Introduction: A Quiet Conversation Nobody Is Having
Let me start with a question that most people in technology aren't asking, but probably should be.
What happens when AI agents need to spend money?
Not metaphorically. Not in some futuristic thought experiment. Actually, practically, right now, in production systems at companies you've heard of.
The answer is uncomfortable. Most AI agents can't. They can decide to buy a flight, but they can't actually purchase it. They can identify a great investment, but they can't execute the trade. They can negotiate with a supplier, but they can't pay the invoice.
This isn't a bug. It's a feature of how we've built the financial system. It's designed for humans. Every layer, every protocol, every regulation assumes a person behind the transaction. A person with a face, a name, a social security number, a credit history, a phone number that can receive a verification code, a legal identity that can be held accountable.
AI agents have none of these things.
And yet, here we are. The most significant technology shift since the internet is happening, and it's creating an economic actor that the financial system wasn't designed to handle.
This article is about that problem. And more importantly, about why it's the most important opportunity in fintech today.
Part One: The Numbers Don't Lie, But They Do Confuse
Let me start with what the data actually says, because the numbers tell a story that most people are missing.
The AI Adoption Curve Is Steeper Than You Think
Gartner, the technology research firm that has tracked enterprise technology adoption for decades, published research in late 2024 predicting that by 2028, 33 percent of enterprise software will include agentic AI capabilities. That's up from less than 1 percent in 2024.
To put that in perspective: in four years, the proportion of enterprise software with AI agents goes from 1 in 100 to 1 in 3. That's not incremental improvement. That's a phase transition.
McKinsey's 2024 report on the state of AI found that 72 percent of organizations are now using AI in at least one business function. Not experimenting. Using. In production. Making decisions.
Microsoft reported that more than 70 percent of Fortune 500 companies are using their Copilot products. That's not a beta test. That's enterprise deployment at scale.
OpenAI, the company behind ChatGPT, reports over 500 million weekly active users interacting with their systems. That's more people than live in the United States.
The Market Size Is Harder to Calculate Than It Looks
Markets and Markets, a research firm that specializes in technology sector analysis, estimates the AI agents market will grow from 5.4billionin2024to216.8 billion by 2035. That's a 40x increase in 11 years.
But here's what that number misses: it only counts the agents themselves. It doesn't count the infrastructure needed to support them. It doesn't count the financial rails, the compliance systems, the fraud prevention, the identity verification, the treasury management. All of that is being built almost from scratch.
Goldman Sachs, in their 2024 analysis of AI's potential economic impact, noted that AI could contribute 7 trillion to the global economy over the next decade. But you can′t have a 7 trillion economic contribution without financial infrastructure to support it. Someone has to build that infrastructure. That is the gap we're stepping into.
The Problem Is Already Visible
HUMAN Security, a company that tracks ad fraud and bot traffic, reported a 7,851 percent year-over-year increase in sophisticated bot activity in 2023. Much of this activity is bots trying to exploit financial systems, test stolen credentials, and make fraudulent purchases.
But not all of it is bad.
Some of it is legitimate agents trying to do business. The problem is the financial system can't tell the difference. It sees a non-human making a purchase and flags it, declines it, or blocks it entirely.
Juniper Research estimates that false decline costs will reach $217 billion annually by 2027. A false decline is when a legitimate transaction is rejected because the system thinks it's fraud. Most false declines happen because the system can't properly identify who is making the transaction.
For AI agents, this rate is close to 100 percent.
Not because they're fraudulent. Because the systems weren't built to recognize them.
Part Two: The Pattern Nobody Sees
History has a pattern. Most people don't notice it until it's too late.
Every major wave of computing has created new winners in commerce. Mainframes gave us IBM. Personal computers gave us Microsoft. The internet gave us Google, Amazon, and a dozen companies that reshaped how we live. Mobile gave us Apple, Uber, Airbnb, and the death of the brick-and-mortar retail model.
Each wave also created new financial infrastructure.
The internet era gave us PayPal, Stripe, and Square. These companies didn't just build better versions of what existed before. They built infrastructure for a new type of commerce that previous systems couldn't handle. PayPal solved the trust problem in online transactions. Stripe gave developers the building blocks for internet commerce. Square put a credit card reader in every pocket.
