The next fintech isn't a bank

A few weeks ago, we made a simple experiment - build a working bank prototype in 24 hours. It worked. Then the next question: a production-grade MVP in two weeks - something that used to take a year and a half. Turns out you can do that too.

That experiment crystallized something that had been forming for a while. Corporate banking - the most lucrative segment in financial services, where $1.2 trillion in annual profits concentrate - is running on infrastructure so outdated that the market prices it in: banks trade at a 70% valuation discount to other sectors. The market already knows these profits sit on a rotting foundation. The question is who builds the replacement.

This memo is an attempt to articulate why the answer is becoming clear.

Next great fintech won't be a bank at all. It'll be an application. The bank will be the blockchain. The regulation will be the stablecoin. And the experience layer on top will be built entirely for AI agents.

The problem

Corporate banking is where innovation goes to die. The reason is structural.

A large corporation's financial stack isn't one system - it's dozens. Accounting, treasury management, procurement, order management, production planning, HR, payroll. These systems were purchased and integrated over decades. They form a single, deeply entangled organism. Banks globally spend $600 billion a year on technology - more than the high-tech industry itself - yet dedicate up to 75% of that to maintaining legacy systems. An estimated 43% of banking systems still run on COBOL, a language from 1959. US bank labor productivity is actually declining despite all that spend.

There have been plenty of attempts to crack this. Through ERPs, through AP/AR solutions, through standalone banking products. Some have achieved a degree of success. But nobody has reimagined the entire stack. Two fundamental barriers have blocked every attempt.

1. Trust. Corporations can't hand their financial operations to a young company. They need banking licenses in multiple jurisdictions, regulatory compliance across every geography, deposit insurance, capital reserves, institutional credibility built over years. This is why people trust high-street banks - not because the software is good (it's terrible) but because these institutions have decades of history. Trust in banking has always been a function of time, not technology.

2. Integration. The sheer number of systems, the cost of connecting them, the time required. Building even a basic SMB-grade banking product cost several million dollars and eighteen months of work from a team of ten. Corporate-grade? Add years on top.

Corporate banking sucks. 58% of businesses already use fintech for core cash management - they're routing around their banks. 89% of finance teams still rely on Excel despite having planning software. Only 18% close their books in three days or fewer. And until recently, there was no realistic path to fixing it at the root.

Two things changed.

Unlock #1: Blockchain + stablecoins

Stablecoins settled $33 trillion in on-chain transactions in 2025 - more than Visa. That number alone should reframe how we think about payment infrastructure.

The market cap crossed $320 billion. B2B stablecoin payments surged 733% year-over-year. And critically, it's now regulated: the GENIUS Act created the first US federal framework for payment stablecoins, MiCA is fully operational across the EU. Stablecoins backed by government-approved reserves, operating under real regulatory frameworks - this is legitimate financial infrastructure.

Here's what this means for corporate banking: blockchain is the new payment rail, and stablecoins are the compliance layer on top. Together they allow you to build banking infrastructure for real companies - storing and moving money - without the enormous capital investment in licenses, multi-jurisdictional regulation, and reserves.

The economics are dramatic. A corporate cross-border SWIFT transfer costs $35-50 per transaction plus 2-4% in FX spreads and takes 1-5 business days. The same transfer on stablecoin rails costs under $1, settles in seconds, and runs 24/7. For a company making 500 international payouts a year, that's $450,000 annually on SWIFT versus $2,000-4,000 on stablecoins.

The trust question shifts. You're no longer asking companies to trust a startup. You're asking them to trust regulated infrastructure that has been validated, stress-tested, and increasingly backed by sovereign authority. The first fundamental barrier - trust and licensing - dissolves.

Unlock #2: AI and the speed of everything

The second unlock is speed - and it operates at two levels.

1. Development speed. The economics of building financial software have fundamentally changed. GitHub's controlled study found AI-assisted developers complete tasks 55% faster. McKinsey found coding tasks completed up to 2x faster. 25% of Y Combinator's Winter 2025 batch shipped with codebases that are 95% AI-generated - and that cohort is the fastest-growing in YC history. Over 30% of all new code at Google is now AI-generated. What used to represent five to ten years of integration work will take months.

