Unlock the future of autonomous commerce

Consumer behavior is undergoing a dramatic shift as generative AI reshapes how people search, discover, and shop. Shoppers are no longer relying solely on Google searches or browsing individual retailer sites; instead, they are turning to AI-driven assistants that can aggregate results across search engines, social platforms, and commerce sites into a single, intelligent interface.

In fact, research shows that 71% of consumers want generative AI integrated into their shopping interactions, and 68% want these systems to deliver aggregated “one-stop” purchase options[1].

This move from traditional search to agent-driven discovery is transforming the foundation of digital commerce. Consumers increasingly expect hyper-personalization, intelligent recommendations, and frictionless checkout experiences: capabilities that agentic systems are especially well positioned to deliver.

For merchants, this shift means that their ability to compete will depend on being “agent-ready”: ensuring product catalogs, pricing, and checkout flows are accessible to AI agents. Retailers that adapt will capture traffic from these emerging channels, while those who don’t risk becoming invisible in the new landscape of digital shopping.

This article builds on our earlier exploration of E-commerce AI, moving from today’s AI-augmented experiences to tomorrow’s agent-led commerce. In the sections ahead, we will outline the emerging infrastructure, protocols, and industry initiatives shaping this transformation, and discuss practical concepts for adopting agentic commerce, keeping you on the leading edge of this shift.

Agentic AI as the Foundation

The application layer of the AI stack enables the operation of intelligent systems. On top of this, agentic workflows create a self-directed, iterative problem-solving system that can plan, execute, and refine its own outputs. These workflows incorporate multiple stages of reflection, tool use, planning, and even multi-agent collaboration to improve performance and robustness. Agentic AI encompasses both the overall process (agentic workflows) and the specialized components (AI agents) that execute specific tasks, acting as an application-level orchestration layer that enables greater autonomy and functionality. Learn more about Agentic AI.

AI Agents in Commerce

AI agents represent a new class of digital assistants that can interpret user intent, invoke the right tools, and take actions on behalf of people or organizations. Unlike traditional chatbots or recommendation engines, agents don’t just suggest options; they can compare products, check availability, initiate checkout sessions, and even execute purchases within the guardrails set by the customer.

In commerce, this shifts the experience from search-and-click browsing to conversational, action-driven journeys where the agent handles much of the complexity. This emerging model, agentic commerce, allows autonomous AI agents to discover, evaluate, and complete transactions with minimal human intervention.

This shift to agent-led commerce presents a fundamental challenge: our digital payment infrastructure was designed for humans, not autonomous AI. Today’s systems rely on a human clicking ‘buy’ to establish intent and accountability. When an agent initiates a payment, it breaks this model, raising critical questions of trust: How can a merchant be sure an agent’s request is accurate and not a hallucination? How can we prove a user gave specific authorization for a purchase? This trust gap is the primary reason new, agent-ready protocols are not just an advantage, but a necessity. To make it work securely and at scale, new protocols are being developed to ensure agents can interact safely with merchants, payment providers, and financial networks.

How does it work?

Merchants make their products “agent-readable” by providing accurate, up-to-date information on titles, prices, availability, images, and policies. AI assistants use this foundation to compare options and answer customer questions, while interfacing directly with merchant APIs to retrieve complete totals, including real-time shipping and taxes; acting as a personalized interface that ensures data accuracy before presenting the final cost in chat.

When a purchase is confirmed, the assistant opens a checkout session with the merchant. The merchant validates the cart, checks inventory, applies business rules, and processes payment through their existing provider.

A key principle of secure agentic commerce is the separation of roles. The AI assistant acts as a ‘shopping agent,’ expertly discovering products and negotiating carts. However, for security, it never handles sensitive payment credentials directly. Instead, it coordinates with a separate, secure component, akin to a digital wallet or ‘credentials provider’, which manages the user’s payment methods. This ensures that while the agent is intelligent, the payment process remains locked down and secure, building trust for both consumers and merchants.

Importantly, merchants remain in full control: they own the customer relationship, handle payments and settlements, and manage post-purchase support. In short, Agentic Commerce allows shoppers to act through trusted assistants while merchants maintain their existing commerce systems. The result is faster discovery, fewer abandoned carts, and a smooth path from question to purchase without rebuilding infrastructure.

What Does It Mean to Be Agent-Ready?

Becoming “agent-ready” doesn’t require rebuilding your commerce stack. It means exposing the right hooks so AI agents can interact with your business seamlessly and securely:

• Structured Catalogs: Share up-to-date product data (titles, prices, inventory, images, policies) in formats agents can interpret.
• Checkout APIs: Provide session-based checkout endpoints so agents can confirm carts, calculate shipping and taxes, and present totals accurately. These can be published through a traditional REST API for compatibility with existing systems, through the Model Context Protocol (MCP) for direct model-to-tool integration, or extended to Agent-to-Agent (A2A) protocols as agent networks evolve.
• Agent-Ready Checkout: Configure your checkout for agentic commerce so that any compatible AI agent can securely initiate and complete a transaction on behalf of your customers.
• Delegated Payments: Support secure, one-time payment tokens (via your existing PSP) so agents can complete purchases without ever holding sensitive credentials.
• Merchant Control: Keep ownership of customer relationships, payment flows, and post-purchase support while allowing agents to drive discovery and transactions.
• Compliance & Trust: Adopt open standards like the Agentic Commerce Protocol (ACP) and Agent Payments Protocol (AP2) to ensure interoperability, fraud protection, and user confidence.

