The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants

Source: Katharina Schumacher and Roger Roberts, with Katharina Giebel; McKinsey & Company (QuantumBlack) · 2025-10-17

Agentic commerce—shopping powered by AI agents acting on our behalf—represents a seismic shift in the marketplace. It moves us toward a world in which AI anticipates consumer needs, navigates shopping options, negotiates deals, and executes transactions, all in alignment with human intent yet acting independently via multistep chains of actions enabled by reasoning models.

This isn’t just an evolution of e-commerce. It’s a rethinking of shopping itself in which the boundaries between platforms, services, and experiences give way to an integrated intent-driven flow, through highly personalized consumer journeys that deliver a fast, frictionless outcome.

The stakes are high: By 2030, the US B2C retail market alone could see up to $1 trillion in orchestrated revenue from agentic commerce, with global projections reaching as high as $3 trillion to $5 trillion, according to McKinsey research.

From a technical standpoint, that means mastering and deploying emerging integration enablers like Anthropic's Model Context Protocol (MCP), Agent-to-Agent (A2A) Protocol, Agent Payments Protocol (AP2), and Agentic Commerce Protocol (ACP) that enable a new era of intelligent, autonomous agents. It also means fundamentally re-architecting approaches to identity management and loyalty. Smart organizations are already beginning to create new, agent-ready sites that provide both strong agentic and consumer experiences.

As we move toward this new era in digital commerce, agentic commerce is being sculpted by the hands of developers, who are creating and refining systems that seamlessly integrate APIs from multiple sources to deliver real-time information and dynamic customer experiences. The pace of technological advancement is staggering. The duration of tasks that large language models (LLMs) can reliably complete with at least a 50 percent success rate has been doubling every seven months since 2019, according to METR.8“Measuring AI ability to complete long tasks,” METR, March 19, 2025. In 2019, for example, leading models could only manage tasks requiring a few seconds of human effort. In 2025, Anthropic’s Claude 3.7 Sonnet boasts a “time horizon” of 59 minutes, capable of completing tasks that would take a skilled human nearly an hour—and recently, Claude 4.5 has extended this to more than 30 human hours. This marks an inflection point where AI may soon be able to tackle multiday or multiweek tasks, opening up new opportunities in task specialization and multiagent collaboration.

Model Context Protocol. MCP is an interoperability standard that allows AI agents and systems to share context, intent, and data about prior activities across models and tools. Unlike static prompts or isolated API calls, MCP enables persistent, structured communication, allowing agents to retain memory, reasoning, and objectives across environments. By standardizing how developers connect LLM-based applications to tools and function calls, MCP empowers agents to take meaningful actions across platforms, unlocking more coherent, autonomous, and context-aware behavior, and laying the foundation for truly agentic ecosystems (Exhibit 3).

Agent-to-Agent Protocol. A2A is a communication protocol and interaction model that empowers autonomous AI agents to coordinate, negotiate, and complete tasks directly with one another, minimizing the need for human intervention. Built for interoperability, A2A enables agents—regardless of vendor, architecture, or environment—to securely exchange capabilities, status, and context through standardized protocols like JSON-RPC and HTTP. This open framework supports long-running tasks, dynamic capability discovery, and multimodal collaboration, including text, audio, and visual outputs. By reducing integration overhead and facilitating cross-agent coordination, A2A establishes the foundation for scalable, multiagent ecosystems. This accelerates enterprise adoption of autonomous workflows and allows intelligent agents to operate efficiently in real-time, cross-platform marketplaces.

Agent Payments Protocol. Google’s AP2 is a groundbreaking open, payment-agnostic protocol designed to empower autonomous and semiautonomous AI agents to make verifiable purchases on behalf of users. By utilizing cryptographically signed mandates that link intent, cart, and payment across users, merchants, and payment networks, AP2 creates an audit trail, not subject to repudiation, ensuring transparency and accountability. This innovative protocol has the potential to revolutionize the way transactions are conducted by unlocking new revenue streams through automated execution of “standing intents,” reducing costs by minimizing fraud and chargebacks, and reshaping competition as agent routing algorithms optimize for cost, speed, and rewards.

Just as traditional online commerce relies on a wide ecosystem of players—e-commerce platforms, search engines, payment providers, review sites, logistics, and fraud prevention tools—agentic commerce will require a similarly diverse and interconnected system (Exhibit 4). At its core, the agentic ecosystem includes AI platforms, autonomous agents, infrastructure, payment systems, and workflow automation. But equally important are the adapters and enablers—such as traditional e-commerce platforms and fraud prevention solutions—that must evolve to support this new paradigm. The adoption and scaling of agentic commerce will depend on the speed with which these players make their systems agent compatible and widely adoptable by consumers, retailers, and brands.

As commerce transitions from direct user interactions to agent-initiated transactions, the existing payments infrastructure will encounter significant structural challenges. Traditionally, components such as gateways and fraud engines have been built around a human-in-the-loop model, in which identity, intent, and authorization are explicit and observable.