AI agents are learning how to spend money, but the wallets designed for them haven't been built yet.

Author and source: GeekPark

In July 2026, Natural, a company just one year old, secured $30 million in Series A funding. Its mission, in one sentence, is a bit hard to summarize—it aims to build a new payment system for AI agents.

This could have been just another routine funding news item. But when you look at everything that has happened over the past few weeks together, it becomes clear that a tipping point is approaching in the AI payments space.

01. Everyone is racing to adopt AI payments

The story begins with a somewhat unusual HTTP status code.

402, "Payment Required," was introduced into the internet protocol specification in 1991 as an interface reserved for a "future internet payment system," and has remained unused for 35 years without ever being formally implemented.

On July 14 last week, the Linux Foundation announced the establishment of the x402 Foundation, officially activating this dormant status code to facilitate autonomous payment settlements between AI agents. Visa, Mastercard, Stripe, AWS, and 40 other members have joined.

On the same day, the blockchain platform XRP Ledger announced that AI agent transactions on the mainnet surpassed 1 million.

Two weeks earlier, on July 2, Stripe, in collaboration with Cross River Bank, launched a dedicated virtual single-use card issuance service for AI agents, enabling autonomous software to complete transactions without accessing users' actual payment credentials; at that time, 1.6 million transactions had already been settled via the x402 protocol.

Three things stacked together have an unusual density.

Natural’s $30 million investment was secured at this stage, bringing its total funding to $40 million when combined with its previous seed round. Forerunner, known for its expertise in consumer investments, shifting its focus to B2B financial infrastructure is itself a significant signal.

02. The "Stripe" of AI

Three members of the Natural team previously had experience in consumer finance.

Kahlil Lalji and Eric Wang previously co-founded Ivella, a consumer finance company incubated by Y Combinator’s S21 cohort, which offered joint bank accounts for couples. After launching, the product served tens of thousands of users, but amid rapidly tightening regulatory conditions, Ivella was acquired by EarnIn in 2024, with the team integrated into the acquiring company. That experience gave Lalji and Wang firsthand expertise in financial compliance, real-time payment splitting, and bank network integration—exactly the kind of “underwater engineering” knowledge most critical for AI-powered payments.

The third co-founder, Walt Leung, was previously an engineering manager at Nextdoor, where he led the platform's technical team.

In 2025, three individuals reconnected in San Francisco to found Natural. The company completed its Series A funding just 193 days after incorporation and currently has a team of 17, with core engineering and product roles recruited from Stripe, Ramp, and Square.

Natural positions itself as an "orchestration layer for AI agents"—not building AI itself, but enabling AI to legally and securely move money within financial systems. Specifically, it has launched six product modules: Wallets are FDIC-insured agent wallets; Vaults are one-way accounts that accept funds but do not allow withdrawals, used to isolate risk capital; Pay, Request, and Transfer correspond to payments, receiving funds, and transfers; and Connect is the integration layer for platforms and markets.

If you look at this product suite in its entirety, it’s not just a payment tool—it’s a comprehensive financial infrastructure built around AI agents, covering the entire lifecycle from account opening and custody to payments and platform settlement.

Over the coming months, Natural plans to gradually roll out Voice (enabling agents to collect payment information over the phone), Accept (allowing agents to receive payments as merchants), and Cards (issuing debit and prepaid cards to agents). By the fourth quarter, deeper financial products such as Charge (billing based on API calls) and Credit (providing credit lines to agents) will also be launched.

The ambition of this product roadmap is clear: it doesn’t want to be a plugin for Stripe—it aims to become Stripe for the AI world.

The lead investor in this round was Kirsten Green, a partner at Forerunner. Known previously for investing in consumer brands, Green is now betting on infrastructure, citing a straightforward reason—Natural isn’t just helping agents “check out”; it’s rewriting the payment infrastructure itself, including how disputed transactions are handled.

However, one fact deserves a calm perspective: Natural is still pre-revenue; the $40 million in funding and $150 million valuation are bets on the projected scale of the agent economy, not on a proven business model. This means its time window is both an opportunity and a constraint—if enterprise customer adoption is slower than expected, this company of just 17 people must precisely balance its burn rate with its product timeline.

03. The Challenges of Current Payment Systems

To understand what Natural needs to do, first understand why today’s payment systems aren’t sufficient.

Kahlil Lalji, co-founder and CEO of Natural, offered a straightforward description: “The way payments are executed is shifting from human execution to agent execution. The astronomical volume of payments will transform the entire payment system.”

