Agentic Commerce

What Is Agentic Commerce?

Agentic commerce describes the shift from AI that surfaces recommendations to AI that executes transaction-related tasks.

Rather than a person navigating websites, comparing options, and manually checking out, an AI agent handles much of the process, from interpreting what you want to executing the purchase, often with reduced human input at each step. Many implementations still require human approval for payments or final commitment, particularly for higher-value transactions.

For example, a user might say, “Find me a nonstop flight to London next week for under $600.” The agent would then search across airlines, check loyalty memberships, weigh trade-offs, and propose or even complete the booking. 

How Do AI Agents Execute Purchases?

An agentic commerce system typically runs through several stages: interpreting the user’s intent, querying product or service data across multiple sources, comparing options against predefined preferences, and executing a transaction once conditions are met.

This architecture depends on machine-readable data and interoperability standards. The Model Context Protocol (MCP), for example, allows agents to connect to tools and data sources such as inventory systems and product catalogs, giving them structured access to the information needed to act. 

It’s worth noting that a separate payment infrastructure is required to authorize and settle transactions, potentially utilizing agentic wallets. Emerging examples include Google’s Agent Payments Protocol (AP2) and Mastercard’s Agent Pay, though these are still developing rather than settled standards.

Can AI Agents Transact Securely on Your Behalf?

When agents transact autonomously, the question of who controls the spending matters enormously. Guardrails, spend limits, and delegated authorization frameworks define what an agent can and cannot do without human approval. 

Software-only security is insufficient here. Agents that can access real resources, execute autonomously, and respond to external inputs create a compounding attack surface where a single compromised step can trigger a chain of damaging consequences.

Ledger’s 2026 AI Security Roadmap addresses this directly with a hardware-anchored security stack built for the agentic economy. The human sets policies upfront, which a Hardware Security Module enforces on every subsequent agent action. Agents operate autonomously inside those boundaries, and anything outside the approved policy is rejected at the hardware level before it can execute.