Commerce as the Transaction Layer for the AI Economy
AI agents can recommend products, compare prices, and answer buying questions. But they cannot manage catalogs, orchestrate payments, calculate taxes across jurisdictions, or fulfill orders. Commerce infrastructure is the transaction layer the AI economy needs. Spree delivers it: open-source, API-first, and agent-ready.
Key Takeaways
The gap: AI agents are transforming product discovery, but every transaction still needs commerce infrastructure: catalog management, pricing, order orchestration, payments, tax compliance, identity, APIs, SDKs, and agent readiness.
The proof: OpenAI tried building this into ChatGPT with $13 billion in funding, partnerships with Shopify and Stripe, and the most capable AI model in the world. They pulled the plug after six months.
The pattern: Six competing protocols (ACP, UCP, MCP, TAP, A2A, AP2) all assume a commerce platform exists underneath. The protocols define how agents talk to commerce. They do not replace it.
The solution: Spree provides that layer: open-source, API-first, and purpose-built for the agentic commerce era.
Last verified: March 2026.
Every AI application will eventually need to transact
The United States economy is 68% consumer spending. That’s roughly $20 trillion in annual personal consumption, according to the Bureau of Economic Analysis (Q4 2025). US retail ecommerce alone hit $1.47 trillion in 2025, with the Census Bureau reporting that digital channels now represent 16.6% of all retail sales and growing.
Here’s what those numbers mean for anyone building on AI: every AI application that touches a consumer workflow will eventually need to handle a transaction. Subscriptions, product purchases, service bookings, marketplace commissions. The question is not whether AI needs commerce. The question is who provides the infrastructure.
Right now, the answer is playing out in real time. OpenAI, Google, Shopify, Stripe, Visa, and Anthropic are all racing to define how AI agents transact. Six competing protocols launched in the past twelve months. The AI economy’s transaction layer is being built right now, and the platforms that provide it will shape a market McKinsey projects at $3 to $5 trillion in global agentic commerce by 2030.
That’s not a niche. That’s the foundation of the next economy.
Why couldn’t the smartest AI company build a checkout?
OpenAI had everything: $13 billion in funding, a partnership with Shopify, Stripe as a co-developer, and the most capable AI model in the world. In September 2025, they launched Instant Checkout inside ChatGPT. Users could ask for a product recommendation and buy it without leaving the conversation.
Six months later, they pulled the plug.
Fewer than a dozen Shopify merchants ever went live, out of millions eligible. Walmart reported conversion rates 3x lower for in-chat purchases compared to clicking through to Walmart.com. Forrester principal analyst Emily Pfeiffer said she was “shocked at the promises versus reality.” On March 4, 2026, OpenAI removed Instant Checkout entirely and pivoted to a discovery model that sends users back to merchant websites.
OpenAI’s failure wasn’t a technology problem. It was an infrastructure problem. They had no sales tax engine covering thousands of US jurisdictions. No fraud detection system. No real-time inventory synchronization across retailers. No returns management. No multi-vendor order splitting. These are capabilities that dedicated commerce platforms spent decades building. You don’t replicate them by adding a buy button to a chat window.
And OpenAI isn’t the first. Over the past decade, Meta, Google, Twitter, Pinterest, Snapchat, TikTok, and publishers all tried to own the checkout. Every single one retreated. The pattern is consistent, and the lesson is always the same: commerce is infrastructure, not a feature.
Nine capabilities that separate commerce infrastructure from features
Before Stripe, payments were a feature. You added a PayPal button and moved on. Stripe proved that payments are infrastructure by showing the depth hidden behind a simple API call: multi-currency support across dozens of countries, fraud detection, subscription management, marketplace payouts, tax calculation, and regulatory compliance per jurisdiction.
Commerce is at the same inflection point payments hit in 2010. Companies are discovering that “add a buy button” is to commerce what “integrate a payment gateway” was to payments: a surface-level solution that breaks under real-world complexity.
