Google Pay is redesigning its infrastructure for a form of commerce in which an AI agent, rather than a person, assembles and initiates a purchase. The central idea is to replace checkout pages built for human eyes with stable, machine-readable transactions.
The update introduces the Universal Commerce Protocol, a merchant integration layer, dynamic Android callbacks and broader WebView support. Together, they aim to let agents confirm inventory, handle fulfilment details and adjust an order without navigating a visual website.
The architecture could make automated booking and procurement more reliable, but it also creates new questions about authority, security and dependence on a central platform.
A common protocol gives agents a machine checkout
The Universal Commerce Protocol is intended to standardise communication between agents, merchants and payment systems. Developers would no longer need a custom integration for every store an agent may use.
Google’s new server-side Merchant Commerce Platform manages those merchant connections and analyses transaction patterns. It abstracts backend complexity for developers while giving Google a central role in agent-driven commerce.
Dynamic callbacks allow an order to change during checkout. Shipping costs can update after an address change, or tax can be recalculated without restarting the entire transaction.
Expanded WebView support enables payment inside third-party applications, including social and conversational environments. An agent can therefore move from product discussion to completed transaction without sending the user through several separate interfaces.
Machine customers change how products are discovered
A traditional customer journey is measured through clicks and page views. An agent evaluates structured product information, price, availability and fulfilment conditions through an API.
Merchants must make inventory and policies machine-readable in addition to creating persuasive pages for people. If an agent cannot interpret a product’s attributes or confirm that it meets the user’s requirements, the merchant may become invisible to this channel.
This creates a form of optimisation for machine buyers. Accurate feeds, consistent identifiers and clear return conditions can matter as much as search copy.
The Merchant Commerce Platform also concentrates valuable transaction data. Enterprises should assess what Google can infer from agent purchases, how information may be used and whether the protocol can be implemented independently.
A universal standard reduces integration cost, but a proprietary clearing point can create lock-in. CIOs need exit plans, portable transaction records and clarity about which layer controls the customer relationship.
Human approval becomes a programmable policy
An autonomous agent could make an incorrect or malicious purchase at machine speed. Google’s cross-device biometric authentication allows an agent to request confirmation on another device before the transaction completes.
A user might ask an agent on a laptop to arrange travel, then approve the final amount on a phone. That checkpoint provides evidence of consent without forcing the person to perform every intermediate step.
Approval should follow consequence. Low-value repeat purchases may fit a pre-authorised budget, while unusual merchants, high amounts or changes in delivery details should require fresh verification.
Businesses need rules for spending limits, approved categories, refunds and failed fulfilment. Each transaction should record the agent identity, user instruction, merchant data and approval event so disputes can be investigated.
Biometrics are not a complete safeguard if the agent assembled the wrong product or concealed important terms. The confirmation screen must present the actual item, price and conditions in language the user can understand.
Google Pay’s update is an early sign that agentic commerce requires more than adding a chatbot to a storefront. Payment systems, merchant data and consent mechanisms must be redesigned for software that acts before a human reaches checkout.
The winners will make machine purchasing convenient without making responsibility invisible. A common protocol can simplify commerce, but trust will depend on explicit authority, portable data and human checkpoints that remain meaningful when the customer journey is executed through APIs.