There is a specific engineering challenge behind this. Traditional payment systems were designed with the assumption that "a person is always behind the operation." Swiping a card requires the cardholder, online banking transfers require the account owner, and even corporate procurement needs an authorized person to confirm at some point. The entire logic is manually closed-loop.

But AI agents don’t work this way. Theoretically, they don’t require human intervention throughout the entire process—from identifying suppliers, comparing prices, to completing purchases. The problem is, when it comes time to make a payment at any stage, the AI agent lacks an “identity”—it cannot open a bank account, has no credit history, and possesses no traditional KYC information.

Brian Armstrong, founder of Coinbase, put it plainly: “They can’t open bank accounts, but they can own crypto wallets.”

Stripe has introduced virtual single-use cards—a阶段性 engineering solution that uses a brand-new, disposable virtual card for each transaction, locking AI behavior within strictly authorized boundaries. This approach is clever, but it also reveals one thing: existing systems are patching for AI, rather than being natively designed for it.

Natural aims to establish "payment infrastructure built for AI" as a distinct category.

04. B-end before C-end

McKinsey provided a forecast figure—agent-based commerce is expected to grow to a revenue range of $3 trillion to $5 trillion by 2030. This figure is large enough that nearly all major payment players have begun to make statements. Mastercard has named its related product line “Agent Pay” and described in its statement: “These agents can continuously transact with each other at machine speed, executing chains of transactions, including microtransactions.”

"Machine speed" is crucial.

Traditional payment systems operate on a daily settlement cycle, and fraud detection relies on statistical patterns of human behavior, with risk control models assuming that humans leave identifiable traces behind each transaction. AI agents, however, behave entirely differently—they can initiate dozens of transactions within milliseconds, exhibiting no human hesitation, habits, or geographic location anchors.

The security architecture needs to be rewritten, and there is little disagreement within the industry on this point. The disagreement lies in who should lead the effort and which path to take.

Fund x402 has chosen the path of an open protocol: activating HTTP 402 to establish a standardized settlement layer governed by mainstream institutions, making it appear more like "industry infrastructure" than a proprietary moat of any single company.

Natural’s approach focuses more on a vertical stack of financial tools—providing businesses with a complete suite to securely enable AI agents to execute payments.

The two paths are not mutually exclusive, but which one can achieve scale first largely depends on companies' actual willingness to make purchases.

Javelin Strategy & Research added a dose of reality to their year-end study: Agent commerce has had "minimal impact to date," and mainstream consumer adoption is more likely to occur in 2027 or 2028. B2B and backend automation are the real battlegrounds today; consumers will need to wait.

This judgment is likely correct. AI automation in scenarios such as comparing freight suppliers, automatically renewing enterprise SaaS subscriptions, and supply chain payments will arrive much earlier—and with lower technical difficulty—than convincing consumers to “let AI shop for me.”

Early customers of Natural should primarily focus on these enterprise service scenarios.

05. Outstanding Issues

Technology can move fast, but legal frameworks won't keep up.

An analysis by Fenwick & West highlights a structural issue: existing financial and consumer protection laws were designed with the assumption that humans make decisions. When an AI agent autonomously executes a transaction that results in an error or dispute, who is liable—the user who issued the instruction, the company that developed the AI, or the platform providing the payment channel? U.S. law currently offers no clear answer.

Stablecoin payment pathways lack the chargeback protection mechanisms found in traditional credit cards, which is a clear disadvantage for consumers. The x402 protocol promotes an open settlement layer, but openness does not equate to protection.

These issues won’t be resolved in the short term, but they will become significant concerns for businesses considering AI-powered payment tools—especially in scenarios involving large transactions or sensitive supplier relationships, where the phrase “AI autonomously completes” introduces legal ambiguity that many finance departments will hesitate to accept.

To make the business model viable, Natural may need to implement a more nuanced permission hierarchy between "fully autonomous" and "human-verification nodes," rather than simply selling the concept of "fully delegating payments to AI."

From this perspective, Stripe’s approach to virtual single-use cards may be conservative, but it precisely aligns with enterprises’ trust thresholds. Giving an AI a “controlled card” with strictly limited usage, where each transaction has defined spending limits and scope, is more readily accepted by today’s risk management teams than allowing the AI to make unrestricted payments.

On the day the AI agent economy truly matures, payment infrastructure will undoubtedly be one of the critical bottleneck areas.

Natural’s $40 million funding, Stripe’s single-use cards, and the 40 members of the x402 Foundation are all betting on the same direction. But who will survive the legal gray zone and who will first achieve scale in real-world business scenarios—no one can answer that yet this summer of 2026.

The market window has opened, but the foundation isn't finished yet.