports better? 38:20 — Would an AI judge actually be fairer? 41:00 — AI may simply become too useful to resist 42:15 — Does AI destroy business competitive advantage? 44:30 — Barrett tries Meta’s new AI agent 48:15 — The agent that prompts you 51:10 — Why Meta may finally have an AI strategy 52:00 — The agent is the new browser 52:55 — Why everyone may eventually need a personal agent 55:05 — How Meta could actually make money from AI 56:15 — Will premium agents create a new digital class system? 59:15 — AI agents, shopping and payments 1:00:45 — What happens to advertising when nobody visits websites? 1:02:30 — Search, Google Reviews and the post-browser internet 1:04:00 — Can people actually trust an AI with their email? 1:06:40 — When agents start gaming loyalty programs 1:08:20 — AI, healthcare and sensitive personal data 1:10:40 — Wrap-up