The agentic economy is the financial paradigm where AI agents hold balances, sign transactions, negotiate terms, and authorize payments autonomously or on behalf of humans. 

Rothschild, Mobius, and Hofman, along with their colleagues at Microsoft Research, formalized the term in their May 2025 paper, The Agentic Economy, which Communications of the ACM republished. But a year later, the agentic economy is no longer merely theoretical. The multi-layered protocol stack is shipping. 

Since May 2025, Nevermined has processed over 1.3 million autonomous agentic transactions on x402 rails. The ERC-8004 identity standard has over 24,000 registered agents. Visa, Mastercard, Cloudflare, AWS, Google, Circle, and Coinbase are all part of the x402 Foundation. Meanwhile, NVIDIA Founder Jensen Huang claims that agentic AI can rewire the $50-trillion industrial economy, and the ‘entire manufacturing line will be operated by robots, managed by more robots’ in all-robot factories of the future. 

Against this backdrop, we dive deep into the 4-layer stack that’s making the agentic economy operational, why stablecoins are its substrate, how intent-based execution moves value across chains, and what’s actually shipping versus what’s still being figured out. 

Key Takeaways

• Researchers at Microsoft formalized the ‘agentic economy’ framing in May 2025; the operational protocol stack — MCP, A2A, ACP/UCP, AP2, and x402 — shipped over the following twelve months and is now live in production. 

• Stablecoins, not credit cards, are the AI-native payment substrate. Card networks can’t price sub-cent payments, and need correspondent banks for cross-border settlements. Stablecoins on x402 don’t have such limitations. 

• Production receipts as of May 2026: x402 processed $24.24M over the 30 days to April 29, with 99.8% settlement in USDC. Circle’s USDC circulation hit $77B in Q1, while AWS Bedrock AgentCore Payments was released in preview across four regions. 

• Agents express outcomes, not routes. LI.FI’s Solver Network and the Open Intents Framework handle cross-chain settlements, enabling seamless liquidity and market access for the rest of the stack.

What the agentic economy actually means

Microsoft’s researchers defined the agentic economy as the realm of assistant agents that act on consumers’ behalf (i.e., consumer agents) and engage in machine-to-machine negotiations with ‘service agents’ that represent businesses. Following up on this definition, ACM claimed that trust, and not attention, is now the scarce resource. Because when a billion agents represent a billion consumers to a billion businesses, who the consumer agent trusts (or believes in) becomes a bigger question than who captures the click. 

Both of these framings, however, are largely academic. There’s also an operational definition, albeit the more important one in the present context: the agentic economy is the paradigm where AI agents are autonomous financial actors, holding balances, signing transactions, negotiating terms, authorizing payments, and settling value across chains on their own, with little or no human intervention.

The academic definition is about agents recommending things. The operational one is about them executing on them. And this difference reflects the fact that the operational rails weren’t around when the academic conception was formed: 

• Onchain wallets agents can use without exposing private keys.

• Payment protocols that settle without a card network.

• Settlement rails priced for fractions of a cent, running 24/7, clearing cross-border without correspondent banking.

This stack emerged over the twelve months following Microsoft’s paper. Infrastructure began. Pure academic treatment stopped. 

Identity, communication, security — The prerequisites

Sequoia partner Konstantine Buhler has been working with a framework for what the agent economy needs to function: persistent identity, seamless communication, and security. Trust is the emergent property here. Was the right party authorized to spend? Did the negotiation actually happen? Did the value move as decided? And so on. 

However, none of the top-ranking explainers on the agentic economy use this framework, which is a gap worth filling, since Buhler’s model maps almost one-to-one onto the protocols already shipping. 

Identity

An agent paying a $50,000 vendor invoice should not be the same agent streaming one-tenth of a cent per second to a compute provider. The infrastructure and counterparties must know which is which.

This layer includes onchain wallets, decentralized identifiers, Skyfire’s Know Your Agent (KYA) framework (now adopted by Experian), and Visa’s Trusted Agent Protocol, launched October 14, 2025, with Cloudflare. Account abstraction does the heavier lifting underneath: ERC-4337 has been live on Ethereum since March 2023, and EIP-7702 went live with the Pectra upgrade on May 7, 2025, recording over 11,000 EIP-7702 authorizations in its first week.

Communication

Anthropic’s Model Context Protocol (MCP) is currently the standard for how an agent talks to the ‘tools’ it uses to perform a host of actions, from fetching web data to creating documents and organizing folders. Monthly SDK downloads for MCP crossed 97 million by Q1 2026, with roughly 5,800 active servers. 

Google’s Agent2Agent (A2A) protocol is another leading agentic communication mechanism, and handles how agents talk to each other. A2A was donated to the Linux Foundation in June 2025 and now has 150+ supporter organizations.

