Can AI Agents Buy Things Autonomously? Inside the Agentic Commerce Protocol

One in six Black Friday purchases in 2024 was made by an AI agent. Not assisted by AI — made by AI. That number is rising fast, and most payment stacks were not built for it.

The Agentic Commerce Protocol, or ACP, is the technical framework that makes autonomous AI purchasing possible. Understanding it is no longer optional for payment leaders building infrastructure for the next three years.

What Is the Agentic Commerce Protocol (ACP)?

The agentic commerce protocol ACP is a set of technical standards that allows AI agents to browse merchant catalogs, select products, and complete purchases on behalf of users — without requiring manual input at each step.

Think of it as the handshake layer between an AI assistant and a merchant's commerce stack. ACP defines how an agent authenticates, queries inventory, applies user preferences, initiates a payment, and receives confirmation. It turns a conversational AI into an active buyer, not just a recommender.

This matters because the existing checkout infrastructure was designed for humans. It assumes a person is reading a screen, clicking buttons, and typing card details. ACP redefines the transaction initiation point, moving it upstream into the AI layer entirely.

Why Is Agentic Commerce Happening Now?

Three forces converged at once. AI assistants crossed a capability threshold where they can reliably follow multi-step instructions. Consumer behavior shifted — 58% of consumers now use generative AI to discover products. And major platforms including ChatGPT, Google Gemini, Perplexity, and Microsoft Copilot began opening commerce surfaces to third-party merchants.

The result is that 20% of eCommerce tasks in 2025 are already agentic. Consumers are not waiting for merchants to be ready. They are asking their AI assistants to "order the usual" or "find the cheapest flight under $400 and book it." If the merchant's catalog and payment infrastructure cannot respond to that request, the sale goes elsewhere.

This is not a speculative trend. It is current revenue being captured or lost depending on whether a merchant's checkout is agent-accessible.

How Does the Agentic Commerce Protocol Work?

At a technical level, ACP governs a sequence of structured interactions between an AI agent and a commerce system. Each step maps to a familiar payment concept, but the actor is different.

Step 1: Catalog Discovery

The AI agent queries the merchant's catalog using structured data. This is analogous to a user searching a website, but the agent needs machine-readable product data, pricing, availability, and shipping options in a format it can parse and compare programmatically.

Merchants without well-structured product feeds are effectively invisible to agentic commerce flows. The agent cannot recommend or purchase what it cannot read.

Step 2: Intent Resolution and User Preferences

The agent matches catalog results to the user's stated or inferred preferences. This is where personalization enters the flow. An agent managing a user's grocery replenishment, for example, already knows their preferred brands, dietary restrictions, and budget limits.

For payment leaders, this step introduces a new requirement: the agent needs to know the user's stored payment methods, preferred currency, and any spending authorizations. Pre-authorized credentials and stored tokens become foundational, not optional.

Step 3: Autonomous Checkout Initiation

Under ACP, the agent initiates checkout programmatically. It calls the merchant's payment API directly. There is no human-facing UI in this flow. No form to fill. No button to click.

This is the step that breaks most traditional payment stacks. Legacy checkout flows were built assuming a browser, a session, and a human on the other end. Agentic flows require API-first payment acceptance that can handle non-human-initiated transactions at scale.

Step 4: Authorization and Confirmation

For low-value or pre-authorized transactions, the agent completes the purchase autonomously. For higher-value purchases, most current implementations route a confirmation request back to the user before charging. The user approves with a single interaction — a tap, a voice command, or a message reply.

This confirmation layer is also where fraud controls need to operate. The transaction has no traditional behavioral signals like browsing history or time-on-page. Fraud detection must work from different inputs: agent identity, authorization scope, transaction velocity, and device binding.

What Does ACP Change for Payment Infrastructure?

The agentic commerce protocol ACP does not just add a new channel. It restructures several foundational assumptions in payment operations.

Tokenization Becomes a Requirement

Agentic flows depend on stored, tokenized payment credentials. The agent cannot prompt a user to enter card details mid-conversation. Merchants without a robust tokenization layer cannot participate in autonomous commerce at all.

This makes network token portability critical. If a user's stored token is locked to one provider and that provider has an outage, the entire agentic flow breaks. Merchants need tokens that survive provider switches and route across their full multi-PSP stack.

Approval Rate Optimization Gets Harder

Human-initiated transactions carry rich behavioral context that issuers use to assess risk. Agentic transactions strip most of that context away. The result is a different risk profile that some issuers will decline more frequently until they calibrate their models to agent-initiated flows.

Smart routing becomes more important, not less. Merchants need routing logic that identifies which providers authorize agentic transaction patterns most reliably and adjusts in real time. A static routing setup built for human checkout will underperform on agentic volume.

Fraud Controls Must Adapt

Traditional fraud rules look for anomalies in human behavior. Agentic transactions are, by definition, non-human — which means they trigger false positives in legacy fraud engines that were not built for this pattern.

