The AI Agent Economy in 2026

Agentic commerce applies agentic AI to buying and selling: an agent finds, compares and purchases products within limits the shopper sets, while brand-side agents answer questions and resolve support requests. As of 2026-10-28, the buying side has open protocols from OpenAI and Stripe and from Google and Shopify, plus checkout inside ChatGPT, Google AI Mode and Gemini. Stellagent counts 90+ companies across 7 ecosystem layers. Analyst forecasts for 2030 differ in size but agree on direction, and Deloitte reports some retailers already receive 15% to 20% of referral traffic from AI chat interfaces. For developers, the practical work is exposing structured catalog, pricing and inventory data through APIs, supporting the emerging protocols, and adding permissions, audit logs and fraud controls.

Key concepts

Agentic commerce

Use of autonomous AI agents that carry out multi-step tasks such as researching, comparing and purchasing products within defined guardrails. Unlike chatbots, agents plan and execute actions on the customer's behalf. S3 · bigcommerce.comS5 · gominimal.ai

Agentic Commerce Protocol (ACP)

Open standard published by Stripe and OpenAI in September 2025, released under Apache 2.0. It focuses on checkout and payments, and ChatGPT was the first AI platform to use it. S2 · stellagent.aiS5 · gominimal.ai

Universal Commerce Protocol (UCP)

Google's standard announced in January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. It covers product discovery through post-purchase support and powers checkout in AI Mode and Gemini for US retailers. S2 · stellagent.aiS5 · gominimal.ai

Agent Payments Protocol (AP2)

Google protocol announced in September 2025 that handles the payment step. Shopper instructions are recorded as signed mandates so the store and payment provider can prove what was approved. S5 · gominimal.ai

Model Context Protocol (MCP)

General open standard for connecting AI applications to outside systems and tools. It is listed in the protocols layer alongside UCP, ACP, AP2, A2A and Visa TAP. S2 · stellagent.aiS5 · gominimal.ai

Agent-readable product data

Structured, timely and consistent product, pricing and inventory data that agents query through APIs. Unstructured, delayed or inconsistent data can keep a product out of the agent's candidate set. S4 · invisibletech.aiS6 · youtube.com

How an agent-driven purchase flows

1. Shopper states an intentThe shopper expresses a goal, such as needing a new laptop under $1,000, or a request like running shoes under 120 euros that arrive by Friday.
2. Model parses structured intentThe model extracts constraints including budget, specifications, brand preference and delivery window from the prompt, replacing the session-based browsing flow.
3. Agent calls commerce APIsThe shopping agent queries product catalogs, pricing services, availability feeds and shipping estimators. Evaluation happens programmatically rather than visually.
4. Agent completes checkoutThe agent builds the cart, selects shipping and payment methods and completes the transaction within the rules the shopper set, such as a budget or delivery date.
5. Post-purchase actions followAgents can handle refunds and subscriptions, and brand-side agents handle tracking questions, returns and address changes after an order.

2030 market forecasts and shopper adoption

Forecasts for 2030 differ in scope and size. Bain projects the US market at $300-500 billion, representing 15-25% of total eCommerce sales. Morgan Stanley forecasts that 10-20% of US eCommerce sales will be agent-driven. Gartner projects that 20% of digital commerce transactions will be executed through AI platforms, via on-platform checkout or AI agents. McKinsey estimates as much as $1 trillion in orchestrated US retail revenue, and as much as $3 trillion to $5 trillion globally. J.P. Morgan estimates up to 25% of U.S. online sales, concentrated in recurring, low-risk categories such as groceries and subscriptions. On the demand side, Kearney estimates that 60% of shoppers expect to use AI agents within the next 12 months and 73% are familiar with AI tools. Among European shoppers, 63% use AI to compare options such as brands, models, prices and reviews, and 55% use AI to learn about a category or product.

The protocol and platform landscape

Stellagent organizes the ecosystem into 7 layers with 90+ companies, and finds the most intense competition in Protocols and Standards. Google unveiled UCP in January 2026 at NRF 2026; Shopify, Walmart, Target, Wayfair and Etsy co-developed it, with over 20 endorsers including Visa, Mastercard, Stripe and American Express. ACP, from OpenAI and Stripe, is open-sourced under Apache 2.0 and focused on checkout and payments; Stripe merchants can integrate with a single line of code. OpenAI paused its Instant Checkout feature in March 2026. AP2 records shopper instructions as signed mandates, and MCP connects AI applications to outside systems. Shopify's Agentic Storefronts, launched in December 2025, send a merchant catalog to ChatGPT, Microsoft Copilot, Google AI Mode and Gemini from one admin setup. VC funding concentrates in the payments and identity layer, leaving checkout execution as a notable gap.

Traffic shifts and what stores must expose

Deloitte surveyed more than 330 senior retail executives globally and reports that some retailers already see 15% to 20% of referral traffic coming from AI chat interfaces rather than traditional search or apps. Airia reports that nearly 6% of all searches flow through AI-powered answer engines, and that retailer traffic from AI sources has surged 1,200% while traditional search traffic is down 10% year-over-year. In this model, agents query APIs instead of navigating storefronts, so backend systems rather than homepage design determine visibility. Retailers that do not optimize metadata, permissions and checkout risk being invisible to assistants such as ChatGPT and Gemini. BigCommerce recommends open, composable architectures with APIs, real-time inventory and rich product data, plus governance controls such as permissions, audit logs and fraud safeguards. Deloitte adds that oversight, exception management and applied judgment are becoming core operating requirements.

UCP versus ACP

Timeline

1. 2025-09
   Stripe and OpenAI publish ACP; Google announces AP2.gominimal.ai
2. 2025-11
   CB Insights publishes a market map listing over 90 companies.stellagent.ai
3. 2025-12
   Shopify launches Agentic Storefronts as a sales channel.gominimal.ai
4. 2026-01
   Google unveils Universal Commerce Protocol at NRF 2026.stellagent.ai
5. 2026-03
   OpenAI pauses its Instant Checkout feature.stellagent.ai

Common pitfalls

• Unstructured, delayed or inconsistent product data means an agent cannot reliably include the item in its candidate set, even though the storefront exists.
• Betting on a single protocol is risky. Analysis cited by Stellagent suggests UCP and ACP will coexist, and supporting both is likely the pragmatic path for merchants.
• Checkout features can change. OpenAI paused Instant Checkout in March 2026, so avoid hard dependencies on one platform's checkout flow.
• Giving agents access to carts and checkout without permissions, audit logs and fraud safeguards undermines safe operation at scale.

Key takeaways

• Forecasts for 2030 range from 10-20% to up to 25% of US online sales; plan for a meaningful but uncertain share.
• Support both UCP and ACP where possible, and expose structured catalog, pricing and inventory data through APIs.
• Pair agent access with permissions, audit logs and fraud safeguards, and keep human oversight for exceptions.

FAQ

How this page is made. Built from the full text of 8 sources found with the AISA Tavily API (advanced search), refreshed weekly. A language model organizes what those sources state into the blocks above; each block links the sources it draws on, and figures that do not appear in the sources are removed.