Seventeen people. Less than a year in business. A $150 million valuation, as reported by Axios in May 2026.
Natural's $30 million Series A, led by Forerunner's Kirsten Green and announced in late July, is the kind of round that makes you look twice—not just at the company, but at what's happening around it. Because while Natural bills itself as building "payments for AI agents," the more revealing story might be who else is making the same bet, and how quickly the entire financial services industry has pivoted to chase a market that's still more projection than reality.
Consider: Visa, Mastercard, Stripe, and Coinbase have all publicly declared, in the span of six months, that autonomous software agents need their own payment infrastructure. And that they, naturally, intend to provide it.
Natural—which is constructing what it calls primitives for ledgering, money movement, multi-bank settlement, fraud detection, and agent identity—has now raised approximately $40 million in total disclosed funding as of July 2026 (including a $9.8 million seed round announced in October 2025). The valuation is striking. The speed is striking. But what's perhaps most notable is the validation embedded in the round itself: a thesis that autonomous software will soon transact at scale, and that existing financial rails weren't designed to handle it.
Whether that thesis is correct is another question entirely.
The Infrastructure Scramble
This isn't a niche anymore, if it ever was. The land grab is already well underway.
Mastercard launched Agent Pay for Machines on June 10—a system designed specifically for micro-transactions and always-on machine payments. Visa published its "Agentic Commerce" framework in March and updated its Core Rules in February 2026 to establish an "Intelligent Commerce" program with specific provider requirements. American Express rolled out its ACE Developer Kit in April, complete with protections for registered agent purchases. Stripe introduced its Machine Payments Protocol on March 18, then expanded with an Agentic Commerce Suite at its Sessions conference in April—featuring partnerships with Meta and Google, support for Mastercard Agent Pay and Visa Intelligent Commerce, and new Shared Payment Tokens.
Then there's Google, which donated the Agent Payments Protocol (AP2) to the FIDO Alliance in late April. Shopify and Google co-developed something called the Universal Commerce Protocol (UCP) back in January. AWS launched Bedrock AgentCore Payments in preview in early May, partnering with Coinbase and Stripe. Coinbase announced "Coinbase for Agents" in mid-June and called AI payments its "most high-conviction bet" a month later. Circle launched its AI Infrastructure on May 11, 2026, and Adyen launched its suite on June 16, 2026. Plaid, Marqeta, Fiserv, FIS—all have shipped agent-focused products or partnerships in the past half-year.
This isn't a corner of fintech anymore. It's the entire industry repositioning itself, and fast.
Five Protocols Walk Into a Bar
The technical infrastructure layer, meanwhile, is fragmenting almost as quickly as it's being built.
AP2 uses a mandate-based authorization model and now sits under FIDO governance. UCP, from Google and Shopify, provides commerce schema and structural scaffolding. Stripe and OpenAI's Agentic Commerce Protocol (ACP)—first announced in September 2025—powers programmatic checkout flows. Etsy went live with it; Shopify merchants are reportedly coming soon. Stripe's Machine Payments Protocol (MPP), launched in March, handles streaming payments and micropayments over stablecoin rails. And x402 revives the old HTTP 402 status code as a payment layer for what proponents call the autonomous web, with Coinbase, AWS, and others building support.
Each protocol reflects a different bet on what agents will actually need. AP2 prioritizes structured mandates and authorization evidence. UCP assumes commerce needs unified schemas before payments can scale. ACP embeds checkout directly into conversational interfaces. MPP focuses on sub-cent, API-native flows. x402 treats every HTTP request as potentially monetizable.
The problem? None of these are interoperable by default, and developers building agents don't particularly want to implement five different payment flows. Natural's pitch—offering payment primitives that work over any rail—addresses this friction. But so do Stripe's Shared Payment Tokens, Adyen's "universal translator" for agent channels, and Plaid's Guaranteed Payments infrastructure.
The race isn't just to build the pipes. It's to own the abstraction layer that sits above them.
The Trillion-Dollar Projection
The market projections are, to put it mildly, optimistic.
McKinsey estimated last October that AI agents could orchestrate $3 to $5 trillion in global consumer commerce by 2030, with U.S. B2C retail alone reaching $900 billion to $1 trillion. Gartner projected in January that 60 percent of brands will use agentic AI for one-to-one interactions by 2028. Grand View Research pegged the "agentic commerce" market at $5.71 billion in 2025, growing at a 35.7 percent compound annual rate through 2033.
But current transaction volumes? They tell a very different story.
A Forbes piece in June cited Coinbase figures claiming 169 million payments across 590,000 buyers and 100,000 sellers in x402's first year. Yet a16z crypto, analyzing the same data in April, argued that adjusted agent-driven x402 volume is closer to $1.6 million per month after filtering out what they described as "inorganic activity"—wash trading, test transactions, and other noise. Visa and Artemis published their own analysis in July that excluded similar artifacts, offering a more conservative baseline for onchain agentic payments as of late April.
Shopify reported that AI-referred orders grew nearly 13-fold year-over-year in Q1 2026, and AI-referral sessions grew eightfold. But the company didn't break out what percentage involved autonomous agent transactions versus plain old AI-assisted browsing.
The methodologies differ wildly across sources. What counts as "agent-driven" versus "agent-assisted" isn't standardized, and volume claims are contested. Perhaps that's the point—the infrastructure is being built for a future that hasn't fully materialized yet. Whether that's visionary or premature depends on who you ask, and when.
Regulators Are Paying Attention
Meanwhile, the regulatory frameworks are moving faster than you might expect.
