In September 2025, ChatGPT users in the US started buying products without leaving the chat, through Instant Checkout, built by OpenAI with Stripe. Four months later, in January 2026, Google announced the Universal Commerce Protocol (UCP), a standard for transactions between AI agents and merchant systems.
This represents the current state of commerce, not a preview of some distant future.
The Emergence of Non-Human Customers
For two decades, businesses have optimized websites exclusively for human visitors. A second type of customer has now arrived, and it operates on entirely different principles.
McKinsey projects that AI agents will orchestrate between $3 and $5 trillion in global commerce by 2030, with agent-driven retail in the US alone potentially reaching $900 billion to $1 trillion.
The infrastructure is arriving fast. ChatGPT Instant Checkout launched in September 2025 with US Etsy sellers and now also covers select Shopify merchants such as Glossier, SKIMS and Spanx, with OpenAI saying more than a million Shopify merchants are coming. Google AI Mode is testing agentic shopping features that let users browse, compare and purchase without leaving search. In McKinsey's research, 44% of people who had tried AI-powered search called it their primary and preferred way to search.
Most businesses remain focused on chatbot implementations while this fundamental shift unfolds around them.
Understanding Agent Commerce
Agent commerce refers to AI systems that discover, evaluate, negotiate, and purchase products on behalf of consumers. These are not customer service chatbots; they are autonomous buying agents that complete transactions.
Three interaction models have emerged. In Agent to Site (A2S), an AI agent visits a website, scans product pages, compares prices, and initiates checkout programmatically. The visitor is software evaluating offerings against competitors in milliseconds, not a human clicking through pages.
In Agent to Agent (A2A), a consumer's shopping agent communicates directly with a retailer's AI agent. These agents negotiate and coordinate transactions. One agent specifies requirements ("running shoes under $150 with free returns"), and the other responds with matching inventory and processes the transaction.
In Brokered Agent to Site, intermediary platforms connect consumer agents with merchant systems, similar to how OpenTable works for restaurant reservations but applied across all commerce categories. Restaurant bookings, hotel reservations, and grocery orders all flow through agent-to-agent communication.
The consistent pattern across these models is that humans establish intent while agents execute transactions. Your website becomes the point where transactions either occur or fail.
The Universal Commerce Protocol
In January 2026, Google announced UCP (Universal Commerce Protocol), an open-source standard co-developed by Google and Shopify with Etsy, Wayfair, Target and Walmart, and endorsed by more than 20 payment and retail companies including Visa, Mastercard, American Express and Stripe.
UCP functions as the HTTP equivalent for agent commerce. Just as HTTP standardized browser-server communication, UCP standardizes how AI agents interact with merchant systems.
UCP Capabilities
UCP enables unified checkout. Rather than building separate integrations for ChatGPT, Perplexity, Google AI Mode, and other platforms, merchants implement UCP once and gain compatibility with all compliant agents.
Identity linking operates through OAuth 2.0-based authentication, allowing agents to securely connect to customer accounts. Loyalty programs, saved addresses, and payment methods become accessible to authorized agents without credential exposure.
Order management functions through real-time webhooks for status updates, shipment tracking, and return processing. Agents manage the complete transaction lifecycle on behalf of customers, extending beyond the initial purchase.
Discovery works through a manifest published at /.well-known/ucp that communicates product offerings, capabilities, and return policies to agents. This enables programmatic business discovery and understanding.
The Protocol Ecosystem
UCP operates on REST and JSON-RPC transports and integrates with other emerging protocols. MCP (Model Context Protocol) is Anthropic's standard for connecting AI to external tools and data. ACP (Agentic Commerce Protocol) is the OpenAI and Stripe protocol specialized for checkout flows. A2A Protocol is Google's specification for agent-to-agent communication. AP2 (Agent Payments Protocol) provides cryptographic proof for agent-initiated payments.
These protocols function as complementary layers rather than competitors. UCP handles commerce flow, ACP specializes in checkout, MCP connects agents to tools, A2A enables negotiation, and AP2 secures payments.
