Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)
Hey Fintech Nerds 👋,
I’m still out this week, so a shorter newsletter again. Thanks to everyone who reached out and sent your well-wishes 🙏
🧠 The Checkout is Dead Pt 2 is another rant I had ready to go. And it is possibly one of the most important concepts for anyone in agentic commerce to fully wrestle with. The checkout isn’t just dead. Its long-term future is entirely invisible.
Lots of news this week - super brief summary in the 👀 Things to Know section this week. Monzo is withdrawing from the US. Ramp’s CLI and stablecoins. Chime’s Prime (an actually incredible bunde).
🍸 I’ll be in Washington D.C. the week of the 13th April, and running a Fintech Nerdcon x Fintech Brainfood Happy Hour. And you can 👉 RSVP by clicking here 👈 if you’re in town! (Massive thanks to Ballad Spahr for sponsoring).
Here's this week's Brainfood in summary
📣 Rant: 🧠 The Checkout is Dead pt2: Invisible Commerce
💸 4 Fintech Companies:
Prava - Payments infrastructure for Agents
Signet - The Control Center for Agentic Payments
Conto Finance - Another Control Center for Agentic Payments
Halofy - The infrastructure for agentic commerce
📚 Good Read: Tether Is Not a Stablecoin Company
Weekly Rant 📣
🧠 The Checkout is Dead pt2: Invisible Commerce
I had an Aha moment when I saw a hackathon project last week. Someone built a parking AI agent that will pay whatever local parking authority you happen to be near based on your phone’s location.
In that moment, you never saw the checkout; you also never created an “intent” to check out. An event occurred in the world (you parked somewhere), and a payment was completed.
When I wrote the checkout is dead in May of 2025, it was about the beginning of the commerce experience moving into the LLM and the chat. Since then an alphabet of protocols ACP, UCP and AP2 have appeared, and the payments industry has begun to ship agentic checkout experiences.
But where we’re going, we don’t need checkouts.
The 5 Levels of Agentic Commerce
Stripe talks about the five levels of agentic commerce (borrowing from the 5 levels of autonomous driving).
Human chooses items - agents complete the purchase
Agent researches items - presents to them to the human - human chooses and agent buys.
Human delegates an action (e.g. buy coffee beans for under $20) and the agent completes the task and purchase
Agent manages complex buying tasks - replenishing inventory or managing subscriptions, enforced with policies
The agent anticipates needs and buys things
Notice what all five levels have in common? A checkout. The agent gets progressively more autonomous, but the purchase event is still visible, still a thing that happens. Even level 5 assumes the agent is reacting to some human intent.
Right now, we’re somewhere between level 1 and 2 of this list. But this list presupposed a human in the loop of purchases until level 5. It assumed a world model where commerce was visible and used the human web.
There’s just one issue with this whole thing.
Walmart said conversion rates for purchases made directly inside ChatGPT were three times lower than when users clicked through to its website.
Oh dear.
And it gets worse.
OpenAI confirmed it was phasing out Instant Checkout in favor of app-based checkout handled by merchants.
I often joke there are more agentic commerce protocols than transactions, and I keep making it because it’s true. And, I sort of get it. AI has that annoying habit of being useless until its suddenly dominant. And no payments company wants to get caught the wrong side of that trend.
But it’s clear agentic commerce is not working.
Maybe we’ve been thinking about this all wrong.
What if that mental model was missing the phase shift?
Like how when water becomes vapor, it starts to play by very different rules. When agents can write code and use computers, their capabilities expand dramatically.
Agents are automating life - Why not payments?
Last week I wrote about how my OpenClaw helped me do my taxes:
I needed to find all of my email receipts. Because my OpenClaw agent already has secure(ish) read-only access to my Gmails, it automatically stores them in a folder when they land.
So I had a package ready to send to my accountant. A CSV of descriptions and categories, already populated. It used my Google Maps history to produce a car mileage report. (I’m getting it the Xero API next).
Sure, I could have done this with an LLM. But I would have had to grant it access. Search for the pieces. Generate the prompt. Wait for the output. The magic of the agent is that it was just sitting there. It had already done the work in the background. I could even have it send it to the accountant without me but I wanted to review.
I’m, at best, a novice user of OpenClaw and an intermediate vibe-coder; what the real nerds are doing is staggering. Automating entire business processes and even companies.
My podcast producer, Pet, has an agent who takes the transcript of every episode and uploads it to TokenizedPod.com for search engines to index and reference.
There’s even open source tools like Paperclip for organizing your agents into an entire company of job roles.
Every month, agents swallow another category of knowledge work. And almost every one of those workflows eventually hits a moment where something needs to be paid for. A subscription. An API call. A vendor invoice. A domain renewal.
That's where they stop.
We already have one example of invisible payments at scale. Uber. You step out of the car and money moves. No cart, no confirm, no "pay now." But Uber had a cheat code — it owns both sides of the marketplace. It has the rider's card, the driver's location, the fare calculation. All one app.
The question is: what if every agent could do what Uber did, without being Uber?
Agentic Payments Need an Alternative to the Checkout
Uber works because it’s one merchant (Uber), that has the same app for the riders and the drivers. They have all the data they need to fully complete an invisible checkout. But having an agent per app could be exhausting.
The ideal state is you can spin up as many agents as you need, who can find and discover ways to pay, to take actions in the real world.
This creates a gap as I explained in 🧠The Intention Layer :
Every AI agent hits the same wall. Your agent is executing a task—researching a question, comparing prices, gathering data from multiple sources. It encounters an API that charges per call. A paywall blocking a document it needs. A service that requires payment to proceed. The agent stops. It can't pay. It asks you to intervene, or it fails silently, or it routes around the obstacle and delivers a worse result.