The EU AI Act's transparency rules begin taking effect in August. Systems that interact with natural persons or generate content will face new disclosure requirements, and the Code of Practice for general-purpose AI providers will operationalize obligations that could affect how agents present payment requests. The EU's "AI Omnibus" entered into force in late July, with service desk FAQs clarifying agent obligations under Article 50.
In the United States, the Treasury's Financial Stability Oversight Council launched its AI Innovation Series back in March. The Senate Banking Committee sent oversight letters in June asking how agentic tools interact with existing payment laws. Senator Warner released a discussion draft of something called the AI AGENT Act in late June, focused on creating frameworks for secure AI agents. The GENIUS Act, separately, addresses stablecoin classification as permitted payments—which intersects directly with crypto-native agent rails.
Singapore's IMDA launched its Model AI Governance Framework for Agentic AI in January, with updates in May covering accountability for autonomous agents. The IMF published a note in April flagging unresolved liability, authorization traceability, and correlated agent behavior risks in payments. The Bank for International Settlements issued a working paper on agentic AI and payments risk management the same month. Federal Reserve Governor Bowman emphasized risk management in remarks at the FSOC AI Series in May.
These aren't speculative policy discussions. They're active frameworks being operationalized now, and companies building in this space will need to comply with rules that didn't exist 18 months ago. That timeline matters.
The Unsolved Problems
For all the infrastructure being laid down, some fundamental questions remain stubbornly unanswered.
Liability for autonomous transactions, for one. If an agent executes a fraudulent payment, who's responsible? The agent developer? The user who deployed it? The payment infrastructure provider? The merchant? The IMF and EU materials flag this repeatedly, and none of the current protocols provide a definitive answer.
Authorization traceability is similarly murky. AP2 and UCP incorporate structured evidence and mandate systems, but recent academic papers have documented injection attacks, replay vulnerabilities, and context-binding failures. "Five Attacks on x402," published in May, and "Whispers of Wealth: Red-Teaming Google's AP2," published in January, outline protocol-level risks that could enable payment hijacking. A July paper catalogued cross-platform vulnerabilities in what it termed "agentic commerce platforms."
Agent identity itself remains contested territory. Antom's AMP includes something called KYA—Know Your Agent. Rain's Agent Control Layer, launched in June, offers programmatic spending guardrails. But there's no industry standard. Visa, Mastercard, and Amex are each building agent-scoped token systems with different permission models. Stripe's Shared Payment Tokens provide yet another approach. Natural emphasizes infrastructure-layer identity and observability as core primitives.
These are competing visions of how to solve the same problem. The market hasn't decided which will win, assuming any of them do.
Then there's dispute resolution. Traditional chargeback systems assume a human is contesting a charge. What happens when two agents disagree about a transaction? Or when an agent's behavior was technically correct according to its mandate but commercially irrational? Nobody seems quite sure yet.
What This Means, If Anything
Natural's approach—abstracting payments complexity away from agent developers and offering a unified layer for ledgering, settlement, risk, compliance, and identity—echoes what Stripe did for online payments a decade ago, or what Plaid did for bank connectivity. CEO Kahlil Lalji wrote back in March that agentic payments require these primitives at the infrastructure layer, not pushed into application logic. The company announced plans for merchant acquiring and direct bank integrations by the end of Q2, targeting three banks in the following year. Whether that timeline has held is unclear.
But Natural isn't building in isolation, which is perhaps the most interesting part. Stripe already has distribution through its existing merchant base and partnerships with Meta, Google, Etsy, and Shopify. Visa and Mastercard control the card networks and are extending them with agent-specific capabilities. Coinbase has positioned AI payments as existential to its business and is integrating deeply with AWS Bedrock. Adyen, with its "universal translator" positioning, serves enterprise merchants directly. Plaid sits between banks and applications and is layering agent tooling on top of guaranteed payments infrastructure.
The market is fragmenting between card-network approaches—managed trust, existing compliance frameworks, agent-scoped tokens—and crypto-native rails, with their sub-cent settlements, API-native flows, and stablecoin liquidity. Jack Forestell, Visa's chief product and strategy officer, called the agentic web "the most important shift since early digital commerce" back in March. Mastercard framed Agent Pay for Machines as enabling fractional-cent programmatic payments on its global network. Coinbase's bet is that crypto rails will become the default for agent-to-agent settlements. Stripe's strategy appears to be supporting everything—cards, stablecoins, streaming micropayments—and letting merchants choose.
Natural, Stripe, Coinbase, Visa, and Mastercard are all racing to define what agent payments infrastructure looks like, which means the category is both crowded and wide open at the same time.
For founders building AI agents, the near-term choice is whether to integrate directly with emerging protocols (AP2, UCP, ACP, x402) or to abstract through a provider. For payment infrastructure entrepreneurs, the question is which layer to own—protocol, orchestration, risk and compliance, identity, or settlement. For financial services incumbents, the calculation is whether to build, buy, or partner before the category ossifies.
The Only Certainty
Here's what's clear: This market didn't exist 18 months ago. Now it has five competing technical standards, regulatory frameworks in multiple jurisdictions, $40 million venture rounds for year-old companies, and Fortune 500 incumbents launching dedicated products. The infrastructure is being built for transaction volumes that are still largely projected rather than realized, but that hasn't stopped the land grab.
Perhaps that's rational. Whoever controls the abstraction layer when autonomous agents do start transacting at scale—if they do—will capture an outsized share of what could be a multi-trillion-dollar payments shift. Or perhaps the entire category is still too early, and most of these bets will need to be rewritten as the actual use cases clarify and the technology catches up to the hype.
Either way, the payments industry has decided that agentic commerce is inevitable. The only question left is who builds the rails—and whether anyone will actually use them.
