The Complete Guide to Agentic Commerce — Definition, How It Works, Protocols, Key Players & Market Forecasts [2026 Edition]

What Is Agentic Commerce?

Agentic commerce is a new form of commerce in which AI agents autonomously search for, compare, negotiate, and purchase products and services on behalf of users.

In traditional e-commerce, humans search for products in a browser, read reviews, compare prices, add items to cart, and enter payment details. In agentic commerce, AI handles most of this process. A user simply says, "Get me camping gear for next week, budget $500," and the agent searches across multiple e-commerce sites, finds the optimal combination, gets approval, and completes the purchase.

The key difference from conventional rule-based bots and recommendation engines lies in the ability to reason, plan, and operate across multiple systems. McKinsey defines the concept as "AI that anticipates consumer needs, navigates options, and executes transactions."

There's another critical point to understand: AI agents don't browse web pages. Beautiful design, clever copywriting, strategic product placement — all meaningless to agents. What agents understand is structured data, API responses, and machine-readable protocols. This fact is precisely why the technology stack underpinning agentic commerce is necessary, and why it demands a fundamental shift in thinking from e-commerce businesses.

The Full Picture of Agentic Commerce

Agentic commerce is not a single technology or product — it comprises seven domains evolving simultaneously to form a unified ecosystem.

1. Platforms & Players — The AI that shops and the commerce infrastructure that accepts it
2. Protocols & Tech Infrastructure — The common language connecting agents to commerce
3. Payments & Trust Layer — The mechanisms for safely flowing money and trust
4. Data & Marketing — Product data strategy for being chosen by AI
5. Regulation & Governance — Legal and standardization frameworks
6. Market Size & Forecasts — Projections ranging from $144 billion to $9 trillion, and why they differ
7. Industry Use Cases — Which industries will change first

These domains don't exist in isolation — they are deeply interdependent. Protocols define payment security, platforms drive protocol adoption, data quality determines visibility on platforms, and regulation sets the rules for everything. The following sections explore each of these seven domains in depth.

Platforms & Players

The "shopping side" of agentic commerce already has major AI platforms in position.

The largest by users is ChatGPT. With 900 million weekly active users, it commands 87.4% of AI referral traffic. After pulling back from Instant Checkout, it pivoted to product discovery, with merchant apps from Instacart, Target, and Expedia handling payments within ChatGPT.

Google is rolling out UCP-based Agentic Checkout within AI Mode, starting with Wayfair, Chewy, Quince, and Shopify merchants. It is also launching Business Agent (a virtual sales associate that speaks in the brand's voice) and Direct Offers (exclusive offers within AI Mode).

Amazon Rufus is available to 300 million active customers and processes 274 million daily queries. While generating approximately $12 billion in incremental annual sales, it blocks all external AI agents. Only Rufus, Alexa+, and Buy for Me can access its 600 million product listings — a closed ecosystem by design.

Perplexity has integrated search and purchasing through Buy with Pro, while Microsoft Copilot Checkout leverages PayPal infrastructure to connect with Urban Outfitters, Etsy, and others. Purchase rates via Copilot are reported to be 194% higher when shopping intent is present.

On the "accepting side," commerce infrastructure is adapting rapidly. Shopify enabled Agentic Storefronts by default in March 2026, connecting 5.6 million stores to ChatGPT, Copilot, and Google AI Mode. AI-attributed orders grew 11x, and early adopter Tatcha achieved 3x AI conversion rate and 38% higher AOV.

The most important lens for reading this landscape is the open vs. closed tension. Amazon blocks all non-proprietary AI in a deliberate isolation strategy, while Shopify opens its stores to every external AI agent. This isn't about which approach is "right" — it reflects a structural tension rooted in platform characteristics. PayPal is converting its 400-million consumer network into agentic commerce infrastructure, and Walmart's Super Agent "Sparky" commits to both open and closed ecosystems on an MCP foundation.

With over 90 players in the 2026 agentic commerce market, a comparison of major AI shopping agents helps e-commerce businesses determine which channels to prioritize.

Protocols & Tech Infrastructure

Connecting AI agents to commerce systems requires a common language. From late 2025 through 2026, the protocols forming this "common language" rapidly assembled into a layered architecture. Eight major layers are now identifiable.

At the top, the commerce flow layer is where Google and OpenAI are battling for dominance.