The infrastructure layer for commerce includes nine interlocking capabilities. AI can replicate any individual storefront feature. It cannot replicate the integrated stack.
| # | Capability | Why it’s infrastructure, not a feature |
|---|---|---|
| 1 | Catalog management | Products have variants, attributes, taxonomies, inventory levels, supplier relationships, and digital/physical distinctions. A social platform’s content model has zero overlap. |
| 2 | Multi-dimensional pricing | Multi-currency, customer-specific pricing, volume tiers, promotional rules, price lists with market rules. Not “one number per product.” |
| 3 | Order orchestration | Orders split across vendors, route to fulfillment centers, track through shipping, and manage returns. The transaction is the beginning, not the end. |
| 4 | Payment processing | Multi-provider support, PCI compliance, fraud detection, refunds, chargebacks, marketplace payouts. Stripe handles the payment tracks. Someone still orchestrates which provider, which currency, which rules. |
| 5 | Tax and regulatory compliance | Multi-jurisdiction tax calculation, VAT handling, industry-specific regulations (HIPAA, ITAR, age verification). Unglamorous but deal-breaking. |
| 6 | Identity and permissions | B2B buyer organizations, approval workflows, budget limits, role-based access, guest checkout, account management. Neither a social platform’s user model nor an AI’s session context can cover this. |
| 7 | API layer | Complete, versioned, well-documented APIs for every commerce operation. REST with OpenAPI specs means any system (or agent) can discover and call endpoints programmatically. |
| 8 | SDK and developer experience | Client libraries, documentation, local dev setup, testing tools. The on-ramp that determines whether developers actually build on your platform. |
| 9 | Agent readiness | Machine-readable documentation, context bundles, tool-use patterns, protocol support. The on-ramp for AI agents, not just human developers. |
OpenAI’s Instant Checkout attempted capability 4 (payments via Stripe) and a thin slice of capability 1 (product data via Shopify). That left seven infrastructure layers untouched. No wonder it failed.
What do AI agents actually need from commerce infrastructure?
When an AI agent completes a purchase on behalf of a user, it needs something fundamentally different from what a human shopper needs. Humans browse visual storefronts. Agents reason over structured data.
Machine-readable product data comes first. Agents need structured catalogs (title, variants, pricing, availability) exposed through clean APIs. Not a rendered webpage. Not a product image carousel. An OpenAPI specification they can auto-discover and call.
Programmatic checkout is non-negotiable. The entire purchase flow, from adding items to a cart through applying discounts, setting shipping options, and processing payment, must be executable via API. No form filling. No human-in-the-loop clicks.
Real-time inventory and pricing matters because agents transact at machine speed. Stale catalog data from an overnight sync means failed transactions, customer frustration, and lost trust.
Delegated authentication is emerging as a core requirement. Agents act on behalf of users. The commerce system needs to verify that this specific agent is authorized to transact for this specific buyer. Visa’s new Trusted Agent Protocol (TAP), now in early pilots, provides cryptographic proof of exactly this.
Context about the commerce system itself is what separates platforms agents can use from platforms they fumble with. Agents need to understand what the platform does, what endpoints exist, and what business rules apply. This is where AGENTS.md files and thorough documentation become critical differentiators.
Post-purchase visibility closes the loop. Agents need webhooks and event-driven APIs to track orders, handle returns, and communicate status back to users.
Every one of these requirements points to the same conclusion: agents don’t replace commerce platforms. They depend on them.
Why do new commerce protocols prove the infrastructure thesis?