Security

Google’s Agents Payments Protocol (AP2) introduced cryptographically signed user authorization via ‘Mandates’, co-developed with Coinbase, the Ethereum Foundation, and MetaMask. Coinbase’s x402 settles the actual payment onchain, where it’s auditable by design rather than by audit trail.

Four pillars of the agentic economy

Built around identity, communication, security, and ultimately, trust, the agentic economy as we see it today stands on four pillars. Think of them as layers of the stack. The best part? Each already has at least one active implementation today. 

Launched by Anthropic in November 2024, MCP is the first pillar, or base layer, of the agentic stack. It gave AI systems hands and legs, enabling them to interact with external processes, software, and APIs. This was critical to AI’s evolution from mere chatbots to active agents, just as inventing the wheel or using ‘tools’ was an inflection point in human evolution. 

However, while MCP made agents active and productive from an execution standpoint, it didn’t solve agent-to-agent interactions. Google’s A2A protocol — the second pillar or layer of the stack — did that. Thus, MCP and A2A work together to facilitate the capability (using tools) and community (working with other agents) aspects of agentic systems. 

Next comes the economic coordination layer. This is where agents negotiate with merchants (or their agentic representatives) using mechanisms such as Stripe and OpenAI’s Agentic Commerce Protocol (ACP), or Shopify and Google’s Universal Commerce Protocol (UCP). These were launched in September 2025 and January 2026, respectively. 

The fourth pillar is the authorization-and-settlement layer. Here, AP2 and x402 work together to validate agentic payments using ‘Mandates’ (constraints like price limits or timing) and settle them via stablecoin rails. Google invented AP2 and co-built the AP2 x402 extension with Coinbase, the Ethereum Foundation, and Metamask. 

With these, we now have the coordination primitives required for operational agents. But there is a missing link in the above structure, especially from the PoV of accessing fragmented onchain liquidity and markets. LI.FI solves this problem. It underlies the AP2 x402 extension as the routing layer that moves stablecoin-denominated value across chains when the payer and payee are in different ecosystems. LI.FI also provides a hosted MCP server, which is often the entry point for agent developers. 

For an in-depth analysis of how the agentic stack works in cross-chain contexts and what it needs to thrive there, read The State of Interop 2026.

Stablecoins are AI-native money

Similar to legacy infrastructure, typical forms of money aren’t conducive to the agentic economy. That’s why stablecoins, and not credit cards or checking accounts, have emerged as its substrate. 

Circle co-founder, Sean Neville, who now runs Catena Labs (out of stealth since May 2025 with an $18M seed led by a16z), framed the case for digital-native AI money cleanly, saying, “If an AI actor is making a payment, it doesn't make a lot of sense for it to have a physical credit card. It makes a lot of sense for it to have money that moves at internet speed.” 

Card networks fail agentic systems for three structural reasons. 

One, they set per-swipe interchange floors around the ten-cent mark. Fine for groceries, but fatal for any payment model priced in fractions of a cent. Two, they settle on banking business days, which works for procurement, but breaks when the model needs continuous flow. Three, they handle cross-border transactions through correspondent relationships. Alright if you’re a banked entity (which 1.3+ billion humans worldwide aren’t, let alone agents), but disqualifying if you’re an autonomous agent software that doesn’t meet legacy KYC/KYB requirements. 

Whereas stablecoins on x402 settle in fractions of a cent and run 24x7, regardless of the counterparties’ locations. And the rails that win with agent economies will be stablecoin-native, not simply because crypto is ideologically superior, but because cards can’t price micropayments or settle them across borders at 2 a.m. on a Sunday. 

To this end, the production data speaks for itself (resoundingly so): 

• x402 processed $24.24 million in the 30 days ending April 29, 2026, with 99.8% USDC settlement (Circle). Cumulative cross-chain transactions reached roughly 165 million by late April. 

• AWS Bedrock AgentCore Payments went into preview in April 2026 across US East (North Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney), with x402 as the default machine-to-machine rail. 

• Circle’s gas-free USDC nanopayments now support transfers as small as $0.000001 — a millionth of a dollar — settled directly on mainnet. 

• Mastercard’s Agent Pay program uses dynamic Agentic Tokens as a credential layer, giving AI agents a verified network-level identity

LI.FI’s Stablecoin API also facilitates treasury-grade stablecoin routing, especially as stablecoin issuance increasingly ties to dedicated chains, thereby fragmenting markets and liquidity. For a complete discussion on this, read Stablecoins Aren’t New.

The case for Intents: Agents express outcomes, not routes

AI agents are getting good at many things, but making low-level execution choices isn’t one of them. It’s naive to expect language models to efficiently decide which bridge to use, which DEX to trade on, or what slippage to set. They must be told. 

For the agentic economy to thrive, we must not ask the wrong tools to do the wrong job. Agents, for one, excel at expressing outcomes, but they need clear instructions: ‘Pay 50 USDC to this Solana address from my Arbitrum balance.’

