The Great Entry Migration: agent commerce infrastructure takes shape
- industry
Target reader: Global ecommerce founders and operators · decision-makers evaluating when to enter agent commerce
We call this shift the Great Entry Migration: the first place consumers go to find products is moving from search boxes and marketplace shelves into conversation. Search and marketplaces remain central. Brand competition now extends from earning the click to making an agent's shortlist.
Protocols, shopping surfaces, and checkout paths are being reorganized
The timeline records what a named platform introduced, launched, or changed on a stated date. Platform facts stop at the action itself. Market interpretation follows separately.
| Date | Platform action | Source |
|---|---|---|
| Sep 2025 | Google introduced AP2, using signed mandates to define authorization for agent payments | Google Cloud |
| Sep 29, 2025 | OpenAI and Stripe launched Instant Checkout and open-sourced the Agentic Commerce Protocol (ACP) | OpenAI · Stripe |
| Oct 2025 | Doubao connected to Douyin Mall | TMTPost |
| Oct 14, 2025 | Walmart and OpenAI announced a partnership to bring product discovery and shopping into conversation | Walmart |
| Dec 5, 2025 | Salesforce reported that AI and agents influenced 20% of purchases worldwide during Cyber Week | Salesforce |
| Jan 11, 2026 | Google and Shopify introduced Universal Commerce Protocol (UCP) | Google · Shopify |
| Jan 2026 | Microsoft launched Copilot Checkout | Microsoft · GeekWire |
| Mar 2026 | OpenAI retired the initial Instant Checkout model, kept merchant checkout, and shifted its focus to product discovery | OpenAI · Retail Dive |
| Mar 2026 | Google added a simplified UCP onboarding path to Merchant Center | Google Merchant Center |
| May 11, 2026 | Qwen and Taobao connected product recommendations, ordering, fulfillment, and after-sales service | Xinhua News Agency |
| May 2026 | Doubao introduced Help Me Choose, connecting product discovery to Douyin ecommerce | TMTPost |
| May 13, 2026 | Amazon retired the Rufus chatbot and put Alexa at the center of its shopping-agent strategy | CNBC |
| May 2026 | Google introduced Universal Cart and transferred AP2 stewardship to the FIDO Alliance | Google I/O · FIDO Alliance |
| Jun 2026 | Volcano Engine rebranded its web search as Doubao Search, a retrieval service for large models with site-authority tiering on results | Volcano Engine docs |
| Jul 15, 2026 | Douyin ecommerce's revised technical-service-fee rules took effect, setting fee rates by product category and transaction channel, with Douyin Mall and Doubao as named channels | Douyin ecommerce rules |
| Aug 5, 2026 | Shopify reported that AI-driven traffic and orders had each reached three times the prior-year level | Shopify Q2 2026 |
| Aug 10, 2026 | Douyin Local Services' specific-channel software service fee policy took effect, with Doubao as a named channel and channel attribution shown on order settlement statements | Douyin Local Services |
Different paths are converging on product discovery
Protocols are separating by function. ACP, UCP, and AP2 address product exchange, commerce coordination, and payment authorization. The protocols govern different layers. Merchants still share one requirement: accurate product data that agents can retrieve and compare.
Discovery and checkout are separating. OpenAI retired the initial Instant Checkout model while continuing to invest in product discovery. Merchants keep their own checkout, and discovery moves further into the conversational interface.
Two market paths are developing in parallel. Open protocols connect multiple merchant systems. Integrated platforms bring product data, payments, and fulfillment into their own interfaces. Both paths begin with retrievable product data.
Two markets, one direction. Western platforms are building protocols and connection standards. Chinese platforms are writing the AI entry point into source tiers and settlement rules. The actions differ; the investment target is the same: machine-readable, checkable product and brand data.
The market is still early. The operating signals are already visible
Current purchasing behavior. Salesforce reported that AI and agents influenced 20% of purchases worldwide during Cyber Week 2025. The measure covers influence on a purchase, not a transaction completed entirely by an agent.
Latest merchant measurement. In the second quarter of 2026, Shopify reported that AI-driven traffic and orders had each reached three times the prior-year level. The company also found that 75% of AI-attributed purchases occurred in categories outside its top 100.
Current scale. eMarketer estimates that AI-related retail ecommerce in the United States will account for roughly 1.5% of ecommerce in 2026, nearly four times the 2025 level. A 1.5% share marks an early market. Growth at nearly four times the prior-year level puts it on the operating agenda.
Long-range forecast. McKinsey estimates a $3–5 trillion global opportunity for agent commerce by 2030. This is a long-range forecast, not current transaction volume.
ChatGPT referrals still trail established channels in volume and conversion
A 12-month study covering 973 ecommerce sites with a combined $20 billion in annual revenue compared more than 50,000 ChatGPT-referred transactions with 164 million transactions from established channels. ChatGPT referrals accounted for less than 0.2% of site visits during the study period. Organic search converted at a rate roughly 13% higher, and the gap was wider for affiliate links.
These findings locate the new channel in its current phase: small in volume, fast-growing, and still forming its buyer base. The immediate work is to build reusable product data, a measurement baseline, and a correction record. Near-term orders form one part of the investment case. Reusable operating assets carry the longer-term value.
Shopify describes AI as a complement to search. Brands can continue to operate established channels while building a distinct capability for the new shortlist entry point.
Early entry starts the operating record
Entry-point migration changes what a product needs to reach the shortlist. Catalog data must be machine-readable, price and availability retrievable, and specifications, use cases, and policies complete enough for comparison. Together, these conditions shape whether a product enters the agent's consideration set.
Earlier connection starts three long-term foundations sooner: product data agents can retrieve, a baseline that makes later results comparable, and a continuous record of corrections and retests. Each later result has a prior point of comparison, and each correction carries an existing history forward.
Four foundations brands can build now
- Build machine-readable product data. Complete product identity, price, availability, delivery coverage, and specifications so agents can retrieve and compare the catalog.
- Establish the first measurement baseline. Use real shopping questions to record whether the brand is mentioned, cited, recommended, or absent. Retain the test time and original responses.
- Bring agent-commerce sources into attribution. Add source parameters to product-data endpoints so existing analytics can identify the endpoint and product.
- Prepare for protocol and platform paths. Protocol rails and integrated platform entry points follow different models. Both depend on accurate, retrievable product data.
The Great Entry Migration can now be tracked as a dated infrastructure process. Protocols and transaction paths will keep changing. The immediate operating priority is to make product data retrievable, establish comparable measurements, and preserve a record that each later cycle can extend and correct.
