Curious about your perspective on agent trust layers. What's the biggest technical hurdle you see in implementing verifiable agent state?