In the agent age, you won't apply for financial products. Your agent will subscribe to them.

Last quarter I argued the winning bank won't have branches, it will have APIs. That was about agents being able to read financial services. This is the next step, and it's already arriving: agents won't just read financial products, they'll buy them.

‍

It's 7am. A SaaS company's treasury agent has already read the week's cash position across three bank accounts, the receivables aging in the ERP, and a supplier early-payment discount that expires Friday. It runs the math: paying early earns 2%, the gap to fund it is €180k for eleven days. It opens Defacto, requests the line, accepts the terms, and pays the supplier. By the time the CFO reads her morning summary, the discount is captured and the line is already scheduled to repay itself from Thursday's incoming wire.

Nobody logged into a portal. Nobody uploaded an invoice. The product wasn't sold to a person. It was subscribed by an agent.

This stopped being a forecast. At Stripe Sessions 2026, the entire narrative was agentic commerce: agents that buy, pay, and transact on behalf of their users. Robinhood is wiring agents into its users' activity. In France, insurers are preparing to let AI agents sell their first contracts (Les Echos). The frontier moved from "an agent can browse" to "an agent can transact." Payments went first because paying is the simplest action. Financial products are next, because they're the highest-value one.

The reframe: from findable to subscribable. A human-sold financial product assumes a human reads the page, fills the form, and signs. An agent-subscribable product assumes a machine evaluates terms against a goal and commits within a mandate. That's a different product, not a different funnel. It needs machine-readable terms, real-time eligibility, programmatic underwriting, and guardrails the agent operates inside.

To be clear about the near term: no SME is going to hand an agent blanket authority to take on debt unsupervised next quarter, and they shouldn't. The realistic on-ramp is agent proposes, human approves. Picture a virtual CFO sitting on a company's data: it runs the diagnostics, surfaces the best option to improve treasury, and the business clicks "ok." Set and forget, with limits. Full autonomy is the destination; assisted subscription is how we get there. But both rest on the exact same foundation: a financial product a machine can actually evaluate and act on.

Here's the part nobody sees, and it's the whole game. For an agent to subscribe to credit in one call, someone has to have turned a regulated, risk-laden, capital-intensive process into something an API can complete safely. That means underwriting and eligibility that run in real time, KYB and AML built into the flow instead of bolted on after, capital and risk management standing behind every euro, and pricing a machine can read.

This is the hard part, and it's exactly what McKinsey and QED just named the new moat. AI has collapsed the cost of building features: what took five years now takes six months. When the product is cheap to copy, the durable advantages are distribution, proprietary data, and regulatory posture. Or as they put it: "a feature is no longer a fintech," and "perhaps the best investment a fintech can make right now is in compliance."

So the agent-age story and the compliance story are the same story. Productizing a financial service for an agent is compliance-by-design plus real-time risk plus proprietary data. The agent layer is the demand. The productization layer is the moat.

Which is the layer we've been building. We've spent 5 years turning short-term B2B credit into an API. This quarter it turned into concrete steps toward the agent age:

• A credit card for SMEs: productized credit, fully flexible, no account opening. 100 clients in private beta.
• Our ECS license filing: owning more of the value chain means cheaper funding, less dependency on partner banks, and precisely the regulatory moat McKinsey describes.
• An MCP server: an agent can query eligibility and subscribe to a loan directly through our infrastructure.

We're not betting that agents will want financial products. Stripe, Robinhood, and a wave of insurers are already proving the demand. We're building the layer that lets a financial product be safely subscribed to by one.

The question for anyone building financial products is no longer whether agents will transact. It's whether your product can be acted on by something that will never read your homepage, never sit through your sales call, and never tolerate a five-day wait. If it can't, you're not a product in the agent age. You're a feature.

And a feature is no longer a fintech.