Google's UCP (Universal Commerce Protocol) follows a decentralized design where merchants simply publish a JSON profile at /.well-known/ucp on their own domain, making it accessible to any AI agent. Over 20 companies including Shopify, Walmart, Visa, and Mastercard have expressed support, and Cart API and Catalog API were added in March 2026, significantly expanding its scope.

Meanwhile, ACP (Agentic Commerce Protocol), driven by OpenAI and Stripe, initially aimed to complete payments entirely within ChatGPT. However, conversion rates at Walmart were only one-third of Walmart.com, and in March 2026 Instant Checkout was phased out. The focus shifted to discovery partnerships with Target, Sephora, Nordstrom, and others. A detailed comparison of both protocols is covered in a separate article.

Below that, the catalog / discovery layer provides the delivery infrastructure for getting product data to AI agents. OpenAI's ACP Feeds, Google Merchant Center, Shopify Catalog (Agentic Storefronts), and PayPal Store Sync operate in parallel. For e-commerce businesses, this represents a new channel strategy starting point: which AI platforms to deliver product data to.

The payment delegation layer features AP2 (Agent Payments Protocol), co-developed by Google with over 60 companies, which uses cryptographically signed digital contracts called "Mandates" to guarantee authorization for agent transactions. Stripe's SPT (Shared Payment Token) takes a simpler approach with single-use disposable tokens. In both cases, agents never touch raw card information.

In the agent-to-agent communication layer, Anthropic's MCP (Model Context Protocol) has rapidly gained traction as a standard for connecting to e-commerce systems. Shopify has shipped four MCP servers, and Stripe, commercetools, and PayPal have also published MCP servers. Google's A2A (Agent-to-Agent) handles agent coordination and has been transferred to the Linux Foundation.

At the bottom, the machine payment rails layer features two contrasting approaches. Coinbase's x402 revives the HTTP 402 status code for stablecoin-based settlement, and the x402 Foundation was established under the Linux Foundation in April 2026. MPP, co-developed by Stripe and Tempo, is a session-oriented streaming payment system, with 50+ services including OpenAI, Anthropic, and Google Gemini already implemented.

Payments & Trust Layer

You don't need to hand your credit card number to an AI agent. Agentic commerce payments are built on tokenization and cryptographic signatures that guarantee transaction legitimacy without a human pressing a button.

Regardless of which AI platform wins, payments flow through existing card networks. Leveraging this structural advantage, Visa and Mastercard in particular are leading the buildout of agent-compatible infrastructure.

Visa has achieved agent authentication through cryptographically signed HTTP messages via its Intelligent Commerce platform and Trusted Agent Protocol (TAP). Over 100 companies are participating globally, with 30+ developing in sandbox, and hundreds of agent transactions have been completed in production.

Mastercard provides tamper-proof cryptographic audit trails through Verifiable Intent and Agentic Token. In March 2026, it co-released an open-source cryptographic framework with Google and completed live transactions in Latin America. Its Selective Disclosure design shares only minimal information with each transaction participant. The strategic differences and commonalities between the two provide an important lens for understanding this market's structure.

Agent identity verification is also being standardized. As a counterpart to the financial industry's KYC (Know Your Customer), the KYA (Know Your Agent) framework is emerging — a mechanism for cryptographically verifying an agent's identity, scope of authority, and behavioral history.

The technical foundation is provided by Cloudflare's Web Bot Auth, which represents a shift from the "request-based" approach of robots.txt to a "proof-based" approach using cryptographic signatures, proposed as an IETF standard. Amazon Bedrock AgentCore, Mastercard Agent Pay, and American Express are progressing integration, and Visa TAP is also built on this foundation.

Data & Marketing

Agentic commerce changes the premises of marketing. Where traditional e-commerce optimization targeted "human eyes and judgment," agentic commerce requires optimization for "AI data parsing."

The foundation of all strategies is making product data Agent-Ready. Schema.org-compliant JSON-LD, global identifiers like GTINs, semantic product attributes. For AI agents to accurately understand and recommend products, human-facing marketing copy alone isn't enough. The architectural evolution from Headless Commerce → Composable Commerce → Agentic Commerce aligns with this same "data-first" trajectory.

If SEO is the discipline of "being found by search engines," AEO (AI Engine Optimization) is the discipline of "being chosen by AI engines". Optimizing the data sources that LLMs reference when recommending products and registering product feeds with AI platforms determine visibility on the new "AI surface." ChatGPT referral conversion rates have reached 7% on transactional sites, exceeding Google's 5%.

