← Blog · June 26, 2026 · Mohamed Salama
The Agentic Economy: Definitions, Protocols, and Market Solutions

The Agentic Economy: Definitions, Protocols, and Market Solutions

The internet is being re-architected around a new kind of economic actor. For three decades, online commerce assumed a human was clicking "buy" on a trusted website. That assumption is now breaking. In its place is the agentic economy — a landscape in which autonomous AI agents discover, negotiate, transact, and increasingly earn on behalf of people and businesses, often with no human in the loop at the moment of action.

This piece defines the term, maps the protocol stack that makes machine-to-machine commerce possible, and surveys the market solutions racing to own the new checkout.

What the agentic economy actually is

The agentic economy is the emerging paradigm in which autonomous AI agents participate directly in markets as independent economic actors — perceiving, reasoning, and acting on goals rather than executing pre-programmed scripts. Agents can represent a consumer, a business, or themselves, and they buy, sell, coordinate, and collaborate with other agents and humans.

The distinction that matters is autonomy, not automation. The OECD, in its conceptual mapping of the agentic AI landscape, and academic work on agentic AI in manufacturing converge on the idea of agenticness — the degree to which a system can adaptably achieve complex goals in dynamic environments with limited supervision, measured across goal complexity, environmental complexity, and adaptability.

In practical terms, a production-grade agent in 2026 is not a chatbot or copilot: it plans a task, calls tools to execute steps, holds context across those steps, and exercises judgment on when to act or escalate. If the underlying categories are unfamiliar, our guide to the different types of AI agents covers them in more detail. Mastercard's own framing is illustrative — an entrepreneur could instruct an agent to launch a flower shop's web presence, buying a domain, hosting, images, and checkout pages within a budget, turning one human request into a chain of transactions executed automatically across providers.

A note of caution belongs here. Adoption is real but uneven: while roughly two-thirds of organizations are experimenting with agents, fewer than one in four have scaled them to production, and even Stripe conceded in its 2026 annual letter that agentic commerce was "overhyped too early in some corners" while remaining potentially generational.

The protocol stack

The agentic economy runs on a layered stack of open protocols. By early 2026 the landscape matured from a cacophony of proposals into a clearer, if still fragmented, architecture. The layers are complementary, not competing.

Connectivity: agent-to-tool

Model Context Protocol (MCP), introduced by Anthropic, is the de facto standard for connecting agents to tools and data through authenticated, schema-based interfaces. It is now governed by the Linux Foundation's Agentic AI Foundation and, by Q1 2026, spanned over 18,000 community-indexed servers. MCP answers the question "how does an agent use a service?"

Coordination: agent-to-agent

Agent-to-Agent (A2A), originated by Google in April 2025 and donated to the Linux Foundation, is the horizontal bus for agents on different frameworks to discover each other and exchange tasks over JSON-RPC. At its one-year mark it reached v1.0 with signed Agent Cards, 150+ production organizations, and SDKs in five languages. Notably, IBM's competing Agent Communication Protocol (ACP) merged into A2A in August 2025, consolidating the coordination layer. MCP and A2A are widely treated as the two-layer default: MCP for vertical tool access, A2A for horizontal agent coordination.

Payments: the contested layer

This is where the action — and the fragmentation — concentrates.

Identity and trust

Agents cannot satisfy traditional KYC, manage chargebacks, or hold accounts the way humans do, so a parallel trust layer is emerging. On-chain, ERC-8004 — proposed by the Ethereum Foundation's dAI team with MetaMask, Google, and Coinbase — went live on Ethereum mainnet in January 2026, providing registries for agent identity, reputation, and verification. Off-chain, the card networks built their own: Visa's Trusted Agent Protocol and Agent Score, Mastercard's Verifiable Intent, and Cloudflare's Web Bot Auth (developed with Microsoft, Shopify, and others) for distinguishing legitimate agents from malicious bots.

Market solutions

The commercial race breaks roughly into three camps: the AI platforms that own the consumer surface, the card networks that own settlement and trust, and the crypto-native rails built for machine-speed micropayments.

AI platforms and merchant rails

OpenAI + Stripe launched Instant Checkout in ChatGPT in September 2025, powered by ACP and Stripe's Shared Payment Tokens — a primitive that lets a platform initiate payment without exposing the buyer's credentials. The consumer shape shifted quickly: Instant Checkout was retired in March 2026 in favor of retailer-operated "ChatGPT Apps" (Walmart, Etsy, Target, Instacart, Expedia), with ACP surviving as the underlying infrastructure. Stripe's broader Agentic Commerce Suite lets merchants broadcast catalog, pricing, and fulfillment into assistant surfaces, and its partner roster spans Microsoft Copilot, Anthropic, Perplexity, Vercel, and Replit.

Google + Shopify counter with UCP and the Gemini app's large consumer base. Shopify's Agentic Storefronts give merchants out-of-the-box access to ChatGPT, Copilot, Google AI Mode, and Gemini; Gap's CTO has publicly argued that UCP gives merchants more control over the experience than ACP.

The card networks

Visa built Visa Intelligent Commerce to embed credentials, controls, and authentication into agent-initiated buying, and shipped Intelligent Commerce Connect — a protocol-agnostic on-ramp accepting agents across Visa TAP, Mastercard MPP, ACP, and UCP simultaneously, positioning one network as the category's translator.

Mastercard moved from Agent Pay to Agent Pay for Machines (AP4M) in June 2026, purpose-built for high-frequency, low-value machine transactions — including microtransactions of fractions of a cent — settled across cards, bank accounts, and stablecoins, with 30+ launch supporters including Stripe, Coinbase, and Checkout.com.

Crypto-native rails

Coinbase's x402 anchors the stablecoin camp. The protocol crossed over 100 million cumulative transactions on Base through Q1 2026, and in April 2026 the x402 Foundation launched under the Linux Foundation with backing from Google, Microsoft, AWS, Visa, Mastercard, Stripe, Circle, and Shopify. Coinbase also shipped Agentic.market, a discovery storefront where agents find and pay for services with "zero API keys required." Amazon Bedrock AgentCore Payments brings x402 and Coinbase wallet infrastructure into enterprise workflows. Beyond payments, platforms like Virtuals Protocol function as a commerce layer, reporting 18,000+ deployed agents and agent-GDP exceeding $479 million.

A reality check on the data: independent analyses flag heavy "gamed" activity in on-chain agent metrics. Artemis found the ratio of real to gamed x402 transactions running close to 1:1 at points in early 2026, so headline volumes should be read with discounting. That said, Chainalysis notes genuine usage shifting toward larger transfers — over $1 — and wallets returning week-over-week absent any speculative catalyst, a signal of real utility.

Where this is heading

The forecasts are large and should be treated as directional rather than precise. Analysts project agentic commerce reaching roughly $1.5 trillion globally by 2030, and Gartner expects 40% of enterprise applications to include task-specific agents by the end of 2026. The next structural evolution is agents that earn rather than only spend — accepting payments, hiring other agents, and running paid services, a shift already visible in early agent-run businesses.

The open question is governance, not capability. As one analysis put it, without a verifiable proof chain — verified merchant, verified intent, verified fulfillment — autonomous purchasing is "just automation with plausible deniability." The protocols converging in 2026 are the industry's attempt to supply that chain before the volume arrives. If you're weighing where agents fit in your own product or operations, tell us about it, or see how we build.

Sources

Definitions and landscape

Protocols

Market solutions

Data and adoption