This is the chapter the resource exists for. AI agents can already hold wallets and sign transactions on every major chain, and the same wave of tokenisation work that put deposits, money-market funds, and central bank money onto programmable rails has, almost as a side effect, put those instruments within reach of agents. The question is whether the legal and regulatory perimeters around each instrument accommodate, ignore, or implicitly block agentic use, and where the architectural opportunities sit for a tokenisation product team building today. Most of the action is in the silence: regimes were written for human and corporate holders, and the perimeters are now being implicitly tested by flows that look like person-to-person flows but are not. The chapter walks the question instrument by instrument, names the architectural primitives that have emerged, and identifies the open regulatory pressure points.
Definition
Agentic commerce is the use of AI agents (autonomous software processes, typically LLM-driven, executing tools under a delegated authority) to hold, send, receive, or commit money on behalf of a principal, where the principal is a human, a corporation, or another agent. Agentic commerce is not a category of instrument. It is a class of flow that runs across many instruments, and the choice of instrument matters more than the choice of agent stack.
The interesting question is the legal one. The agent is not, in any current jurisdiction, a legal person. It cannot enter into a contract in its own name, be sued in its own right, or satisfy know-your-customer (KYC) obligations as an account holder. What it can do, on every public chain and on the permissioned chains that allow the relevant transfer pattern, is hold a wallet, sign a transaction, and move tokens. The friction sits at the boundary between what the agent can technically do and what the law treats it as having done.
The legal frame is principal-and-agent. Common-law agency principles and their civil-law equivalents in japan, south korea, Hong Kong's hybrid system, and Singapore's common-law-derived framework treat an agent as binding the principal when the agent acts within the authority granted. None of these doctrines require the agent to be a natural person. The principal grants authority. The agent acts within it. The principal is bound by authorised acts and not bound by acts outside the authority unless the third party can rely on apparent authority. The doctrines have answers in principle for what happens when the agent malfunctions, hallucinates, or acts outside its authority. The case law on LLM-driven agents is thin.
A working test for whether a flow is genuinely agentic in the sense this chapter cares about: if the human principal is reviewing and approving each transfer before it is signed, the agent is a UI. If the principal has delegated authority for a class of transfers and the agent decides which specific transfers to make within that authority, the agent is operating agentically. The legal interface is the delegation, not the chain.
That distinction is load-bearing for what follows. The product question is not whether to give an agent its own KYC, license, or balance-sheet capacity. The product question is whether the principal's authority can be expressed cleanly enough that a relying party (a counterparty, an issuer, a chain-level protocol) can verify, at transaction time, that the agent's act is within scope. The Bank for International Settlements (BIS) has framed the same point at the macro level in its Annual Economic Report 2024 chapter on AI and tokenisation, arguing that the convergence of agents and programmable rails will be operative before regulators write text addressing it. The rest of this chapter is the operator-level version of that argument: which instruments accommodate it today, which architectural primitives have emerged to support it, and where the regulatory perimeters are silent in ways that will become live as flows scale.
The next part walks the five gating questions that determine whether agentic commerce is operationally workable on a given tokenised rail, and then the instrument-by-instrument view from payment stablecoins through wholesale CBDC.