Underlying Infrastructure for AI Agents to Make Payments

For agents to transact safely on behalf of users, a new infrastructure is required: one that ensures trust, interoperability, and compliance. Just as important, these standards must support complex flows and flexible configurations for any commerce type, from physical and digital goods to subscriptions and even asynchronous purchases. To meet this need, two of the most visible open initiatives in this area are the Agentic Commerce Protocol (ACP) and the Agent Payments Protocol (AP2).

Agentic Commerce Protocol (ACP)

Co-developed by OpenAI and Stripe, the Agentic Commerce Protocol (ACP) is an open standard that defines how AI agents, people, and businesses can work together to complete purchases securely and efficiently [3].

ACP is designed to be powerful, flexible, and easy to adopt while protecting user data and allowing merchants to decide whether to accept an order. Crucially, merchants always remain the merchant of record, maintaining direct customer relationships and control over payments, refunds, and support.

An early public implementation of ACP is Instant Checkout in ChatGPT, which enables eligible purchases directly within the ChatGPT experience. OpenAI launched it in September 2025 for U.S. Etsy sellers, and current help documentation says it is available in the United States for eligible Etsy items and select Shopify merchants, including Glossier, SKIMS, and Spanx.

For products without Instant Checkout enabled, ChatGPT can still direct users to the merchant’s website to complete checkout. ACP is open-sourced under the Apache 2.0 license, helping merchants and developers begin integrating with the protocol.

Agent Payments Protocol (AP2)

The Agent Payments Protocol, or AP2, is an open protocol initiative introduced by Google Cloud with participation from leading financial and technology organizations. It is designed as a payment-agnostic framework for secure, interoperable agent-led payments across payment methods. Because its public specification is still in an early stage, AP2 is best understood as an emerging protocol initiative rather than a mature industry standard [2].

At its core, AP2 addresses the risk of agent error by shifting from ‘inferred action’ to ‘verifiable intent.’ Instead of simply trusting an agent’s interpretation of a user’s request, the protocol uses mandates: cryptographically signed digital authorizations. Think of a mandate not as an instruction, but as a tamper-evident, non-repudiable, cryptographically signed authorization. When a user approves a purchase, they create a verifiable audit trail that helps protect the user, the merchant, and other parties in the event of a dispute.

These mandates enable two key modes of payment:

• Real-time purchases, where the customer approves a transaction on the spot.
• Delegated purchases, where the customer sets predefined rules and the agent executes later.

AP2 is designed to work alongside emerging agent protocols. Its public documentation describes it as an extension for Agent2Agent (A2A), with MCP-related integration also part of the broader architecture. In adjacent open-source work, the A2A x402 extension shows how agent-to-agent crypto payment patterns are being explored, but it is better presented as complementary ecosystem experimentation than as widespread AP2 adoption.

Although still in its early stages, AP2 seeks to provide shared standards for consent, fraud prevention, and dispute resolution, enabling interoperability across cards, bank transfers, and emerging digital assets. This delegated mode is especially significant: it marks the point where agents can transact independently, executing purchases later based on pre-set mandates or even negotiating with other agents. In effect, AP2 points toward a future of more autonomous, agent-to-agent commerce beyond today’s primarily real-time, user-driven transactions.

The Future of E-Commerce in an Agentic World

ACP and AP2 go beyond checkout in chat interfaces and point toward a broader commerce model in which agents may become more deeply embedded in everyday purchasing workflows. Each participant in the ecosystem benefits: merchants reach high-intent buyers while continuing to run on their existing infrastructure, AI agents embed commerce directly into their applications without becoming merchants of record, and payment providers expand transaction volume by securely processing agentic payments.

On top of this foundation, commerce begins to extend far beyond typing into a chatbot:

• Conversational and Multimodal Shopping: Chat and voice assistants handling shopping tasks across mobile, web, and smart devices.
• In-App and Vertical Agents: AI agents in fitness, travel, or productivity apps recommending and transacting directly with merchants.
• Ambient and Voice Interfaces: Smart speakers, cars, and wearables enabling purchases through natural interactions.
• Delegated and Autonomous Transactions: Agents executing later, reordering supplies, booking tickets, or negotiating offers based on user mandates.
• Agent-to-Agent Commerce: Buyer agents and seller agents may increasingly transact directly, especially in B2B and supply chain contexts, with protocols such as AP2 aiming to support secure payment authorization in those workflows.

Together, these capabilities expand e-commerce into a truly agentic model, where transactions are contextual, continuous, and increasingly autonomous. Businesses that prepare for this evolution today will be positioned not just to adapt, but to lead in the next era of digital commerce.

Partner with Krasamo to Build AI Agents

Navigating this transition from today’s e-commerce to a fully agentic model requires both a strategic vision and deep technical expertise. As businesses prepare for this future, partnering with specialists who understand the AI stack is crucial for success.

Agentic commerce is reshaping how businesses connect with customers. At Krasamo, we help organizations design and implement the AI-driven systems that make this shift possible.

• Design AI workflows tailored to your business needs.
• Build and deploy AI agents that can transact, recommend, and automate securely.
• Integrate with existing applications and commerce platforms to extend functionality without disrupting operations.
• Enhance e-commerce and payment systems with agent-ready protocols and frictionless checkout experiences.
• Leverage our expertise in AI, IoT, and mobile development to create scalable, future-ready solutions.

Let’s explore how Krasamo can help you stay ahead in the age of autonomous commerce. Contact us today.

References:

[1] What matters to today’s consumer 2025 :consumer behavior tracker for the consumer product and retail industries

[2] Intro to Google Agent Payments Protocol (AP2)

[3] Agentic Commerce Protocol