Agentic Commerce Update: The Ecosystem Takes Shape

Back in October, I wrote about the emergence of two competing protocols for agentic commerce: Google’s Agent Payments Protocol (AP2) and OpenAI’s Agentic Commerce Protocol (ACP). I asked which would win. Three months later, we have an answer—and it’s more interesting than a simple winner-take-all outcome.

The short version: ACP has emerged as the dominant implementation standard for consumer commerce, but the ecosystem has expanded far beyond OpenAI. Microsoft, Stripe, Shopify, PayPal, and Salesforce have all built major implementations. Meanwhile, a new layer of protocol-agnostic middleware is emerging to serve merchants who don’t fit neatly into the platform ecosystem.

What Happened to AP2?

Let’s address the obvious question first. Google’s AP2 has impressive backing—60+ organizations including Adobe, Mastercard, American Express, PayPal, Coinbase, Salesforce, and ServiceNow. But as of today, consumers cannot use AP2-based payments in any publicly known, real-world product or service.

The protocol specification is published. Reference implementations exist. Partners are building. But actual consumer-facing checkout? Not yet. Cobo, a crypto custody provider, has announced plans for a live AP2 demo in February 2026. ServiceNow is integrating AP2 for enterprise procurement workflows. But no one is shipping consumer commerce on AP2 today.

This confirms the prediction I made in October: AP2 is “building the foundation” while ACP is “shipping product.” AP2 may well become the enterprise and B2B standard—its cryptographic mandates and audit trails are designed for exactly that use case. But ACP owns consumer commerce right now.

What Changed Since October

In my previous post, ACP had just launched with Etsy and was promising Shopify integration. I noted that “if you’re already using Stripe, you can enable agentic payments with as little as one line of code.”

That prediction played out, but the ecosystem evolved faster than expected:

Salesforce integrated both protocols with Agentforce Commerce. Enterprise merchants can now connect via ACP through Stripe and via Google’s AP2 for agent-led payments. This confirms my prediction that “the winning strategy might be supporting both protocols.”

Microsoft entered with Copilot Checkout in January 2026, adopting ACP and extending it across Copilot.com, Bing, MSN, and Edge. More significantly, they introduced automatic enrollment for Shopify merchants—no application, no technical integration required.

Stripe launched the Agentic Commerce Suite in December 2025. This is the “middle path” I anticipated—merchants connect their catalog to Stripe, select which AI agents to sell through, and Stripe handles the ACP infrastructure. Partners now include Wix, WooCommerce, BigCommerce, Squarespace, and commercetools.

Why Salesforce Is Uniquely Positioned

I want to spend more time on Salesforce because their approach illustrates something important about where enterprise agentic commerce is heading. While OpenAI, Microsoft, and Stripe are building point solutions—checkout here, payments there—Salesforce has spent two decades building something else entirely: a unified platform that connects sales, service, marketing, commerce, and now AI agents on a single data foundation.

Data Cloud is the foundation. This is Salesforce’s hyperscale data engine that aggregates and harmonizes data from both Salesforce applications and external sources. The platform processes over 2 quadrillion records per quarter and converts unstructured data—PDFs, chat transcripts, emails, voice recordings—into unified customer profiles. With zero-copy integrations to Snowflake and Databricks, enterprises can use their existing data lake investments without duplication. The result: AI agents grounded in a complete, real-time view of every customer interaction across every touchpoint.

Agentforce 360 is the orchestration layer. Announced in October 2025, this platform connects humans, agents, and data in one system. Unlike point solutions where each agent operates in isolation, Agentforce agents share context through Data Cloud. A commerce agent knows what the service agent discussed yesterday. A marketing agent knows what the sales agent promised last week. This isn’t just data sharing—it’s institutional memory for AI.

Multi-agent orchestration through MCP and A2A. Salesforce partnered with Google and 50+ technology vendors on the Agent2Agent (A2A) protocol, contributing the concept of an “Agent Card”—a lightweight JSON contract that communicates an agent’s capabilities, identity, compliance tags, and trust score. With native MCP support in Agentforce 3, agents can now connect to external tools and systems through MuleSoft while maintaining governance. An Agentforce commerce agent can delegate to a ServiceNow support agent or an SAP inventory agent through standardized protocols.

