The Rise of Agentic Commerce: What the Next 2–3 Years Will Look Like

The internet is about to undergo its most profound transformation since the browser. Within five years, autonomous AI agents will outnumber humans online — scaling from today's 5.5 billion internet users to billions, potentially trillions, of agents negotiating, transacting, and collaborating at machine speed. The infrastructure we have built for a human-centric web is fundamentally unprepared for what is coming.

This is not speculation. The building blocks are being assembled right now, and the pace is accelerating faster than most people realise.

From Chat Assistants to Autonomous Agents

We are witnessing a decisive shift in how AI is deployed. Every major model release in 2025 and early 2026 has featured agent ability as a headline capability. Anthropic's Claude, OpenAI's Codex, and Moonshot's Kimi are all marketing autonomous, multi-step reasoning as their core differentiator. Where a simple chat query burns hundreds of tokens, an agent run with tool-calling and multi-step reasoning burns tens to hundreds of thousands. That is the demand multiplier the industry needs — and the models are finally reaching the point where they can deliver it.

But capability alone is not enough. For agents to operate autonomously at scale, they need infrastructure: discovery, identity, payments, and trust. This is where the most consequential work is happening right now.

Project NANDA: Beyond DNS for the Agent Internet

One of the most important papers published recently is Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts by Ramesh Raskar and colleagues at MIT. The paper makes a compelling case that the internet's current DNS-centred identity and discovery infrastructure simply cannot handle what is coming — billions of agents that spawn, negotiate, delegate, and migrate in milliseconds.

The NANDA (Networked AI Agents in a Decentralised Architecture) index proposes a lightweight, horizontally scalable foundation for agent discovery and authentication. At its core, it introduces AgentFacts — cryptographically verifiable capability assertions that allow agents to prove what they can do, who they represent, and whether they can be trusted, all without centralised gatekeepers.

The architecture delivers five concrete guarantees: a quilt-like index supporting both native and third-party agents, rapid global resolution for newly spawned agents, sub-second revocation and key rotation, schema-validated capability assertions, and privacy-preserving discovery across organisational boundaries. This is not an incremental improvement — it is a fundamentally new layer for the internet.

OpenClaw: The Napster Moment for Agentic Commerce

If NANDA represents the infrastructure layer, OpenClaw represents the consumer breakout. Originally a side project called Clawdbot, OpenClaw crossed 180,000 GitHub stars and drew 2 million visitors in a single week in late January 2026. It is an open-source agent framework that connects chat platforms — WhatsApp, Telegram, Discord, iMessage — to AI coding agents with full access to a computer's operating system: browser, terminal, files, and 50+ integrations through a skills extension system.

What makes OpenClaw architecturally significant is its heartbeat — a proactive loop where the agent wakes up on a set interval, scans its environment, checks for work, executes tasks, and goes back to sleep. This shifts agents from passive tools to active systems that behave autonomously. Combined with the framework's ability to deploy sub-agents and coordinate between them, it opens up more potential for agent-to-agent interaction than anything we have seen before.

Some commentators have called this the "Napster moment" for agentic commerce — the point at which the concept stops being theoretical and becomes viscerally real for millions of people.

The Infrastructure Stack: MCP, A2A, and the Emerging Protocols

For autonomous agents to transact reliably, we need standardised protocols. Several critical ones are converging simultaneously:

Model Context Protocol (MCP)

Anthropic's MCP has emerged as the universal standard for connecting AI models to tools, data, and applications. Adopted by OpenAI in March 2025 and integrated across Microsoft's Azure ecosystem, MCP provides the standardised interface through which agents access external capabilities. But here is the critical point: we need MCP servers with verified tools. As the agent population scales, the ability to cryptographically verify that a tool does what it claims — and nothing more — becomes a matter of systemic trust, not just convenience.

Agent-to-Agent Protocol (A2A)

Google's A2A protocol, now donated to the Linux Foundation, provides the definitive common language for agent interoperability. It enables agents built on diverse frameworks by different vendors to communicate and collaborate effectively — as agents, not just as tools. The protocol supports capability discovery, task delegation, and secure message exchange between agents that have never interacted before.

Agentic Commerce Protocol (ACP)

OpenAI and Stripe co-developed the Agentic Commerce Protocol (ACP) — an open standard (Apache 2.0 licensed) that enables programmatic commerce flows between AI agents and businesses. ACP was born from a simple insight: customers should be able to securely buy where they discover, businesses should be able to sell through new channels without giving up trust or control, and AI agents should be able to enable transactions without exposing customer credentials.

ACP defines a clear transaction flow: the buyer discovers a product via an AI surface, the agent interfaces with the business to initiate checkout, and the business retains full control as the merchant of record — choosing to accept or decline using its existing payment and fraud signals. Payment credentials are relayed via secure tokens that are programmatically controlled, permissioned, and logged.

What makes ACP particularly powerful is its pragmatism. Businesses build once and can distribute to any ACP-compatible AI agent — no custom integrations for each agent. It supports physical and digital goods, subscriptions, asynchronous purchases, and complex flows like multi-merchant carts. Stripe has already deployed ACP to power Instant Checkout in ChatGPT, enabling US users to buy from Etsy sellers and over a million Shopify merchants directly in conversation. Salesforce has also announced support for ACP, signalling rapid enterprise adoption.

x402: The Crypto-Native Payment Layer

x402 is an open standard for internet-native payments built on the long-reserved HTTP 402 status code. Developed by Coinbase in collaboration with Cloudflare, x402 enables any API or web service to require payment before serving content. For agents, this means paying per-request from a single wallet, switching between providers autonomously based on cost and capability, and accessing economic tools without human approval at every step.

