Most discussions about agentic systems focus on capability: how smart agents are, what tasks they can complete, how autonomous they can be.

The harder problem is delegation.

Delegation changes the trust model. When a human clicks a button, intent is implicit. When software acts on someone’s behalf, intent has to be made explicit.

That shift exposes a set of problems that show up regardless of industry:
• authority has to be defined, not assumed
• limits have to be enforceable, not implied
• disclosure has to be minimised, not convenient
• outcomes have to be explainable after the fact

Commerce makes these problems unavoidable because money, liability, and regulation are involved.

When something goes wrong — a disputed transaction, an incorrect purchase, a regulatory inquiry — the question won’t be which agent framework was used. It will be whether there is clear, verifiable evidence of what was authorised, under what conditions, and by whom.

This is where many early agentic implementations may start to strain.

Session-based trust doesn’t survive once authority is delegated. Platform boundaries blur. Agents act across systems. Humans are sometimes in the loop, sometimes not. And the assumptions that made traditional commerce flows feel safe no longer hold.

At that point, trust can’t remain contextual. Authority needs to survive delegation, independent of which platform invoked the action or which interface rendered the experience.

Once delegation becomes explicit, the question is no longer whether trust needs to scale — but how.