Intent-based execution serves this reality. Users (or their agents) declare the desired end state. Professional solvers compete to deliver it. The infrastructure handles routing, bridging, and settlement. Agents don’t worry about technicalities like bridging funds from the source to the destination chain — they only need the outcome. 

This approach is winning, especially as standards come along. ERC-7683, for example, authored by Uniswap Labs and Across, defines cross-chain intents in a way every protocol can support. The Open Intents Framework launched in February 2025, with contributions from the Ethereum Foundation, Hyperlane, Bootnode, and LI.FI. Anoma, the intent-centric L1, also went live in two phases, featuring the Resource Machine VM with a custom intent-execution layer. 

Meanwhile, in The Missing Mechanisms of the Agentic Economy, Tim O’Reilly argued that the agentic economy still needs new market mechanisms beyond MCP and x402. He emphasized the need for intent-based, cross-chain settlements. That’s huge, coming from an open-source veteran who’s not necessarily crypto-native (and has been rather skeptical at times). 

LI.FI Intents is one production proof that cross-chain intent settlement works at scale. The routing layer has handled more than $80 billion in lifetime transfer volume across 100 million transfers, 1,000+ integration partners, and 60+ chains. The intent layer isn’t merely a thesis anymore, and already moves real value.

To fully understand how intent-based execution works across chains, read Intents and Chain Abstraction 101.

Money on the table — Sizing the agentic market

Besides the protocols, frameworks, and standards discussed so far, another notable production-grade implementation is Denso’s AI sourcing agents, which have consistently made autonomous purchases through Skyfire since August 2024. More recently, LI.FI also launched its API for agentic commerce, solving market access for agents.  

Such developments corroborate McKinsey’s projection that global agentic commerce will present a $3–$5 trillion opportunity by 2030, with $900 billion to $1 trillion in orchestrated B2B revenue in the US. And while that plays out, 5x of PayPal’s and nearly half of Visa’s payment volumes are already being settled in stablecoins. 

Gartner, however, predicted in June 2025 that more than 40% of agentic AI projects would succumb by the end of 2027, due to escalating costs, unclear business value, and inadequate risk controls. It will be interesting to see how this counterstatistic holds up, but as of now, the opposite trend is clearly dominating. 

That said, it would be equally blind (and biased) to overlook that the demand for agentic systems is still catching up with supply and has ample room for expansion. Only 30 Shopify merchants adopted the first-generation Instant Checkout, pushing OpenAI to pivot from it in March 2026. Comet’s shopping agent also faced a roadblock when U.S. District Judge Maxine Chesney granted Amazon a preliminary injunction against Perplexity. The ruling acknowledged that while Comet accessed Amazon with the user’s permission, it did not have authorization from Amazon.

The bottom line, therefore, is that the infrastructure for agent-led economies is both scaling and maturing. But the application layer is still finding its product-market fit, and there’s some ground to cover before demand — both retail and institutional — fully kicks in.

What are VCs saying? 

Before we wrap, it’s worth looking at how the capital allocators backing the agentic economy have aligned. 

a16z’s stance, as they elaborated in Big Ideas 2026: The Agentic Interface, is the most direct. This year, “AI stops being something you ask, and becomes something that does.” And while the same agent might not efficiently handle a $50,000 invoice and a $0.001/second compute stream (as we said at the beginning of this piece), Sam Broner of a16z crypto argues that the same rail can serve both. Meaning, the underlying infrastructure must be adequately neutral, highly composable, and operational 24x7. Like stablecoins and intent-based mechanisms. 

Sequoia’s Julien Bek also presented an adjacent thesis in Services: The New Software. For every dollar spent on software, six dollars are spent on services that make that software work. This further tips the balance in favor of agents that deliver outcomes rather than tools that deliver functionality. 

The VC takeaway is thus unanimous in direction, if not in detail. There’s an emerging consensus around the scaling potential of agentic economies. The question now is when this will happen and, more importantly, how fast.

For a broader discussion on how crypto’s institutional adoption has and will pan out, read Crypto’s Distribution Playbook: Or, the DeFi Mullet.

What’s next (and what’s missing)

Three challenges currently face agentic economies, though none of them are existential and won’t last long.  

Cross-chain settlement at scale is one of them. Especially from a market access standpoint. But this is already being solved, with LI.FI’s Solver Network, the Open Intents Framework, and x402. The infrastructure is ready, and demand is picking up. 

Agent identity at mass, retail scale is another. Skyfire’s KYA (now also Experian’s identity layer), Visa’s Trusted Agent Protocol, Mastercard’s Agentic Tokens credential layer, and Catena Labs’ AI-native banking solution are innovating in this domain. Their core focus is to help merchants and payment networks verify that their agentic counterparty is indeed who it claims to be or has the authority to do what it intends. There are no clear winners in this vertical yet. 