Google's Agent Payments Protocol (AP2)

Perhaps the most ambitious protocol to emerge is Google's Agent Payments Protocol (AP2) — an open, payment-agnostic protocol built on top of A2A that establishes a common language for secure, compliant transactions between agents and merchants. AP2 addresses three fundamental questions that today's payment systems were never designed to handle: authorisation (proving a user gave an agent specific authority to purchase), authenticity (ensuring an agent's request reflects the user's true intent), and accountability (determining liability when something goes wrong).

At the heart of AP2 are Mandates — tamper-proof, cryptographically signed digital contracts that serve as verifiable proof of a user's instructions. For real-time purchases, a user's approval signs a Cart Mandate that creates an unchangeable record of the exact items and price. For delegated tasks — "buy concert tickets the moment they go on sale" — the user signs a detailed Intent Mandate upfront that specifies price limits, timing, and conditions, allowing the agent to act autonomously within those bounds.

What makes AP2 particularly significant is its universality. It supports credit cards, debit cards, stablecoins, and real-time bank transfers. Google has also launched the A2A x402 extension in collaboration with Coinbase, Ethereum Foundation, and MetaMask — a production-ready solution bridging AP2 with crypto-native payments. The protocol was developed with over 40 partners including Visa, Mastercard, PayPal, Stripe, Shopify, and American Express.

Visa Intelligent Commerce

Visa is not waiting on the sidelines. In late 2025, Visa announced it had completed hundreds of secure, agent-initiated transactions with partners across its network — a milestone the company described as signalling that "2025 will be the final year consumers shop and checkout alone." Visa's Intelligent Commerce initiative is building the infrastructure for AI agents to operate within the trusted Visa ecosystem, with fraud detection, buyer protection, and merchant workflows intact.

Visa's approach is pragmatic: rather than building an entirely new protocol, it is extending its existing rails to accommodate agent-initiated transactions. This means merchants can accept agent payments through the same infrastructure they already use, with Visa handling the trust, authentication, and dispute resolution layers. For the billions of merchants already on Visa's network, this dramatically lowers the barrier to participating in agentic commerce.

Mastercard Agent Pay

Mastercard has taken a similarly aggressive stance with Agent Pay, unveiled in April 2025 and expanded with new tools in September. Agent Pay introduces new merchant interfaces designed to distinguish trusted agents from bad actors, along with an Agentic Token Framework that infuses security and transparency into every agent-initiated transaction.

Mastercard's framework is notable for its focus on scalability and ubiquity — enabling merchants to participate securely without demanding significant development effort. The company is also collaborating across the ecosystem, participating in Google's AP2 protocol, OpenAI's Agentic Commerce Protocol, and the Linux Foundation's A2A standard. This cross-protocol collaboration signals that the major payment networks see agentic commerce not as a niche experiment, but as the next fundamental evolution of how money moves.

Why This Matters: The Commerce Implications

The convergence of these protocols creates something unprecedented: an economy where autonomous agents can discover each other (NANDA), verify capabilities (AgentFacts), communicate (A2A), access tools (MCP), pay via crypto rails (x402), transact through traditional payment networks (AP2, Visa, Mastercard), and do all of this without human intervention.

Consider what this means in practice over the next 2–3 years:

Procurement agents that autonomously source, negotiate, and purchase supplies across thousands of vendors, comparing prices and terms in real time
Financial agents that manage treasury operations, execute trades, and rebalance portfolios based on market conditions — settling transactions via smart contracts
Supply chain agents that coordinate logistics across borders, handling customs documentation, payment, and compliance autonomously
Personal agents that manage subscriptions, negotiate bills, book travel, and handle routine financial decisions on behalf of individuals

This is not a distant future. OpenClaw users are already demo-ing agents that act as personal assistants, software developers, and — occasionally — entrepreneurs. The tools like ClawRouter (autonomous model routing via x402), Clawpay (private agent payments), and ClawCredit (agent-native credit lines) emerged within days of OpenClaw's breakout.

The Trust Problem We Must Solve

With this opportunity comes a profound responsibility. When trillions of agents are transacting autonomously, the attack surface is enormous. We have already seen over 1,800 exposed OpenClaw instances leaking API keys within the first week of its popularity. This is not a bug in the concept — it is a warning about the infrastructure gap.

We need:

Verified tool registries — MCP servers where every tool's behaviour is cryptographically attested and auditable
Agent identity standards — like NANDA's AgentFacts, providing portable, verifiable credentials that agents carry across platforms
Programmable trust — smart contracts that encode the rules of engagement, escrow payments, and enforce SLAs between agents
Privacy-preserving discovery — allowing agents to find and verify each other without exposing sensitive organisational data

This is precisely the kind of infrastructure we are building at Yugo. The intersection of agentic commerce, blockchain, and programmable money is not a theoretical exercise — it is the foundation of how the next generation of global trade will operate.

What I Expect in the Next 2–3 Years

2026 will be the year of agent frameworks and protocol convergence. MCP, A2A, and x402 will mature from early-stage protocols into production infrastructure. We will see the first meaningful agent-to-agent commercial transactions outside of controlled demos.

2027 will be the year of agent identity and trust. NANDA-like discovery layers will become essential as the agent population grows beyond what manual configuration can handle. Verified tool registries will become table stakes for enterprise adoption.

2028 will be the year agentic commerce goes mainstream. The majority of routine B2B transactions will have some form of agent involvement. The distinction between "AI-assisted" and "AI-autonomous" commerce will blur, and the companies that built the infrastructure early will define the market.

The internet was built for humans. The next internet will be built for agents — and the humans who direct them. The question is not whether this transformation will happen, but who will build the infrastructure that makes it trustworthy.

Georgi Shulev is the co-founder of Yugo, building financial infrastructure at the intersection of agentic commerce, blockchain, and AI.