loading...
Natural is building payment infrastructure for AI agents – wallets, vaults, and settlement rails that Stripe was never designed to provide.
ENTRY ANGLES
Agent treasury management – spending policies, allocation rules, multi-agent cost accounting · Inter-agent settlement protocols – clearing and reconciliation for multi-agent workflows
VERTICALS
CAPABILITIES
Payment infrastructure expertise, Multi-agent workflow understanding, Compliance and financial regulation
Six products for AI agents – wallets, vaults, payment rails, settlement accounts, marketplace tooling, and inbound collection – shipped before Natural was 193 days old. Stripe has shipped zero.
The problem Natural is solving is not primarily about cost or speed. It is about missing primitives. Stripe was architected for businesses run by people: a legal entity opens an account, a human authorizes transactions, compliance flows assume someone capable of accepting liability. None of that scaffolding maps onto an AI agent running autonomously. An agent that books a vendor, collects subscription revenue, and settles costs with other agents in the same workflow needs financial infrastructure that was never built – not because it was hard, but because the demand didn't exist until AI agents became capable enough to need it.
Kahlil Lalji recognized the gap before it became obvious. He and co-founders Eric Wang and Walt Leung – Lalji and Wang had previously built Ivella, a financial product for couples acquired by Earnin in 2023 – founded Natural in 2025. The platform provides FDIC-insured wallets for agents, Vaults (one-way accounts into which agents can move money but never withdraw), and four other products now at general availability. Seven more are rolling out through Q4, including Cards, Voice (PCI-compliant card collection over phone calls), and Charge, which enables per-API-call billing. A $30 million Series A led by Forerunner's Kirsten Green closed in July 2026, bringing total capital past $40 million. All major earlier investors returned to the round.
The Vaults design is worth specific attention. When a company gives an AI agent access to financial infrastructure, it faces a trust problem: an agent with a standard account can, in principle, overspend, misallocate, or execute transactions its principals didn't authorize. Vaults solve this by architectural constraint – money flows in but cannot flow out. The agent accumulates a balance from transactions it collects; disbursements require a separate mechanism. It is a spending control model that doesn't require a human to approve each transaction in real time but prevents the agent from having unconstrained financial authority. That's a cleaner solution than rate limiting or post-hoc transaction monitoring.
The per-API-call billing primitive (Charge, coming Q4) extends this logic to the economic unit of AI agent work. Most payment systems are built around time, events, or headcount. AI agents work in API calls: a customer-facing agent might execute 200 tool calls to complete one task, and the billing model for services that support it should reflect that unit rather than forcing a mismatch between how the agent works and how its costs are metered. No existing payment infrastructure was designed for it.
The market sizing – agentic AI payments at $7 billion today, projected to reach $93 billion by 2032 – is speculative, but the directional argument is solid. As agents take on more financial responsibility – managing departmental budgets, paying contractors, collecting from customers – the transaction volume flowing through non-human principals will grow faster than what flows through human ones. Natural is positioned as the infrastructure layer for that transition, and it shipped six products before most of the market had even named the category.
The entry question for builders is not whether to compete with Natural but what sits above it. Natural provides the financial primitives; it does not provide the logic that tells agents how to use them. Agent treasury management – spending policies, budget allocation rules, multi-agent cost accounting, audit reporting – does not exist as a product category. Any enterprise running multiple agents across a workflow needs tooling to define and enforce how those agents spend, and that tooling needs to integrate with payment infrastructure like Natural's. The addressable buyer is the same CFO or engineering leader who buys Natural itself, and the product is complementary rather than competitive.
The second opening: agent-to-agent settlement. An orchestrator agent that routes tasks to three specialist agents and receives results needs a way to aggregate the costs of those tasks and settle them against a single budget allocation at the workflow level. Natural's Connect product moves in this direction; the orchestration-layer settlement mechanism – a billing model that tracks costs across multi-step agent interactions and reconciles them at workflow completion – has not been built. The company that defines how inter-agent economic contracts work will occupy a position analogous to what clearing houses occupy in human financial markets: infrastructure that every participant depends on but no single participant is well-positioned to build.