The Agent Visibility Gap
Most websites remain invisible to AI agents. Human visitors perceive design, headlines, and testimonials. AI agents perceive structured data, or the absence of it.
Structured data (JSON-LD) for products, pricing, FAQs and organization details helps agents read a page accurately, as does semantic HTML with clear headings and labeled sections. Agents also benefit from machine-readable APIs for inventory and pricing, consent flows that can be queried in code, and real-time accuracy: outdated prices or wrong stock levels cost an agent's trust.
Most websites offer visuals designed for human perception, content embedded in JavaScript that agents cannot parse, pricing accessible only through multi-step human navigation, and FAQs formatted as prose paragraphs rather than machine-readable Q&A structures.
This gap represents where businesses lose market share without recognizing the cause.
How Much Does llms.txt Matter?
llms.txt is a proposal to publish a markdown summary of a site at /llms.txt. Jeremy Howard of Answer.AI proposed it in September 2024, and it has spread mostly among developer documentation sites.
Its effect is smaller than often claimed. Google's 2026 guidance says Google Search doesn't use llms.txt and that creating one neither helps nor hurts visibility, and the crawler documentation of OpenAI, Anthropic and Perplexity doesn't describe using it for search or citations. It's cheap to add, but it isn't a lever for citation rates. Crawler access, structured data and accurate page content come first.
Agent Experience as a Design Discipline
User Experience (UX) has defined digital design for two decades, focusing on how humans interact with websites. Agent Experience (AX) addresses how AI agents interact with sites.
AX encompasses discoverability (whether agents can locate products without HTML scraping), interpretability (whether content structure enables contextual understanding), transaction-readiness (whether agents can complete purchases without human intervention), and trust signals (whether sites demonstrate the data freshness, accuracy, and credentials that agents prioritize).
Organizations optimizing for AX will capture the agent-driven marketplace. Organizations that neglect AX will observe declining traffic despite stable search rankings.
Implementation Roadmap
Within the next six months, organizations should audit structured data to verify JSON-LD product information and machine-readable FAQs, check that AI search crawlers and agents can reach the site (robots.txt and CDN bot rules), evaluate whether checkout flows are reachable through APIs or locked behind JavaScript, and track AI-referred traffic separately.
Between six and eighteen months, e-commerce organizations should implement UCP endpoints, build or expose APIs for inventory, pricing, and availability, develop agent-friendly content parallel to human-friendly content, and create verification frameworks for agent transactions.
Beyond eighteen months, organizations should evaluate building consumer-facing agents or platform partnerships, redesign content strategy for dual human and agent audiences, and invest in observability systems tracking agent data access, interpretation, and recommendations.
The Cost of Delay
In McKinsey's survey, only 16% of brands systematically tracked AI search performance. The remaining 84% can't tell whether agents recommend their products or a competitor's.
Half of consumers already seek out AI-powered search on purpose. When answers arrive without a click, buying decisions can be made before anyone reaches the merchant's site.
Organizations implementing UCP and agent-ready infrastructure now will capture emerging market share. Organizations waiting for market clarity will spend 2028 investigating collapsed digital revenue.
Anymorph's Position
Anymorph builds for a commercial environment where websites must serve human and agent audiences simultaneously.
The platform generates pages optimized for both audience types, with automatic structured data implementation, maintained content freshness, and agent-friendly formats alongside human-readable content.
Standards like UCP are still settling. Anymorph measures how AI answers mention a brand and what they cite, and uses that to publish pages on the brand's own domain that both people and agents can read easily.
Agent commerce has arrived. The question is whether your website is prepared to serve customers that never click, never scroll, and never read testimonials while still generating revenue.
Sources
- OpenAI, Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol (September 2025)
- OpenAI Help Center, Instant Checkout
- McKinsey, The agentic commerce opportunity (October 2025)
- McKinsey, New front door to the internet: Winning in the age of AI search (October 2025)
- Universal Commerce Protocol
- The llms.txt proposal
- Google Search Central, Optimizing your website for generative AI features on Google Search