Six protocols have emerged in the past year to define how AI agents transact. Look at what they assume.
| Protocol | Backers | What it standardizes |
|---|---|---|
| ACP (Agentic Commerce Protocol) | OpenAI, Stripe | Checkout transactions: cart, payment, fulfillment events |
| UCP (Universal Commerce Protocol) | Google, Shopify | Full shopping journey: discovery, cart, checkout, post-purchase |
| AP2 (Agent Payments Protocol) | Payment authorization for agent-initiated transactions | |
| MCP (Model Context Protocol) | Anthropic (donated to Agentic AI Foundation) | Foundational agent infrastructure: context, tools, integrations |
| A2A (Agent-to-Agent) | Multi-agent coordination for complex tasks | |
| TAP (Trusted Agent Protocol) | Visa, Cloudflare | Agent identity verification and authorization |
Notice what every protocol has in common: they all work over HTTP and REST. They all assume a commerce platform exists on the other end with products to browse, carts to fill, and orders to process. The protocols define how agents talk to commerce infrastructure. They do not replace it.
This is the Stripe parallel made real. Stripe defined how applications talk to payment infrastructure. ACP, UCP, and MCP are defining how agents talk to commerce infrastructure. In both cases, the protocol layer creates massive value, but only because the infrastructure beneath it already exists.
Platforms with complete REST APIs and machine-readable documentation are natively compatible with every protocol in this table. Platforms that rely on GraphQL-only interfaces, proprietary SDKs, or admin-panel-only workflows are architecturally excluded from the agentic commerce wave.
Three forces reshaping who wins in commerce
The convergence happening right now is not incremental. Three structural forces are colliding simultaneously, and they favor a very specific type of platform.
SaaS feature moats are eroding. AI replicates software features faster than SaaS companies can ship them. The companies holding value are those embedded in slow-moving digital transformations (government, enterprise) where switching costs are the moat, not feature lists. Generic SaaS without a legitimate infrastructure story faces the strongest erosion.
The volume shift is already measurable. Shopify reported a 15x increase in AI-driven shopping orders between January 2025 and January 2026. Visa measured a 4,700% surge in AI shopping traffic to US retail sites. The question is whether your platform participates or gets bypassed.
Open source is moving from liability to asset. When AI erodes feature moats, the open-source model becomes more attractive. You cannot replicate an open-source codebase’s transparency, auditability, and extensibility by bolting an AI chatbot onto proprietary software. BSD 3-Clause licensing means full code ownership, no vendor lock-in, and the freedom to deploy anywhere, including on infrastructure you control. In a market where SaaS licensing premiums are compressing, owning the code is a strategic advantage that compounds.
Developer and agent experience is the new moat. Features get replicated. Infrastructure gets adopted. The platforms that win the AI era will be the ones developers and agents reach for first when they need commerce. Documentation quality, SDK completeness, API design, and agent-readiness (AGENTS.md, MCP servers, OpenAPI specs) determine adoption.
eMarketer projects AI platforms will account for $20.9 billion in US retail spending in 2026, nearly quadrupling 2025 figures. The platforms with the strongest developer and agent experience will capture a disproportionate share.
These three forces point in the same direction: the winning position is open-source commerce infrastructure with exceptional API and agent experience. Not the prettiest storefront. Not the most features. The deepest infrastructure with the cleanest interface.
What does this mean for your platform choice?
If your commerce platform was chosen for its storefront templates, its app marketplace, or its ease of setup for a simple online store, it was probably the right choice three years ago. The question is whether it’s the right foundation for the next ten years.
Spree was built as a commerce engine with complete REST APIs: API-first, headless, open-source under BSD 3-Clause. Every operation available in the admin panel is also available through the API. The REST API v3 (Store + Admin) ships with an OpenAPI 3.0 specification, which means any agent framework can auto-generate client code and start transacting programmatically.
Spree covers all nine infrastructure capabilities natively: multi-vendor marketplace with vendor dashboards and commission management, B2B ecommerce with buyer organizations and approval workflows, multi-tenant architecture for franchise networks and white-label SaaS, multi-dimensional pricing with customer groups and market rules configurable from the admin panel (no code required), and payment orchestration across 60+ gateways with no provider lock-in.