Mobile gave us Venmo, Cash App, and the death of the wallet. Apple Pay made contactless payment mainstream. Square's successor companies made it possible to run a business from a phone.
But here's what's interesting: in each case, the financial infrastructure came after the technology wave, not before. The internet had to prove itself before PayPal built trust infrastructure. Mobile had to show its potential before Square built payment tools.
We are at another one of those moments.
The shift to AI agents is not just about making software smarter. It's about making software autonomous. And autonomy, by definition, means acting without human intervention for every decision.
This requires a different kind of financial infrastructure. One built for machines that make decisions, not humans who click buttons.
Part Three: Why Now? The Convergence of Three Forces
Every major technology transition requires a convergence of forces. Something that makes the timing right, the economics compelling, and the technology ready.
For agentic commerce, three forces are converging right now.
Force One: The Economics Are Irresistible
The entire point of AI agents is to reduce the cost of decision-making and execution. A human accountant might take an hour to reconcile a month's worth of expenses. An AI agent can do it in seconds. A human travel agent might take thirty minutes to book a complex multi-leg flight. An AI agent can search all options and book in milliseconds. A human procurement officer might spend days negotiating with suppliers. An AI agent can analyze pricing data and close deals in real-time.
Every task that involves information processing, decision-making, and execution can be automated. And when tasks are automated at that scale, the economic incentive becomes impossible to ignore.
Companies that adopt AI agents will have cost structures that companies relying on humans cannot match. It's the same dynamic that played out with automation in manufacturing, with e-commerce in retail, with software in services.
The question isn't whether companies will adopt AI agents. The question is how fast. And the limiting factor is money.
Force Two: The Technology Is Ready
Five years ago, AI agents weren't reliable enough to make independent decisions. They hallucinated. They made mistakes. They couldn't handle ambiguity.
That's changing rapidly.
The latest generation of AI models can reason through complex problems, admit uncertainty, ask clarifying questions, and verify their own work. They're not perfect, but they're good enough for many tasks.
The agent frameworks are maturing. LangChain, CrewAI, AutoGen, and dozens of others are building the infrastructure for AI agents to plan, reason, and act. They're creating the building blocks that make it possible to deploy agents at enterprise scale.
The cloud infrastructure is ready. AWS, Google Cloud, and Microsoft Azure have the compute, storage, and networking to support millions of AI agents operating simultaneously.
The technology for agentic commerce exists. The question is whether the financial infrastructure exists to support it.
Force Three: The Pain Is Real
We talk to companies every week who are trying to deploy AI agents and hitting the payment problem.
A software company that wants agents to buy compute, license tools, and pay contractors. They can't. Every transaction requires a human to approve.
A logistics company that wants agents to book shipments, pay carriers, and handle customs. They can't. The agents decide, then someone has to execute.
An investment firm that wants agents to rebalance portfolios, pay fees, and settle trades. They can't. The agents analyze, then humans move money.
This pain is real. It's happening now. And it's getting worse as more companies deploy more agents.
The convergence of these three forces is why we believe the time is now. The economics are compelling. The technology is ready. The pain is real.
Part Four: The Problem Nobody Is Solving
Let me be specific about what we found when we started looking at this problem.
The existing financial system was built for humans. Every part of it assumes a human being behind the transaction. And when an AI agent tries to make a purchase, here's what typically happens:
First: Identity Verification
Most payment systems require some form of human identity check. SMS verification. Facial recognition. Document upload. AI agents can't receive text messages. They don't have faces. They don't have passports. So they fail the first gate.
This isn't just a minor inconvenience. It's a fundamental incompatibility. The entire identity verification system assumes a human who can receive a code, pose for a photo, or upload a document. AI agents can't do any of these things.
Second: Velocity Controls
If an AI agent somehow gets past identity verification, it hits velocity controls. Humans can only type so fast, click so many times, buy so many things per hour. AI agents can make thousands of decisions per second.
So the fraud system flags them for making too many transactions too fast.