2. Agentic replacement. The legacy systems corporations run today - the archaic accounting platforms, the treasury tools, the procurement systems - won't be modernized. They'll be replaced by agents. 59% of CFOs now use AI within finance functions, up from 37% in 2023. 57% of finance teams are already implementing or planning agentic AI. McKinsey envisions a near-future where one human supervises 20-30 AI agents managing end-to-end workflows. The platform that wins needs to be designed for agents from the ground up.

Picture how a payment works in this world: one agent creates the payment, another runs compliance checks, a third handles financial planning and cash management, a fourth manages liquidity and executes through the optimal rail at the optimal time. Each company brings its own processes, its own rules - and all of it integrates dynamically.

Why this can't be bolted on

There's a tempting counterargument: why not just add blockchain as one more payment rail and layer AI on top of existing products? Most fintech companies today are doing exactly that. Replacing OCR models with language models. Adding a copilot to existing dashboards. Letting you talk to a chatbot instead of clicking buttons.

The underlying product doesn't change. It's the same system with some additional AI capabilities.

The opportunity is to rethink and rebuild from scratch - AI-native, not AI-augmented. That means rethinking three layers:

1. The interface layer. People won't interact with financial systems through dashboards and data entry forms. They'll work through agents. The entire surface-level experience changes - charting tools, legacy UIs, click-through workflows are not needed when agents do the work and communicate results in entirely new ways.

2. The composition layer. Today's products integrate through APIs - login, credentials, push/pull data. An agentic architecture is fundamentally different. Different agents handle different parts of the workflow. They can be built by different companies, designed to replicate specific business processes, assembled dynamically. The system needs to let many agents collaborate on complex tasks - and those workflows are dynamic, not hardcoded. That's the core advantage of AI: the range of capabilities is fundamentally broader than any static workflow.

3. The trust and access layer. When agents from different companies collaborate on financial workflows, entirely new questions emerge around data access, permissions, and auditability. These problems don't exist in today's products and can't be solved by adding a feature.

Existing products aren't designed for any of this. It's not one more bank with an AI button. It's an entirely new kind of system.

Why companies will actually switch

Because the upside isn't linear - it's multiplicative. Duolingo, with the same headcount, now produces 4-5x as much content. They created 148 new language courses in one year versus 12 years for the first hundred. Shopify grew 20-40% while headcount dropped from 11,600 to 8,100.

The pattern scales. The larger the company, the more standardized routine processes it runs — the greater the surface area for optimization.

Technology adoption doesn't happen linearly. It happens in leaps. The entities that adopt first aren't slightly behind - they're far behind. From A straight to C. NFX calls this AI leapfrogging: the organizations with the most outdated systems have the strongest incentive to skip a generation entirely. In corporate banking, that describes almost everyone.

The opportunity

There's room for a new kind of platform - a financial operating system for companies. A single surface where all payments, all financial decisions, all treasury operations are managed by agents working over a shared ledger.

That ledger is blockchain, combined with whatever additional memory and state management the agentic ecosystem requires. On top of it: an extensible platform for building, deploying, and orchestrating financial agents. Each company customizes its own processes, rules, and integrations. The platform understands your business, configures itself, and runs.

The treasury management market alone is projected to grow from $6.6 billion to $16.3 billion by 2032. Agentic AI in financial services is growing at 43% CAGR. B2B payments represent $98 trillion in annual flows. No single platform pulls it all together today.

Why it should be open source

One of the clearest lessons from the history of technology: open platforms that create standards, earn trust through transparency, and allow extensibility end up dominating their categories. For a financial operating system to earn enterprise trust, it needs to be inspectable. Companies need to bring their own agents, organize their own workflows, run their own rules. Something like this should be open source.

Open Treasury - open-source financial platform that sets the standard for how agentic corporate finance works.

What remains hard

None of this is easy. Enterprise sales are enterprise sales - months-long cycles, sometimes years. Revenue builds slowly. That's a hard business from every angle.

The smart play isn't to sell to the Fortune 500 on day one. It's to find the right wedge - companies that move fast, understand the technology, have urgent needs and short decision cycles. Companies that prove this out first, whose success pulls the rest of the market forward. More on that in a future article.

---

Banks spent the last thirty years digitizing paper. The next ten will be about replacing the banks themselves - not with other banks, but with software that makes banking invisible. The infrastructure is ready. The window won't stay open forever.

If you see the same thing, I'd love to talk. Reach out.