Your Next User Won't Have Hands: Designing Fintech Flows for the AI Agents That Pay on Their Behalf
I've spent two decades designing money apps for people with eyes and thumbs. Both are about to become optional — and I've had to relearn what my job actually is.
Strahil Hadzhiev
AUTHOR
For twenty years, I designed for a body. That body is leaving the flow.
Let me tell you what I used to obsess over.
The exact weight of a confirmation button. The half-second of delay before a success animation, because too fast feels cheap and too slow feels broken. The precise moment in an onboarding flow where a nervous person decides to trust you with their salary. I've spent twenty years reading faces I'll never see — designing for the flicker of doubt, the thumb hovering over "Send," the tiny hit of relief when the money lands.
All of that craft assumes a human body. Eyes to read the reassurance. A thumb to make the choice. A nervous system to feel the fear I'm designing around.
And now, for the first time in my career, the thing paying with your customer's money has none of those.
It's an AI agent. It doesn't have eyes, so my typography is invisible to it. It doesn't have a thumb, so my beautiful tap targets are meaningless. It doesn't get nervous, so every micro-reassurance I've built my craft on lands on nothing. It reads structured data, follows a mandate, and moves money while the actual human is asleep or in a meeting or, as one payments executive put it, off doing something more interesting with their life.
This isn't a distant forecast. Amazon is testing a feature that buys things for you. OpenAI and Stripe wired checkout straight into ChatGPT, so the assistant doesn't just find the product — it buys it. Mastercard shipped "Agent Pay," giving trusted agents a kind of verified ID so a merchant knows it's dealing with a real customer's helper and not a fraudster's script. Visa built its own trusted-agent framework. PayPal opened a toolkit so developers can drop payments into agent workflows. The whole industry is quietly agreeing on the same strange new fact: the buyer is becoming software.
The honest reaction, if you've built your identity on craft like I have, is a small cold panic. If the user can't see anything I make, what am I even for?
I sat with that question for a while. Here's where I landed, and it's the reason I'm more useful now, not less.
The screen didn't disappear. It split in two — and both halves are pure design.
The mistake is thinking the interface vanished. It didn't. It broke into two moments, and both of them are the most human, highest-stakes design work I've done in years.
The first moment comes before the agent ever acts. A person has to hand it authority. They have to say, in effect, "you may spend my money, on these things, up to this much, for this long." That's a mandate. And right now, in most products, it's either nonexistent or buried in a settings screen a lawyer wrote.
Think about what that moment actually is. A human is granting a machine permission to spend without asking again. That is one of the most emotionally loaded interactions in all of finance, and almost nobody has designed it well yet. The good version feels like the best parental-controls flow you've ever seen — calm, specific, generous with examples. "Your agent can rebook flights under 800 euros. It can't touch your savings. It will always ask before anything over 200." The bad version is a checkbox next to a wall of text, and it will end in a support ticket, a chargeback, and a person who never trusts an agent again.
The second moment comes after. The agent has been busy. It spent, negotiated, subscribed, cancelled, re-booked — all while the human was living their life. Now that person opens the app and needs to understand, in one glance, what their machine did on their behalf. Was that charge mine? Why did it pick that vendor? Can I undo it? Can I kill this thing right now if it's gone wrong?
That control room does not exist in most fintech products today. It is, I'd argue, the single most important screen of the next ten years, and it's a blank canvas. It needs the clarity of a great bank statement, the calm of a good thermostat, and the one-tap panic button of an emergency brake. Designing it well is the difference between an agent that feels like a trusted employee and one that feels like a stranger with your card.
So notice what happened to my job. I stopped designing the middle — the checkout, the tap, the moment of purchase — because a machine does that now, headless and instant. And I moved to the bookends: the human granting trust, and the human auditing it. The two moments where a person's relationship with their money is most fragile. That's not less design. It's the part I was always best at, finally stripped of everything that was hiding it.
The industry even says it out loud now: treat the AI agent as a new kind of customer, and design for it as deliberately as you'd design for mobile. Fine. I'll design the rails and the data the agent consumes. But the work I'd stake my name on is the two places a human still stands — deciding whether to trust, and checking whether that trust was kept.
Everything technical changed. Everything human stayed exactly the same.
Here's the part that took me twenty years to earn the right to say, and it's why I think the fundamentals matter more in this era, not less.
Strip away the protocols and the verified IDs and the cryptographic mandates, and what's left is the oldest question in financial design: will you trust me with your money, and how do I make that trust safe? Agents didn't change that question. They raised the stakes on it. When a human made every payment by hand, a bad interface cost them a moment of confusion. When a machine makes a thousand payments on a vague instruction, a bad interface costs them their savings and their confidence in one bad night. One mischarged payment by an agent can sour a person on the whole idea for years. I've watched trust rebuild slowly enough with humans in the loop. With machines, it's more fragile, not less.
Which means all the unglamorous craft I've been practicing for two decades suddenly matters more. Clarity, because a confused human grants a dangerous mandate. Honesty, because an agent that quietly does more than the person expected is a betrayal at machine speed. Reversibility, because the first thing a frightened person needs is the certainty they can undo it. Restraint, because the temptation now is to let the agent do everything, and the designer's job is to insist that some decisions stay human. These aren't features. They're the whole thing. They always were. The checkout screen was just where they used to live.
And there's a new discipline on top of the old ones. When your user is an algorithm, you have to design for how it fails, not just how it succeeds. A human can be tricked, sure — but slowly, one screen at a time. An agent can be manipulated by a hidden instruction buried in something it reads, and act on it in milliseconds. So I design ceilings it can't exceed, a list of places it's allowed to go, a hard stop above a threshold where a human has to nod, and a revoke button a panicking person can find in one second without thinking. Designing the graceful failure is now as important as designing the happy path. Maybe more.
So here's what I actually believe, after twenty years and now this.
The teams that win the agentic era won't be the ones with the flashiest agent demo. Anyone can film a demo. They'll be the ones whose flows survive the first real dispute, the first hijacked instruction, the first frightened customer who needs to hit the brakes — and finds the brake exactly where a calm, experienced hand put it. That's not a job for a template or an afternoon of AI-generated screens. It's a job for someone who has sat with a nervous human and their money enough times to know precisely where the fear lives.
That's the work I do. Not the studio, not the process, not the deck — me, a designer who's spent twenty years learning where people flinch when their money's on the line, now doing that same work for a moment where the person isn't even in the room. If your product is about to hand a machine the keys to your customers' money, you want someone in the room who's designed for that flinch a thousand times.
Your next user won't have hands. Design for the human whose hands they're replacing. That's still the whole job. It always was.
Strahil Hadzhiev
AUTHOR