"Share of Model" is a new metric measuring brand presence within AI models. When someone asks ChatGPT, "What running shoes do you recommend?", what percentage of the time is your brand recommended? BrightEdge's research reveals that brand recommendation rates vary significantly across AI platforms for the same query.

Regulation & Governance

Existing contract law was designed on the premise that "a human presses the button." The legal frameworks haven't caught up with a world where AI agents autonomously execute contracts.

Legal liability for agent transactions is being debated across the industry, including the legal validity of contracts executed by agents, liability for errors, and the demarcation of responsibility between platforms and merchants. The EU Product Liability Directive (effective December 2026) includes software and AI systems as "products" subject to strict liability, with autonomous machine learning and post-sale changes via OTA updates potentially qualifying as defects.

In the U.S., NIST announced the AI Agent Standards Initiative in February 2026, with three pillars: development of international standards, joint investment in open-source protocols, and foundational research on agent security. An April 2026 deadline was set for concept papers on agent identity and authorization, and public-private dialogue is ongoing.

Visa TAP, Mastercard Verifiable Intent, and AP2's Mandate structure are all attempts to cryptographically bridge this gap between law and technology. Legal frameworks will take time to mature, but the market is moving forward in the meantime.

Market Size and Forecasts

Market size forecasts for agentic commerce range from $144 billion to $9 trillion across research firms. This gap isn't a matter of forecasting accuracy — it stems from differences in how "agentic" is defined.

eMarketer counts only sales occurring on third-party AI platforms like ChatGPT, Google AI Mode, and Perplexity, projecting $144 billion by 2029. Retailer-native AI tools like Amazon Rufus are excluded. On the other end, ARK Invest scopes all global online spending facilitated by AI agents, estimating $9 trillion.

Narrowing to the U.S. B2C market, major forecasts converge. Morgan Stanley projects $190–385 billion (10–20% of e-commerce sales), Bain projects $300–500 billion (15–25%). Taking the midpoint yields roughly 15–17% of U.S. e-commerce sales by 2030 flowing through agents.

Often overlooked is the B2B scale. Gartner predicts that by 2028, 90% of B2B purchasing will be mediated by AI agents, with over $15 trillion in B2B spending flowing through agent channels. Looking only at consumer-facing numbers significantly underestimates the total market. A detailed comparative analysis is available in our market size forecast article.

Industry Use Cases

The pace of agentic commerce adoption varies dramatically by industry. Industries with high repeat purchase rates and product standardization show the strongest affinity with AI agents.

Grocery is the industry that will change fastest. The combination of "repeat purchases, low involvement, and perishability" is a perfect match for AI-driven predictive replenishment. Instacart has integrated AI into smart carts, and Walmart's agent has contributed to $55 million in annual food waste reduction. Demand forecasting accuracy is reported to have improved from 55–65% to 85–92%.

In terms of scale, B2B procurement is the "hidden elephant". Routine, repetitive transactions — regular parts ordering, raw material procurement, SaaS license renewals — are the areas most naturally suited to AI agent automation. McKinsey's analysis projects procurement cost reductions of 25–40%, and platforms like SAP Ariba, Coupa, and Pactum are enabling simultaneous multi-site supplier negotiations powered by AI. As Gartner's B2B forecast (over $15 trillion) indicates, this could dwarf the consumer market in scale.

Consumer-facing shopping-focused AI assistants like Phia ($185M valuation) are also emerging, with differentiation along axes of "search-first vs. purchase-first" and "general-purpose vs. specialized."

Preparing for Agentic Commerce

NRF's survey shows that 68% of retailers plan to adopt agentic commerce within 12–24 months. Yet an SAP CX survey puts readiness at just "3–4 out of 10" — a significant gap between aspiration and reality.

The most practical first step is confirming whether your payment service provider (PSP) has an agentic commerce roadmap. Major PSPs like Shopify, Stripe, PayPal, and Adyen are already advancing protocol support, and as PSP-side readiness matures, the integration burden on merchants will be significantly reduced.

Conclusion

Agentic commerce is advancing simultaneously across seven domains: protocol standardization, platform competition, payment infrastructure buildout, data strategy transformation, regulatory framework development, market expansion, and industry-specific adoption. Every research firm agrees on the direction: AI agents will embed themselves at the center of the purchasing process, and companies that prepare early will be the first to benefit from this new channel.

Start with a simple question: "Is our product data in a state that's easy for AI agents to read?" The answer to that question is the starting point for your competitiveness in the agentic commerce era.