Every major trust primitive in Web3 was designed for humans.
Private keys assume a human holds them. Multisigs assume a committee of humans deliberates. Even ERC-4337's UserOperation model, as sophisticated as it is, still imagines a human at the top of the call stack who eventually reviews, signs, and is accountable for what gets submitted.
That assumption is already wrong.
Right now, AI agents are signing transactions, managing treasury allocations, executing arbitrage, rebalancing portfolios, and calling external protocols — with no human in the loop, on timescales no human could operate at. And the infrastructure they're running on was never designed for them.
The gap isn't a missing feature. It's a missing layer. A trust layer that answers five questions the current stack cannot:
