Agents Keep Levelling Up: Why Google’s New Agents Payment Protocol Is Worth Paying Attention To
On September 16, Google announced the launch of Agents Payment Protocol (AP2), a new payments framework designed to allow AI agents to initiate and complete purchases on behalf of users. AP2 enables an agent to search for, initiate and purchase goods and services based on explicit user intent, operating as the transactional layer in the emerging stack of agent-to-agent (A2A) and Model Context Protocol (MCP) standards. If MCP is how agents interact with tools and A2A defines how agents communicate with each other, AP2 is positioned as the mechanism by which agents securely transact money.
AP2 is built on intent mandates, in which users provide explicit purchase instructions with constraints such as price limits. For example, if an item is out of stock, the user can authorize the agent to buy it when available and approve a specific premium. This mandate pairs with a cart mandate; once the agent adds the item to the cart, the user confirms the purchase. Early implementations require human verification, but Google plans to support human-not-present flows, in which cart mandates are pre-authorized for future actions, such as an upcoming ticket release. By framing intent explicitly, the risk of model hallucination with AP2 is low, as it operates on a clear binary between authorized and unauthorized transactions.
Interoperability with A2A adds another dimension, while MCP extends AP2’s utility by standardizing access to payment APIs and financial rails. In enterprise workflows, ServiceNow’s launch of AI Agent Fabric and planned A2A integration underscores AP2’s potential, with Google confirming ServiceNow supports the protocol. ServiceNow agents could interact with ERP agents from providers such as Oracle and SAP, express procurement intent and have ERP agents complete purchases through AP2 – handling authentication and non-repudiation without provider-specific logic. In supply chain workflows, agents from firms such as Akira AI and C3 AI already flag disruptions, draft supplier outreach and generate replacement orders; with AP2, they could also execute those transactions with fewer bottlenecks and complications.
Google has already listed more than 60 firms supporting the protocols, including American Express, Etsy, Mastercard, PayPal, Revolut and Salesforce. Despite strong early momentum, Verdantix notes several limitations and early hurdles that need to be overcome. Adoption, especially from smaller developers, may lag until public software development kits (SDK) and local MCP server hosting is available, primarily due to the absence of framework‑agnostic tooling. Intent‑driven authority also raises governance risks; without strong controls, children or other unintended users could trigger purchases via chatbots.
Although technical maturity and ecosystem adoption of AP2 are still open questions, enterprises and technology vendors should begin assessing where automated procurement and agent-led transactions may introduce efficiency or risk to their operations. For more insights on the latest AI-related news and updates, visit the Verdantix AI Applied Insights page or book an analyst inquiry.
About The Author

Aleksander Milligan
Analyst