Google, Alipay, and Coinbase are building the protocols that let machines buy things. Here is how the plumbing works.

Last week, Alipay quietly crossed a threshold that most of the Western payments industry has not yet noticed. Its AI Pay service, a payment layer purpose-built for AI agents, processed 120 million transactions in a single week. Not 120 million searches. Not 120 million product views. Transactions. Money moved by machines on behalf of humans.

On the same day, Google expanded its agentic shopping experience, letting US consumers buy products from Etsy and Wayfair directly inside AI Mode in Search and the Gemini app. Coinbase unveiled wallet infrastructure built specifically for AI agents. And Expedia told investors on its Q4 earnings call that it is "experimenting aggressively" with agentic commerce integrations, including Amazon's Alexa+.

In our previous analysis, we explored how agentic commerce threatens the marketplace model that companies like eBay and Amazon have built over decades. This piece goes deeper: how does the agentic commerce transaction actually work? What are the protocols, the payment rails, and the trust layers that let an AI agent go from a natural language request to a completed purchase?

Traditional e-commerce was designed for humans. Login forms, CAPTCHA challenges, dropdown menus, card number fields, billing address confirmations. Every one of these steps assumes a person is sitting in front of a screen, clicking through a flow optimised for human cognition.

AI agents do not work this way. They do not have browsers. They do not have fingers. They need APIs, structured data, and machine-readable protocols.

The scale of this shift is already measurable. According to Google, traffic from AI sources to retail sites surged 1,200 percent year over year, while traditional search traffic declined 10 percent. During Cyber Week 2025, 20 percent of global orders were influenced by AI and agents. A WP Engine report found that nearly one third of all web requests now originate from bots.

The problem is not just technical. As Total Retail put it, retailers now face a dual-audience challenge: serving both human shoppers and the AI agents that increasingly influence, or outright make, purchase decisions.

So how does an agentic commerce transaction actually happen? We can break it into three layers: intent, protocol, and payment.

The transaction starts when a consumer tells an AI assistant what they want, in natural language. "Book me a hotel in Paris under 200 euros a night for next weekend." "Order the same coffee I got last Tuesday." "Find me running shoes under 100 pounds with next-day delivery."

The agent interprets the request, extracting parameters like budget, location, timing, and personal preferences. It then queries multiple merchants simultaneously, comparing options across price, availability, reviews, and fit.

This layer is where the major AI platforms are competing. Google Gemini handles it through AI Mode in Search. Alibaba's Qwen App does it inside China's super-app ecosystem. ChatGPT and Alexa+ are building their own commerce capabilities. The entry point varies, but the function is the same: translate human intent into structured, actionable queries.

This is the critical infrastructure layer, and the one where the most consequential competition is playing out right now.

For an AI agent to buy something, it needs a standardised way to discover products, negotiate terms, and structure a transaction. It cannot navigate a website the way a human does. It needs a protocol.

Google's Universal Checkout Protocol (UCP) is the most ambitious attempt to build this standard. Announced at the NRF conference and co-developed with Shopify, UCP is an open-source protocol that establishes a common language for agents and merchant systems to work together, from discovery through to purchase and post-purchase support.

The partner list is striking. More than 20 companies have endorsed UCP, including Visa, Mastercard, Stripe, Adyen, American Express, Target, Walmart, Best Buy, Macy's, The Home Depot, Flipkart, and Zalando. "Open, interoperable protocols are the spark for agentic commerce," Pablo Fourez, Chief Digital Officer at Mastercard, told FinTech Magazine.

UCP is designed to work across verticals and is compatible with existing industry protocols including Agent2Agent (A2A), Agent Payments Protocol (AP2), and Model Context Protocol (MCP).

In China, Alipay has taken a different approach with its Agentic Commerce Trust Protocol, launched in January 2026. Where UCP focuses on commerce interoperability, Alipay's protocol emphasises transparency, secure authorisation, and clear user consent in agent-based transactions. The Qwen App became its first adopter, connecting with Taobao Instant Commerce and Alipay's AI payment service to let users place food and beverage orders through chat interactions.