Industry-specific vertical solutions. Agentforce isn’t one generic agent—it’s specialized for healthcare, financial services, retail, manufacturing, and the public sector. Each vertical includes industry-specific data models, compliance frameworks, and pre-built agent skills. A healthcare commerce agent understands HIPAA. A financial services agent handles regulatory requirements. This vertical depth is difficult for horizontal platforms to replicate.

The early results are compelling. Reddit automated 46% of support cases and reduced response times by 84%. OpenTable now handles 70% of restaurant and diner inquiries without human intervention. Adecco reports 51% of candidate conversations occur outside working hours. Pandora achieved a 10% lift in net promoter score while AI-powered recommendations drive higher orders and average order value.

What makes Salesforce’s position particularly strong for commerce is that Agentforce Commerce is the only unified platform spanning digital commerce, point of sale, and order management—connected to sales, service, and marketing data.

When an AI shopping agent knows a customer’s purchase history, service tickets, marketing engagement, and loyalty status, it can do things no standalone commerce agent can: personalize not just recommendations but the entire relationship.

The Rise of Protocol-Agnostic Middleware

Perhaps the most interesting development is what’s happening below the platform layer. The market isn’t waiting for protocol convergence—it’s solving fragmentation from the bottom up.

AgentCart has emerged as a protocol compliance engine supporting ACP, AP2, and A2A standards simultaneously. Their pitch: “Connect once. Become compatible everywhere.” They support Shopify, WooCommerce, Magento, BigCommerce, Wix, and Squarespace—turning any online store into an agent-ready store without requiring merchants to choose a protocol.

Klarna launched the Agentic Product Protocol in December 2025—an open standard for product discovery designed to complement both ACP and AP2. It provides AI agents access to 100+ million products and 400 million prices across 12 markets. Klarna’s insight: before agents can buy, they need to know what exists. Their protocol handles discovery; ACP and AP2 handle checkout.

Magebit released an open-source ACP module for Magento 2 and Adobe Commerce, giving enterprise merchants on those platforms a path to ChatGPT Instant Checkout without replatforming.

This middleware layer is significant. It means merchants don’t have to bet on a single protocol or platform. The abstraction is moving up the stack.

The Three Paths to Participation

The most important development for merchants is how dramatically the onboarding experience has diversified. There are now three distinct models:

Automatic Enrollment: Shopify merchants are now automatically eligible for OpenAI’s Instant Checkout and automatically enrolled in Microsoft’s Copilot Checkout. No application, no technical work. Etsy sellers have the same status. This is the lowest-friction path and represents a major shift from the technical integration I described in October.

Managed Infrastructure (Stripe): Stripe’s Agentic Commerce Suite provides a single integration point that reaches multiple AI agents. Merchants connect their product catalog, select agents from the dashboard, and Stripe hosts the ACP endpoint. The suite is modular—choose discovery, checkout, or payments components as needed. This is the path for merchants who want reach without platform lock-in.

Direct Implementation: OpenAI’s original approach—building your own ACP endpoints, implementing REST APIs, setting up webhooks—remains available for merchants who want maximum control. This requires more effort than a plug & play solution but offers the most control and customization. Vendors like XCentium and point solutions like AgentCart and Magebit’s module make this path accessible to merchants on legacy platforms

What About Custom Storefronts and Microsites?

A question I’ve heard from several customers: what if we want to run specialized microsites alongside our main commerce operations? Think Addidas CONFIRMED—a marketing-specific experience separate from the main Addidas site. Or Nike SNKRS.

The answer is headless commerce architectures, which preserve flexibility while enabling agentic commerce through protocol middleware.

On-platform options: Salesforce Commerce Cloud offers headless Commerce APIs that enable custom storefronts — Salesforce is particularly strong in B2B settings where customer-specific pricing, contract terms, and multi-stakeholder approval workflows require deep CRM connectivity. This is an area that our B2B commerce team at XCentium concentrates on.

