A company’s treasury agent wakes up before the CFO does.

They’ve already pulled the latest e-invoices, matched them against open purchase orders, and projected the cash position for the next 45 days. There’s a gap in week three. Not a crisis; a €200k shortfall that will slow supplier payments if left alone.

So the agent queries a catalogue of financial products, compares three offers from two lenders, and selects a 30-day credit facility at the best available rate. By the time the CFO opens her laptop, the financing is arranged. She gets a notification. She approves it in one click.

Nobody filled out a form. Nobody called an account manager. Nobody waited five days for a credit decision.

This isn’t a prediction. It’s a description of infrastructure that exists today, being assembled by companies that understand what’s actually happening.

The interface was the moat. [Past tense.]

Software companies spent twenty years building moats out of user interfaces. If your product was complex enough, sticky enough, embedded deeply enough in someone’s workflow, they wouldn’t leave. Salesforce; SAP; the major banking apps.

Billions of dollars in enterprise value resting on an assumption that humans would always be the ones clicking through.

AI agents don’t click. They call APIs. And what they can’t call, they route around. Without announcing it, and without giving the incumbent a chance to respond. Products that aren’t queryable don’t lose customers in any visible way. They just get left behind.

PSD2 showed what happens when you open financial infrastructure to machines. European banks had to build APIs. The world didn’t end.

Financial services could be made programmatically accessible without losing what makes them valuable. The compliance, the risk management, the capital all stayed. What’s changed is who can access the service, and how.

MCP (the Model Context Protocol) is doing the same thing to all software at once, connecting AI agents to every business tool in use.

The companies that see it coming are already embedding credit, payments, and insurance directly into their products. Not as features, but as revenue lines.

Goldman isn’t waiting

In February 2026, Goldman Sachs announced plans to deploy autonomous AI agents, built on Anthropic’s Claude, to handle trade accounting and client onboarding. Not chatbots answering questions; agents resolving exception-heavy back-office workflows that previously required teams of people.

Goldman’s CIO has been explicit about where this goes: a hybrid workforce where AI agents are treated as virtual employees.

Goldman isn’t some startup making a bet. When it moves on something like this, it’s showing the industry where things go next.

The CFO is about to experience what CTOs just went through

Watch what happened to engineering over the last two years. The CTO didn’t disappear when developers got Copilot and Cursor and Claude. The role shifted. From writing code to directing agents that write code.

The best CTOs today spend less time in the weeds and more time making decisions that only a human with context and judgment can make. Their leverage multiplied. Their teams got faster.

The CFO is next. The treasury agent doesn’t replace the CFO. It does what the coding agent did for the engineer: takes over the execution layer so the human can move up to the decision layer.

Europe’s working capital gap is €4.3 trillion. Better services solve this.

European businesses need easier access to working capital. That’s not a market inefficiency waiting for a better bank. It’s an information problem.

The signals that predict a financing need are already there, sitting in ERP systems, payment flows, invoice data. They’re just not being read in real time and acted on.

A treasury agent reads them continuously. It identifies the gap before the CFO does, sources the right instrument for that moment in the cash cycle, and executes. The CFO approves. Then moves on to something that actually requires judgment.

Scale that across every SME in Europe and you’re not looking at a fintech opportunity. You’re looking at the entire flow of business credit, restructured around machines that never miss a signal.

The banks that win this won’t have the best branch network or the most intuitive app. They’ll have the most queryable product catalogue, structured and accessible to agents acting on behalf of businesses they’ve never met. The banks that spent the last decade building beautiful interfaces have a distribution strategy, not a moat.

And agents are about to make that distribution irrelevant.

The actual work of banking, credit risk, compliance, asset-liability management, doesn’t go away. It becomes more valuable. The heavy lifting is the moat. It always was.

What we’re building at Defacto

We didn’t set out to build a treasury agent. We set out to fix a specific problem: small businesses in Europe couldn’t get short-term credit quickly enough. Banks were too slow. The data existed in accounting systems, payment processors, and ERPs, but nobody was reading it in real time to make the necessary credit decisions.

So we built that. Fast, embedded, data-driven credit delivered through the platforms SMEs already use. And in building it, we ended up with infrastructure that sits exactly where treasury agents will need to operate.

Integrations with financial systems of record. A real-time decisioning layer. And the ability to read a business’s financial signals and act in minutes.

The treasury agent became the obvious next step.

The company that knows how to lend to a business, in the right moment, from the right signals, is exactly the company that should be identifying that moment in the first place.

The question isn’t whether agents will consume financial services. It’s whether you’ll be findable when they do.