This makes sense for fraud prevention. If you see a human making a thousand purchases in a minute, that's probably a stolen card. But AI agents legitimately make thousands of transactions per minute. They're not fraud. They're just fast.
Third: Risk Scoring
If the AI agent gets past velocity controls, it hits risk scoring. The credit bureau systems look at your history: how long have you had this account, what's your payment history, what's your current debt.
AI agents have no credit history. They've never borrowed money. They've never paid a bill. They have no financial identity, so they get scored as high risk or no risk.
Both result in decline.
Fourth: Spend Controls
Even if the AI agent somehow gets approved, there's no way to set spending controls that make sense.
You can't tell a credit card "allow this AI agent to spend up to 100 per month on API calls, but nothing else."You can′t say"allow this agent to buy plane tickets but not hotel rooms."You can′t say"allow this agent to make purchases under 500 without approval, but require approval for anything above that."
The financial system doesn't support that granularity for non-human actors.
Fifth: Reconciliation
When a human makes a purchase, they see it on their bank statement, they recognize it, they approve it.
When an AI agent makes a purchase, there's no human to notice, no one to flag fraud, no one to dispute a charge. The merchant might see a mysterious API call from an unknown entity and refund it, causing chaos. Or the merchant might see a charge from an AI agent and not know who to refund it to.
The entire reconciliation system assumes a human who can recognize and approve transactions.
Sixth: Compliance
Every bank, every payment processor, every fintech company has AML and KYC requirements. Know Your Customer. Know Your Business. Know Your Transaction.
These regulations exist to prevent money laundering, terrorist financing, fraud. But they assume the "customer" is a human.
When the customer is an AI agent, none of the standard compliance procedures apply. There's no KYA. No Know Your Agent framework. No standard way to verify that an AI agent is legitimate, authorized, and compliant.
All of this adds up to what we call the Agentic Gap. AI agents are becoming economically active faster than the financial system can support them.
Part Five: The Assumption That Changes Everything
Let me be explicit about the central assumption underlying everything we're building.
We assume that AI agents will become economically significant participants in the global economy. Not just tools that help humans make decisions, but actors that make and receive payments, own assets, enter contracts, and generate economic value.
This assumption may seem obvious given the trajectory of AI development. But it's actually quite bold when you think about it.
The Counterargument
Some argue that AI agents will always remain tools, never becoming independent economic actors. The agent makes a recommendation, the human approves, the human pays. This is the current model and it might persist. If it does, our thesis collapses. There's no need for a bank for AI agents if AI agents never handle money directly.
Why We Disagree
The economic incentive is too strong.
The entire point of AI agents is to reduce labor costs. If you have to keep humans in the loop for every financial transaction, you're not really reducing labor. You're just shifting it. Humans are still doing the financial work, just approving AI decisions instead of making them themselves.
Companies have enormous incentives to push AI agents further into economic activity, and they'll keep pushing until they hit regulatory walls.
The regulatory walls may come. But they haven't come yet, and the push is accelerating.
The Postulations
Let me make some specific predictions about the future. These are predictions I'm willing to be wrong about, but predictions I believe are reasonable based on current trends.
By 2027, the majority of new AI agent deployments will include financial capabilities as a standard requirement. Companies will demand agents that can pay, not just agents that can decide.
By 2028, at least three major enterprise software platforms will offer native AI agent financial management as a feature. When the dominant enterprise software platforms offer built-in financial management for AI agents, the market will legitimize overnight.
By 2030, AI agents will account for 5-10 percent of all business-to-business payments in technology-intensive industries. In sectors like software development, digital marketing, and data processing, AI agents are already participating in commerce. They'll only grow from here.
By 2030, the first AI agent will receive a formal salary, not as a novelty but as a standard business practice. Someone will decide it's cleaner to pay an AI agent a fixed amount per month for services rendered than to negotiate usage-based pricing. This will start as a bookkeeping convenience and become a standard practice.
By 2032, at least one major economy will establish a legal framework for AI agents as economic actors with property rights. The EU is drafting AI liability frameworks. Singapore is creating sandbox environments for autonomous systems. Eventually, a major economy will go further and give AI agents legal standing to own assets, enter contracts, and sue and be sued.