For consumer sites on Shopify there is the. Hydrogen/Oxygen framework that lets you build fully custom React storefronts while maintaining Shopify's backend and automatic ACP eligibility. BigCommerce offers similar headless APIs and is part of the Stripe Agentic Commerce Suite.

Enterprise options: commercetools provides MACH architecture (Microservices, API-first, Cloud-native, Headless) with native MCP support. For brands needing multiple storefronts from a single backend—different experiences for different markets, channels, or campaigns—this is purpose-built.

Custom builds: For brands like Nike with existing custom infrastructure, AgentCart’s protocol compliance engine can add ACP and AP2 support without replatforming. Nike runs a React SPA with Node.js backend powering SNKRS, Training Club, and Run Club. Connecting that to agent protocols doesn’t require rebuilding—it requires middleware.

The Legacy Merchant Opportunity

Here’s something the platform-focused coverage misses: a significant portion of ecommerce is NOT on Shopify, Stripe, or any of the automatically-enrolled platforms.

A substantial segment, roughly 20% of all implementations worldwide do not run on the popular platforms. Most of these implementations are in the enterprise segment because of the need for greater control.

B2B ecommerce migration spend is projected at $4.2 billion in 2025—indicating a large base of merchants on legacy systems. Custom-built solutions rank as the second-largest segment globally in some analyses. And 67% of B2B buyers report switching suppliers specifically to find a “more modern, consumer-like experience.”

These merchants—enterprise brands on Oracle Commerce, B2B companies with customer-specific pricing and multi-stakeholder approval workflows, anyone with complex checkout requirements—are exactly who OpenAI’s original “build your own endpoints” approach was designed for.

XCentium provides merchants a viable path. These enterprises don’t need to replatform to Shopify or rebuild on Stripe. They can connect their existing infrastructure to agent protocols with our assistance.

Early Results

The conversion data is compelling. Microsoft reports that shopping journeys with Copilot led to 53% more purchases within 30 minutes. When shopping intent is present, Copilot journeys are 194% more likely to result in a purchase. Salesforce reports AI assistants are growing retail traffic volumes by 119%.

Microsoft’s Brand Agents—AI shopping assistants merchants can deploy on their own websites via Microsoft Clarity—have shown a 3x higher conversion rate in assisted sessions. This addresses a concern I raised in October: merchants maintaining control of their customer relationships. Brand Agents let you deploy agentic commerce on your own properties, in your own voice.

Key Takeaways

Merchants retain control. Across all implementations, the merchant stays the merchant of record—you own the transaction, customer data, and relationship. AI agents facilitate transactions; they don’t intermediate them.

PayPal and Stripe own the payment layer. PayPal powers payments for both OpenAI’s Instant Checkout and Microsoft’s Copilot Checkout, leveraging its 400 million active accounts. Stripe’s Shared Payment Tokens handle secure credential transfer between agents and merchants.

Protocol fragmentation is being solved by middleware. AgentCart, Klarna’s Agentic Product Protocol, and similar solutions mean merchants don’t have to bet on ACP vs. AP2. The abstraction layer is emerging.

Legacy merchants have options. If you’re not on Shopify or Stripe’s ecosystem, you’re not locked out. Direct implementation paths and middleware solutions now exist for enterprise and custom platforms.

What This Means for You

If you're an enterprise on Salesforce, you're arguably best positioned. Agentforce Commerce gives you both ACP and AP2 support, autonomous merchandising, and intelligent order routing—all connected to the unified customer view that standalone commerce platforms can't match.

If you run specialized microsites alongside your main commerce operations, headless architectures plus protocol middleware (API endpoints) preserve your flexibility.

If you’re on Shopify or Etsy, you’re already in. Check your admin panel for Copilot Checkout settings.

If you use Stripe, the Agentic Commerce Suite is your fastest path to multi-agent reach without building custom infrastructure.

If you’re on Magento, Adobe Commerce, look at AgentCart or Magebit’s module. You don’t need to re-platform.

The strategic question has shifted from “which protocol should I support?” to “which onboarding model matches my technical capacity and business model?” The protocols are converging; the implementation paths have diverged.