Payment operations teams need fraud infrastructure that can distinguish between a legitimate AI agent acting on pre-authorized instructions and an unauthorized bot attempting to exploit stored credentials. The signals are different and the response logic needs to match.

Reconciliation and Reporting Change

When an AI agent places 40 orders in a single session across multiple merchants on behalf of a single user, the transaction data looks nothing like a standard purchase. Reconciliation systems need to handle agent identifiers, authorization scope metadata, and multi-step confirmation records alongside standard payment data.

Most legacy reporting stacks are not built for this. Payment operations teams that rely on manual processes will find agentic volume extremely difficult to reconcile accurately.

How Is the ACP Standard Evolving?

The agentic commerce protocol ACP is still being formalized. Several overlapping efforts are shaping the standard, including Anthropic's Model Context Protocol (MCP), which provides a framework for how AI models interact with external systems including commerce APIs. Google, OpenAI, and independent consortia are contributing parallel specifications.

The practical reality for payment leaders is that a single universal standard does not yet exist. What does exist is a converging set of patterns: API-first checkout, structured catalog access, pre-authorized token flows, and agent identity verification. Merchants that build to these patterns now will be positioned to comply with formalized standards as they arrive, rather than retrofitting later.

The merchants who wait for a finished specification before acting are already behind. ACP-compatible commerce is live today on the major AI platforms, and consumer adoption is accelerating regardless of standardization timelines.

How Does Yuno Support Agentic Commerce?

Yuno's Agentic Commerce product makes merchant catalogs purchasable inside ChatGPT, Claude, Gemini, Perplexity, and Copilot through a single integration. Rather than requiring separate technical work for each AI platform, one integration activates the merchant across all major agentic shopping surfaces.

This matters for payment leaders because it collapses the infrastructure problem. Merchants do not need to rebuild their checkout stack or maintain separate API connections for each AI assistant. The integration sits on top of the existing payment infrastructure and handles the translation layer that the agentic commerce protocol ACP requires.

Merchants using agentic flows see 2 to 6 times higher conversion compared to standard eCommerce flows. That gap reflects the friction reduction when a user can complete a purchase inside the AI assistant they are already using, rather than being redirected to a separate website and checkout experience.

The underlying payment infrastructure still matters. Agentic commerce creates new volume, but approval rates, fraud rates, and reconciliation accuracy on that volume depend on the quality of the payment stack beneath it. Yuno connects 1,000+ payment methods across 200+ countries, which means the routing, tokenization, and fraud controls needed to perform on agentic transactions are already in place.

The Revenue Case for Acting Now

The numbers make the urgency concrete. If 20% of eCommerce tasks are already agentic and a merchant's catalog is not agent-accessible, that is not a future risk — it is current revenue going to competitors whose infrastructure is ready.

The conversion premium is equally significant. A 2 to 6 times improvement over standard checkout conversion is not a marginal optimization. It reflects a fundamentally different purchase experience that removes the friction points where traditional eCommerce loses customers: redirects, form fields, loading screens, and checkout abandonment.

Payment leaders who frame agentic commerce as a technology project are likely to deprioritize it. The more accurate frame is revenue capture. The question is not whether AI agents will buy on behalf of your customers — they already are. The question is whether your infrastructure is positioned to accept those transactions reliably and profitably.

Where to Start: A Practical Checklist for Payment Leaders

Getting payment infrastructure ready for agentic commerce does not require rebuilding everything at once. These are the highest-priority areas to address.

• Audit your tokenization coverage. Identify which markets and payment methods in your stack rely on stored tokens and whether those tokens are portable across providers. Agent-initiated flows require pre-authorized credentials that survive PSP switches.
• Review your API-first checkout readiness. Agentic flows call payment APIs directly. If your checkout requires a browser session or has dependencies on human UI interaction, those are blockers to address before enabling agent access.
• Assess your fraud rules for non-human patterns. Run your current fraud logic against a sample of transactions with no behavioral signals and see where the false positive rate lands. Adjust rule thresholds before agentic volume arrives at scale.
• Check your catalog structure. Agent discovery depends on machine-readable product data. Confirm your catalog exposes pricing, availability, and product attributes in a structured format that AI platforms can parse.
• Map your top three markets for agentic readiness. Start with the markets where your current approval rates are strongest and consumer AI assistant adoption is high. These are the fastest paths to agentic revenue with the least payment infrastructure risk.

The Bottom Line on Agentic Commerce Protocol ACP

The agentic commerce protocol ACP is the technical foundation for a purchasing behavior that is already underway. AI agents are completing real transactions today, and the infrastructure gap between merchants who are ready and merchants who are not is widening with every month.

For payment leaders, the work is not abstract. It is tokenization, routing, fraud controls, and catalog structure — familiar problems with a new urgency. The merchants who solve them now will capture a conversion premium that compounds as agentic volume grows. The merchants who wait will spend the next two years in retrofit mode, trying to recover ground on a channel they could have owned from the start.

Agentic commerce is not coming. It is here. The only question is whether your payment infrastructure is ready to accept it.