Coinbase is approaching the problem from the crypto side with its x402 protocol, which makes payments native to the HTTP layer itself, letting agents pay for resources as naturally as they request web pages.

Once the agent has found what the consumer wants and structured the transaction, it needs to actually move money. This is where identity, authorisation, and trust become critical.

The fundamental question: how does a merchant know the agent is authorised to spend the consumer's money?

Google's answer is to use Google Pay as the credential provider for UCP-powered checkouts, with PayPal integration coming next. The protocol is designed to be agnostic, supporting any credential provider.

Alipay's answer is its AI Pay service, which handles authorisation within the same digital environment as the conversation. The agent interprets the request, verifies the order, and processes the payment without the consumer navigating multiple screens. It has expanded beyond apps into physical retail with Luckin Coffee and even AI smart glasses with Rokid, where developers can integrate payment capabilities using natural language via MCP.

Coinbase's answer is AgentKit, wallet infrastructure that gives AI agents their own crypto wallets with programmable spending rules and guardrails.

Traditional card networks are adapting too. Mastercard is extending its Agent Pay scheme through UCP. Visa has its own Trusted Agent Protocol in development. Stripe and Adyen are both UCP endorsers, positioning their existing card processing infrastructure as the rails these new protocols will run on.

This is not theoretical. As we tracked the developments this week, transactions are already happening.

Google and Shopify have launched UCP-powered checkout for eligible US retailers directly within Google Search and Gemini. "Shopify has a history of building checkouts for millions of unique retail businesses," said Shopify Vice President Vanessa Lee. "We have taken everything we have seen over the decades to make UCP a robust commerce standard that can scale."

Alipay AI Pay is processing 120 million transactions per week across apps, mini-programs, brick-and-mortar retailers, and smart devices. In China, the super-app model means agents can order from Luckin Coffee, book services through Taobao, and pay through Alipay without ever leaving the conversation.

Expedia is integrating with multiple AI platforms, with CEO Ariane Gorin telling investors that agentic commerce "opens up new growth opportunities." According to PYMNTS Intelligence, 25 percent of consumers are already comfortable letting AI agents plan their travel.

For all the momentum, significant challenges remain.

Identity and trust are the biggest. When an AI agent presents a payment credential, how does the merchant verify it is acting within the consumer's authorised parameters? Spending limits, merchant categories, geographic restrictions: these controls exist for human cardholders but have no established equivalent for agents.

Fraud at machine speed is a growing concern. If agents can complete transactions in seconds, fraudulent agents can too. The screening and monitoring infrastructure that catches suspicious human behaviour may not be calibrated for machine-speed transaction patterns.

The checkout gap persists. Practical Ecommerce reports that merchants are still debating whether to block AI bots entirely, let alone optimise for them. The retailers who move fastest to support agent-readable product data and API-first checkout will capture disproportionate agent-driven revenue.

Regulatory uncertainty looms. As agentic commerce scales, questions around consumer protection, liability, and data privacy become urgent. If an AI agent books the wrong hotel or overpays for a flight, who is responsible: the consumer, the agent platform, or the merchant?

The agentic commerce stack is being assembled in real time. Google's UCP gives us the closest thing to an industry standard, but Alipay's Trust Protocol and Coinbase's x402 show that different markets and use cases will likely produce different approaches.

The pattern is familiar to anyone who has watched payments infrastructure evolve. We saw it with mobile payments, where Apple Pay, Google Pay, and Alipay each won in different geographies. We saw it with open banking, where PSD2 in Europe and open banking mandates in the UK created frameworks that the US still lacks. Agentic commerce is following the same trajectory: moving fast, moving globally, and moving unevenly.

What is different this time is the speed. Alipay went from launching AI Pay in 2025 to 120 million weekly transactions in early 2026. Google went from announcing UCP to live checkout in weeks. The infrastructure is not being built and then adopted. It is being built and adopted simultaneously.

For our readers in payments and commerce, the implication is clear. The protocols are being written now. The standards are being set now. The merchants who plug into this stack early will be where the agents shop. The ones who do not will be invisible to an increasingly machine-driven commerce layer.

Is your checkout ready for its next customer: a machine with a wallet and a mandate to buy?