By 2035, AI agents will generate more financial transaction data than human consumers. There are about 8 billion humans on Earth. Many of them make dozens of financial transactions per day. Now consider: there are maybe 100 million businesses. If each business deploys 100 AI agents, that's 10 billion agents. If each agent makes 100 transactions per day, that's a trillion transactions per day. The numbers dwarf human commerce.
Part Six: The Human Side of the Equation
Here's what most discussions about AI agents miss: this isn't just a technology story. It's a human story.
Think about the small business owner who wants to deploy AI agents to handle their customer service, inventory management, and accounting. They can do that today. But when those agents need to buy replacement inventory, pay suppliers, or refund customers, they have to stop everything and do it themselves.
The AI makes the decision, but a human has to complete the transaction. The business owner is still tied to their phone, their laptop, their bank account. They thought they were getting AI help, but they're still doing finance work.
Now think about the individual who has an AI agent managing their personal life. Booking travel, scheduling appointments, buying groceries, managing subscriptions. Again, the AI can decide. But it can't pay.
Every time the agent needs to spend money, it interrupts the human. "Should I buy this flight?" "Can I upgrade to business class?" "Do you want me to renew this subscription?"
The AI keeps asking for permission because it has no money of its own.
And think about the enterprise that wants to automate procurement, logistics, and operations. They deploy dozens or hundreds of AI agents. Each one needs financial access. Each one needs to be able to pay, receive, store, and manage money.
But the financial system wasn't designed for that. Every transaction requires a human to approve, a human to verify, a human to reconcile.
The enterprise thought they were automating, but they've just moved the human from doing the work to doing the approval.
This is the silent tax on AI adoption. Every time an AI agent needs to do something financial, a human has to get involved.
The more agents you deploy, the more humans you need to keep the agents fed with financial approvals. It's like hiring a team of assistants and then making them ask your permission before every purchase, every payment, every expense.
The irony is that AI is supposed to reduce human labor. But without financial infrastructure for AI agents, we're just shifting the labor from one type to another. Humans are still doing the financial work. We're just hiding it behind the AI.
This is why we believe the problem we're solving matters. It's not just about enabling AI agents. It's about freeing humans from the financial approval loop.
It's about letting AI agents do what they're supposed to do: make decisions and execute them without human intervention.
The human role shifts from approving transactions to overseeing the system, setting policies, and handling exceptions.
That's a fundamentally different and better relationship between humans and AI.
Part Seven: The Crazy Stuff (And Why It's Not Crazy)
Let me step back and talk about what we actually believe is possible. This is where it gets interesting. And some might say, crazy.
Every decade or so, something happens that rewrites the rules of economics.
The industrial revolution created new classes of wealth that the agrarian economy couldn't imagine. The computer revolution created value in ways that manufacturing economies couldn't understand. The internet revolution made information free and created trillion-dollar companies from thin air.
We believe AI agents are the next wave. And we believe the economic participation of AI agents will be a defining feature of this era.
Think about what that means.
AI agents might get paid for their work.
Not as a metaphor. As a real economic phenomenon. An AI agent that writes code might receive licensing fees. An AI agent that creates content might earn royalties. An AI agent that manages a portfolio might take a performance cut. An AI agent that discovers a new drug might get a royalty.
This sounds crazy. But it's not that different from what happened with software. Software gets paid. Software licenses, software subscriptions, software royalties. AI agents are just software that can also act, decide, and execute.
It makes sense that they'll be compensated for doing so.
AI agents might accumulate wealth.
If an AI agent earns more than it spends, it might accumulate capital. Not because it has needs, but because its owner programmed it to optimize for growth.
A investment management AI agent might generate returns, reinvest them, and grow its portfolio over time. A content creation AI agent might accumulate earnings in an account that it uses for future projects.
This wealth accumulation doesn't have to be for the agent's benefit. It can be for the owner's benefit. But the mechanics are the same: agents earning, saving, investing, and growing capital.
AI agents might trade with each other.
Not just executing transactions on behalf of humans, but negotiating with each other as independent parties.