For the agent-readiness layer specifically, Spree ships AGENTS.md (machine-readable context for AI agents), an MCP server, documentation bundles for local AI context, and the OpenAPI 3.0 spec that makes every commerce operation discoverable by agent frameworks.
The AI economy needs a transaction layer. The protocols are forming. The infrastructure platforms that developers and agents adopt today will power the next decade of commerce. For teams evaluating their commerce architecture, an open-source, API-first platform with native agent readiness provides the strongest foundation for the decade ahead. Own the layer that matters.
Get started with Spree and explore how an open-source commerce engine positions your business for the AI economy. You can also book a consultation about your specific architecture requirements.
Frequently Asked Questions
What is the “transaction layer” for the AI economy?
The transaction layer is the commerce infrastructure that AI agents use to complete purchases on behalf of users. Just as Stripe became the payment infrastructure for the internet economy, commerce platforms are becoming the transaction infrastructure for the AI economy. AI agents can discover and recommend products, but they need dedicated systems for catalog management, pricing, order orchestration, payment processing, tax compliance, and fulfillment. The transaction layer provides all of these as a unified, API-accessible service.
Why did OpenAI’s Instant Checkout fail?
OpenAI pulled Instant Checkout from ChatGPT on March 4, 2026, after fewer than a dozen Shopify merchants went live. Walmart reported 3x lower conversion rates for in-chat purchases compared to its own website. The core issue was infrastructure: OpenAI had no sales tax engine, no fraud detection, no real-time inventory sync, and no returns management. Dedicated commerce platforms built these capabilities over decades. Even $13 billion in funding and partnerships with Shopify and Stripe could not replicate them in six months.
What are agentic commerce protocols like ACP and UCP?
ACP (Agentic Commerce Protocol, from OpenAI and Stripe) and UCP (Universal Commerce Protocol, from Google and Shopify) are emerging standards that define how AI agents interact with commerce systems. ACP focuses on checkout transactions (cart, payment, fulfillment events). UCP covers the full shopping journey from discovery through post-purchase. Both work over HTTP/REST and assume a commerce platform exists on the other end. They standardize the communication, not the infrastructure.
How does Spree support AI agent commerce?
Spree provides the commerce infrastructure layer that AI agents need to transact. The REST API v3 covers every commerce operation (products, carts, checkout, orders, payments) with an OpenAPI 3.0 specification that agent frameworks can auto-discover. Spree also ships AGENTS.md for machine-readable platform context, an MCP server for local AI integration, and documentation bundles that agents can install locally. These capabilities make Spree natively compatible with emerging protocols like ACP and UCP.
Why does open source matter for AI commerce infrastructure?
AI is eroding SaaS feature moats faster than companies can build them. When features can be replicated by AI, the value shifts to infrastructure depth, transparency, and extensibility. Open-source commerce under BSD 3-Clause licensing gives teams full code ownership, the ability to audit and extend every component, and freedom from vendor lock-in or platform fees. In the AI era, controlling your commerce infrastructure is a strategic advantage, not a concession.
What makes commerce infrastructure different from a storefront platform?
A storefront platform provides templates, themes, and checkout pages for human shoppers. Commerce infrastructure provides the full operational stack: catalog management, multi-dimensional pricing, order orchestration, payment processing, tax compliance, identity and permissions, APIs, SDKs, and agent readiness. Storefront features are the visible surface. Infrastructure is the operational depth beneath it. AI agents need infrastructure, not storefronts.
Which commerce platforms are ready for the agentic commerce era?
Platforms with complete REST APIs, OpenAPI specifications, and machine-readable documentation are architecturally ready for agent commerce. Platforms that rely on GraphQL-only interfaces (which require agents to construct queries rather than call discoverable endpoints) or admin-panel-only workflows (which require human interaction) are structurally disadvantaged. Agent readiness is becoming a meaningful differentiator as AI commerce volumes grow.