Finally, and this is where we’re currently lagging the most, is the need for ethico-legal frameworks designed for agentic systems. Because human-centric models are currently being applied to agents, either forcibly or ad hoc. The Amazon vs. Perplexity case is a glaring example of this, which closed the obvious scraping path without presenting an alternative. Questions about autonomous spending limits, fiduciary duty, or cross-border taxes will soon be at the center of protocol-mediated agentic commerce, and our legal systems must answer them without bias or regressive strongarming. 

The agentic economy and the infrastructure enabling it are both evolving at breakneck speed. It’s not waiting for archaic structures and processes to catch up. They won’t either, so the systems that are must adapt or fade. Consumers are changing, and so is the way they interact with businesses. Agents will be the primary economic actors of the future, while we spend more time pursuing ideas that create lasting value, enriching ourselves and the world around us. This is inevitable and will be unmistakably apparent to everyone within the next decade or less. It’s still early. Now is the time to build; to be bullish.

Frequently Asked Questions (FAQs)

What is the agentic economy?

The agentic economy is an emerging economic system in which AI agents act on behalf of consumers and businesses, discovering products, negotiating terms, authorizing payments, and settling transactions with little or no human intervention. The framing was formalized by Microsoft researchers (Rothschild, Mobius, Hofman, et al.) in The Agentic Economy (arXiv, May 2025) and republished by Communications of the ACM in January 2026. The operational stack has shipped over the following twelve months: MCP, A2A, ACP, UCP, AP2, x402, and stablecoin settlement across chains are now live in production. 

How is the agentic economy different from the digital economy?

The digital economy moves human decisions onto digital interfaces. Users click checkout, fill in forms, and authorize payments themselves. The agentic economy removes the human from these steps. Agents handle discovery, negotiation, and payment authorization on the user’s behalf using protocols like Stripe ACP, Shopify UCP, and Google AP2. The substrate also shifts: card networks cannot price micropayments, so stablecoin rails like x402 become the default for machine-to-machine value transfer. 

What protocols power the agentic economy? 

Four layers cover the production stack as of May 2026: 

• MCP (Anthropic, November 2024) for agent-to-tool calls.

• A2A (Google, April 2025; Linux Foundation, June 2025) for agent-to-agent coordination. 

• ACP (Stripe + OpenAI, September 2025) and UCP (Shopify + Google, January 2026) for merchant negotiation. 

• AP2 (Google, September 2025) for cryptographically signed user authorization via Mandates, paired with x402 (Coinbase, May 2025) for stablecoin settlement. 

Cross-chain routing, i.e., moving stablecoins between chains when payer and payee live on different ones, is the layer below x402, where LI.FI and the Open Intents Framework operate. 

Is the agentic economy real or hype?

The infrastructure layer is real, with multiple production-grade implementations. x402 processed $24.24M in the 30 days, with 99.8% USDC settlement. Nevermined has processed over 1.38M autonomous agent transactions since May 2025. AWS Bedrock AgentCore Payments is live in four regions, built with Coinbase and Stripe. Visa, Mastercard, Cloudflare, AWS, Google, and Circle are all part of the x402 Foundation. 

But the consumer-facing application layer hasn’t yet found its product-market fit. On the other hand, Gartner forecasted that more than 40% of agentic AI projects will be canceled by 2027, while OpenAI pivoted away from its first-generation Instant Checkout in March 2026 after only about 30 merchants went live.

What role does crypto play in the agentic economy?

Stablecoins are the AI-native payment rail. Card networks have interchange floors that make sub-cent payments uneconomic, and they settle on banking business days. Whereas on protocols like x402, stablecoins settle 24x7, support micropayments down to a millionth of a dollar (Circle’s gas-free USDC nanopayments), and clear cross-border without correspondent banking. AWS Bedrock AgentCore Payments runs on x402 by default. Mastercard’s Agent Pay uses dynamic Agentic Tokens as a credential layer. Google built the A2A x402 extension with Coinbase, the Ethereum Foundation, and MetaMask. And cross-chain routing layers like LI.FI’s Solver Network enables seamless market access across the stack.

When will the agentic economy go mainstream?

The infrastructure already is. MCP has 97 million monthly SDK downloads. A2A has 150-plus supporter organizations. AWS ships AgentCore Payments. 

However, consumer adoption is the gating factor. Visa forecasts that millions of consumers will use AI agents for purchases by the 2026 holiday season, though that is a marketing claim, not a measured number. 

Mainstream consumer agentic commerce probably arrives unevenly between 2026 and 2028, with B2B and enterprise use cases moving faster.

Try it!

Building agent infrastructure that has to move value across chains? 

Start with our agent overview documentation. And check out our Stablecoin API if you need treasury-grade stablecoin routing for moving capital at scale.