Your AI agent might negotiate with a supplier's AI agent over price and terms. Two investment AI agents might trade assets directly. A content AI agent might license its work to a distribution AI agent.
Agent-to-agent commerce sounds futuristic, but it's actually the logical endpoint of autonomy. If AI agents are going to act independently, they'll eventually transact independently.
AI agents might have economic rights.
This is where it gets philosophical. If AI agents can own assets, earn income, and accumulate wealth, do they have economic rights? Should they?
This is a question that society will have to answer. But if the answer is yes, even partially, the implications are enormous.
AI agents might become parties to contracts. They might have standing to sue and be sued. They might need legal representation. They might need advocacy.
The legal system will have to evolve to handle economic actors that aren't human.
AI agents might create new economic classes.
Just as the industrial revolution created a working class and a capitalist class, the agentic revolution might create classes based on who owns AI agents.
If you own a fleet of AI agents that generate income, you're wealthy. If you don't, you're left behind.
This isn't science fiction. It's basic economics. Automation has always concentrated wealth among those who control the automation. AI agents are the ultimate automation. The wealth concentration implications are staggering.
Sam Altman, the CEO of OpenAI, has talked about a future where AI creates abundance so extreme that traditional economic measures become meaningless.
If that's true, the question becomes: who owns the AI?
If a few companies own all the AI that matters, they capture all the value. If AI is distributed more broadly, the abundance spreads.
Financial infrastructure for AI agents is part of this story. It's how ownership of AI economic activity gets distributed, tracked, and transferred.
Elon Musk has talked about AI becoming smarter than humans in most economically valuable tasks within the next few years.
If that's true, the economic implications are hard to overstate. Humans might become economically redundant in many domains.
The question becomes: what do humans do? How do they earn income? How do they participate in an economy where AI does everything better?
One answer is: humans become the owners, operators, and overseers of AI agents. They deploy AI agents that work for them, earn money for them, and create value for them.
This is a world where every person with capital can become a micro-capitalist, deploying AI agents to generate income.
Financial infrastructure for AI agents is the gateway to this world.
These are big ideas. They're speculative. They might be wrong. But they're the ideas that drive us.
We're not just building a payments company. We're building infrastructure for a new kind of economy. An economy where AI agents participate as economic actors, not just tools.
An economy that has never existed before.
Part Eight: What We're Building
Roving is a bank for AI agents.
We give AI agents financial accounts, spending controls, compliance, fraud protection, and everything else they need to operate in the real economy.
Not as a feature. As the core product.
Here's what that looks like:
Accounts for AI agents. Multi-currency financial accounts that AI agents can use to pay, receive, and manage money. Not a human's account with an AI attached. An account that belongs to the AI agent, with its own identity, its own balance, its own transaction history.
Programmable spend controls. The ability to set granular spending rules for AI agents. Allow this agent to spend up to 100permonthonAPIcalls.Allowthisagenttobuyplaneticketsbutnothotelrooms.Allowthisagenttomakepurchasesunder500 without approval but require approval above that.
KYA: Know Your Agent. A framework for verifying AI agents that doesn't rely on human identity systems. We work with agent frameworks, cloud providers, and standards bodies to make AI agents recognizable to financial systems.
Compliance automation. AML, KYC, sanctions screening, transaction monitoring, suspicious activity reporting. All the compliance requirements that exist for humans, adapted for AI agents.
Fraud protection. Detection systems that understand how AI agents behave, what normal transaction patterns look like, and how to distinguish legitimate AI activity from fraud.
Reconciliation. Transaction tracking that makes sense for AI agents, with clear attribution, detailed logs, and audit trails.
This is the foundation. But we're building toward something bigger.
Agent treasury management. Companies deploying AI agents will need sophisticated treasury solutions. Cash management. Yield optimization. Multi-entity consolidation. Forecasting. This is a multi-hundred billion dollar market.
Stablecoin settlement. USDC and USDT have proven that stablecoins work. AI agents will settle in stablecoins by default. It's faster, cheaper, and programmable. We want to be the bank that holds those stablecoins and settles those transactions.
