.svg)
Why We Invested in Natural: Payments for AI Agents
We're excited to announce our investment in Natural, the agentic payments platform designed for AI agents to hold, send, collect, and reconcile money autonomously.
Today, an agent can negotiate a freight load, source a contractor, manage procurement, and run a sales cycle. But the moment money needs to move, the workflow breaks and a human has to step in. Natural removes that step. It gives any AI agent a small set of payments primitives: an FDIC-insured account it can hold funds in, the ability to send and collect money, a credit line, and the ability to bill end customers, all behind a clean developer API. Underneath, Natural handles the parts agent builders shouldn't have to: ledgering, routing, compliance, risk, credit, identity, and observability. We believe the companies building agentic products today will, over the next decade, own a meaningful portion of all payment volume globally - but they need payments infrastructure built around their workflows rather than retrofitted from a human checkout flow.
The shift in payments
For the last twenty years, payments infrastructure has been built on one assumption: a human is in the loop. A person clicks the button, types the card number, signs the check, approves the wire. Every layer of the payments stack reflects it. KYC/B is built around human identity. Fraud models are calibrated against human behavioral patterns. Disputes resolution assumes a human chargeback. Checkout assumes a session with a screen and a keyboard.
Agents break that stack in three places. First, the actor moving money is no longer human, so the identity layer is incomplete: you now need to not only verify the user, but also the agent acting on its behalf, the scope of its authority, and the instruction that triggered a given transaction. Second, today's fraud and AML systems were trained on how humans spend, so an agent's perfectly legitimate behavior trips alarms that were calibrated for people. Third, agents need first-class programmatic primitives - accounts, payouts, collections, credit, billing - not human dashboards retrofitted with a few API endpoints.
Why B2B is the right wedge
The mainstream narrative today on agentic payments is anchored on consumer commerce. Consumer agentic commerce will eventually be real, but the need today is far less urgent, the transactions are low-value (typically sub-$100), and betting there means betting against Stripe in the exact surface area it is most motivated to defend, and where it has already partnered with the major model providers.
B2B is the inverse on all three counts. The pain is immediate and operational: a logistics agent that can't settle freight, a property management agent that can't pay a contractor, a back-office agent that can't pay vendors. There's budget authority, real ROI on automation, and meaningfully larger transactions (e.g. 4-7 figure payments). And the volume being displaced is ACH, wire, and RTP - the legacy bank rails Stripe has limited incentive to prioritize in the near-term because they don't generate interchange. Natural is building straight into the part of the market where the pain is most acute and the incumbents are least motivated to compete.
And the market being displaced is enormous: B2B payments globally are roughly ~$125T in annual transaction volume and U.S. ACH alone is ~$80T a year. Even a low-single-digit-percent shift onto agentic rails is north of $1T in TPV.
The demand for B2B agentic payments is also already here. Voice agents (e.g. ElevenLabs, Bland), browser agents (e.g. Browserbase, Anthropic Computer Use), search/research agents (e.g. Profound, Glean, etc), and vertical workflow agents (e.g. Mason, HappyRobot, etc) have all reached significant customer scale. Almost all of them eventually need to move money, and almost none of them have a viable native solution today.
What got us excited
Owning the stack end-to-end. Natural is building ledgering, escrow, identity, agent observability, dispute workflows, settlement, card acquiring, AML, and credit underwriting in-house from day one. It's the harder, longer build, but it also lets them ship at the speed the market demands (especially once agentic volume inflects) instead of waiting on a third-party processor's roadmap or a bank IT cycle. This is also a bet that you can't win agentic payments with a point solution: a single agent workflow routinely spans holding funds, paying a vendor, collecting, drawing on credit, and billing the end customer. Stitching that across five providers means five identity checks and five places to break, while Natural does all of it behind a single API - one ledger, one identity model, and one place any transaction can be traced end to end.
Agent identity as a moat. Every dollar moved on Natural ties back to a verified entity, the specific agent that moved it, and the original instruction that produced it. That last piece matters most: every payment is auditable back to the request that created it, and no other platform has it. Over time that graph - entity to agent to instruction to transaction - becomes a proprietary dataset for fraud and AML detection that is hard to displace. A good analogy is Stripe Radar: Stripe doesn't sell Radar as the headline product, but the fraud advantage it gets from seeing how every card transacts is one of its deepest moats.
Credit first is a strong wedge. Pre-funded wallets create friction and obvious churn points: move money in, watch it drain, top it back up. Credit compresses time-to-value: an agent can transact on day one, and Natural settles at the end of the cycle. It's the right focus on end-user experience, and it also carries better economics (credit roughly triples the take rate on that volume).
Trust is the product. Across agentic categories, the binding constraint on deeper enterprise adoption is trust, not technology. For payments, the trust problem is far more acute: letting an agent move real money on your behalf demands a level of confidence that no other agent workflow requires. Natural's identity, observability, and AML infrastructure is fundamentally a trust product, and that trust is what gives businesses the confidence to let their agents transact in the first place.
Brand as a moat
Brand as a moat is the thesis Torch was built on. We believe category leaders, irrespective of business model and end-user, are the ones that build great brands - and in payments infrastructure specifically, trust and developer love are as important as any single feature, if not more.
Natural is already executing on their brand better than any six-month-old company we've seen. The launch brand, API docs, product design, and public hiring testimonials are some of the best we've seen at any stage, and the team's writing is fast becoming the reference point for developers and operators to think through problems within agentic payments.
The team
For a company this young, the caliber of the people Natural has hired is extraordinary. Six months in, Natural has already recruited high-caliber senior payments and regulatory talent from leading payments companies like Stripe and Ramp.
We're thrilled to partner with Kahlil, Eric, Walt, and the team as they build the money-movement layer for the agent economy.
.svg)
.png)