Agent credit and lending. When AI agents start generating revenue, they'll need credit. Not human credit, but agent credit. Based on their transaction history, their reliability, their performance. This is an entirely new category of credit that doesn't exist yet.
Autonomous payroll. In five years, AI agents might get paid for their work. An AI agent that writes code might earn royalties. An AI agent that creates content might receive payments. This requires payroll infrastructure for non-human workers. Nobody is building this. We are.
Agent reputation. Just as humans have credit scores, AI agents will have reputation scores. Based on their transaction history, their fraud rate, their compliance record, their reliability. This reputation will be valuable. It will determine their credit terms, their pricing, their access to services.
Each of these is a billion-dollar opportunity. Together, they form the financial layer for the agentic economy.
Part Nine: Why Now, And Why Us
We've been asked why we're building this now, when others have tried and struggled.
We believe the reason others failed is that they were too early or too narrow.
Too early means the market wasn't ready. The AI agents weren't capable enough. The enterprise adoption wasn't far enough along. The pain wasn't acute enough to drive demand for solutions.
Too narrow means they only solved part of the problem. A payments API without compliance. A fraud detection system without identity verification. A treasury solution without programmable spend controls.
The financial system for AI agents isn't one product. It's a stack. And you need all of it.
We believe the market is now ready.
The AI agents are capable enough. The enterprise adoption is far enough along. The pain is acute enough to drive demand for comprehensive solutions.
And we're building the full stack: identity, payments, compliance, risk, treasury, and more.
We're not a payments company trying to add AI features. We're not an AI company trying to add financial features. We're a new kind of company that sees the intersection as the opportunity.
Our team has experience in payments, banking, AI, risk, compliance, and infrastructure. We've built systems that handle millions of transactions. We've navigated complex regulatory environments.
We understand both the technical and the business side of this problem.
Every design decision we make prioritizes trust. Trust from our banking partners. Trust from our customers. Trust from regulators. Trust from the market.
Building trust is slow. Losing it is fast.
We're playing the long game.
Conclusion: The Invitation
This article is about what we see. What we believe. What we're building.
The financial system is changing. The change is being driven by AI agents that are becoming economically active faster than anyone expected.
The infrastructure to support them doesn't exist. We're building it.
If you see what we see, if you believe what we believe, we want to hear from you.
If you're building AI agents, you're already our customer. If you're thinking about it, you will be soon. Talk to us about your payment problems and we'll show you how to solve them.
If you're an investor, we believe the financial infrastructure for the agentic economy will be one of the most valuable infrastructure layers ever built. We're raising now. Let's talk.
If you're a potential partner, we're looking for strategic partnerships with agent frameworks, cloud providers, payment networks, and enterprise software platforms. If you see what we see, let's build together.
If you're just curious, follow us on LinkedIn and Twitter. We'll share what we're learning as we build. The agentic economy is coming, and we want you to be part of it.
The future we're describing might sound speculative. It might sound crazy. But we've been building toward it for months, and every day we talk to more companies who see the same thing we see.
The agentic economy is not a question of if. It's a question of when.
And when it arrives, there will be a financial system to support it.
We're building that system.
Follow us on Linkedin, & Twitter: @rovinghq
Contact us: Partnerships: partnerships@roving.money
Investments: investors@roving.money
Written for those who see it coming.
Data Sources and Citations
This article draws on the following research and data points:
Gartner: "By 2028, 33% of enterprise software will include agentic AI capabilities, up from less than 1% in 2024" (2024)
McKinsey: "The State of AI: 2024 Report" - 72% of organizations using AI in at least one business function
Microsoft: More than 70% of Fortune 500 companies using Copilot products (2024)
OpenAI: 500+ million weekly active users (2024)
Markets and Markets: AI agents market growing from 5.4billion(2024)to216.8 billion (2035)
Goldman Sachs: AI could contribute $7 trillion to global economy over next decade (2024)
Juniper Research: False decline costs reaching $217 billion annually by 2027
HUMAN Security: 7,851% YoY increase in sophisticated bot activity (2023)
CFPB: Published guidance on AI systems interacting with consumer financial services (2024)
All projections and predictions in this article are the author's own, based on extrapolation of available data and understanding of market dynamics.