Checkouts and the human web don’t work for agents. We’re making them work better for agents, sure. But agents don’t want a better checkout, they want an alternative.
Agents don't read your ad. They query a registry, get structured results, and pick the best option in milliseconds. Your pricing page becomes a machine-readable header. Your sales funnel collapses into three HTTP calls: discover, authenticate, buy.
And that customer needs a way to pay that's as machine-native as it is.
Think about what agents use to navigate the web. Markdown. CLIs. APIs. The backroom infrastructure of the internet. Not flashy front ends. The payment layer needs to match. Simple. Elegant. Protocol-level.
Agents Need Their own Way to Pay
Last week, I had a small hand in helping to build one way they could do that..
With the launch of the machine payments protocol (MPP) last week, I saw the beginning of what my Tokenized Co-host Cuy Sheffield calls command-line commerce.
Agents can call the Machine Payments Protocol from a command line, and with a simple HTTP request (GET /resource), a server can respond and accept payments via credit cards, wallets, or stablecoins.
Super simple flow
At first, I thought the main uses for this would be the obvious ones. As I wrote last week:
A research agent pays for access to proprietary datasets—clinical trial results, satellite imagery, patent filings.
A development agent purchases compute and testing infrastructure, spins up environments, runs integration suites, tears them down
A video editing agent purchases stock footage, music licenses, and rendering credits
A delivery routing agent pays for real-time traffic data, weather updates, and route optimization APIs, re-purchasing updated feeds every few minutes as conditions change
And all of those use cases are live and working now in MPP, with AI-native, stablecoin-skeptic, developers starting to use them and experiencing genuine magic.
But what surprised me is how many people had their agents buy them a sandwich.
And look, there’s also x402, and Google is releasing new agentic protocols every week. I don’t know if MPP will be the thing, but something like it will.
Which brings me back to where we started. The parking app. The problem with the parking app today is every app has its own checkout, vendor, and payments provider. Getting agents to use the internet is hard, getting them to use custom on-device, often poorly implemented apps is even harder.
Agents will make payments invisible
Walmart's decision to pull out of checkout chatbots proves something important. We’re doing it wrong. Conversion collapsed 66%, because we took something highly optimized for humans and buried it inside a chatbot.
That was the wrong starting point for agentic commerce.
Agents don't want a checkout. The parking app example I gave you shows a world where an event happened — you parked — and a payment was completed. Surely that’s a better world, where you can just say to your agent, “pay for my parking” or delegate all parking to it forever. And it just happens. You don’t have to click anything.
That's the phase shift.
The parking agent never got an intent from a human. The world itself triggered the payment. The agent inferred the rest.
Like when water turns into vapor. Liquid has a surface you can see and touch. Vapor is everywhere and nowhere. Checkout payments are visible, and optimized for human conversion. Agent payments are invisible. Like air.
New payments models need new security models
Invisible payments are also invisible risks.
When your agent spends while you sleep, who audits the trail?
When an agent is compromised, who's liable?
When your agent buys something faulty from a merchant, how are returns handled?
The solution for much of this from the existing card networks has been to make a direct connection between the agent creator (e.g. OpenAI) and the human who owns the underlying card (i.e. you). But in a world where agents have their own budgets, wallets or accounts, things get a lot messier. Especially if those agents are running locally, on open weight models and platforms like OpenClaw.
Agent identity or Know Your Agent (KYA) is the next big protocol question after payments and one many are fervently working on.
(Which I’ve written about before in the Wallet Wars series Part 2: Wallets become the home of AI agents. And Part 3: The coming bot authentication crisis.)
That infrastructure is being built. But it's early. And the gap between "agents can pay" and "agents can pay safely" is where the next crisis lives if we don't close it fast.
The checkout is dead. The payment is invisible.
The first viral use case for MPP wasn't enterprise procurement. Wasn't API monetization. Wasn't any of the obvious things.
People had their agents buy them a sandwich.
And sure you could call that a hackathon quirk. But my brain saw something else.
I saw a world without checkouts.
And maybe, the bug in the whole agentic commerce conversation to date.
We shouldn’t be trying to make better checkouts. We should be replacing them with invisible commerce. Led by agents with identities.
Prioritize accordingly.
ST.
4 Fintech Companies 💸
1. Prava - Payments infrastructure for Agents
Prava helps agent builders (e.g. Merchants, support agents) accept card payments securely, using the Visa Intelligent Commerce tokens. This works with any PSP, and avoids months of integration work directly with Visa.
🧠 Think of this as card on file but for agents. Agent on file. I think of this as like VGS or Basis Theory for agents (which, incidentally, those companies also offer). Imagine a shopping agent on Amazon, helping you find things to buy for your upcoming Fintech Conference. In the chat experience, you might surface some ideas. Prava turns those products into something buyable. The key thing here is that the agent doesn’t have to be ChatGPT or Gemini. It’s everyone elses.
2. Halofy - More infrastructure for agentic commerce
Halofy aims to support any agentic commerce protocol (ACP, UCP, x402), any merchant and any agent from a single infrastructure. It handles the plumbing between different payment rails and provides a single vault for delegating auths, agent identity, and payment completion.
🧠 If that sounds complex, consider this metaphor. Today, companies like VGS and Basis Theory help merchants manage “tokens” in vaults, so you can store your card on file with them, prove your identity, and pay with Apple or Google Pay through the various card networks. Agents multiply that complexity, as does the proliferation of new protocols. Halofy is trying to aggregate and orchestrate all of that.
3. Signet - The Control Center for Agentic Payments
Signet helps companies observe every agent, on every protocol and payments rail in real-time. It monitors every transaction agents make, and enforces pre-defined rules and policies. It audits everything, will push questionable payments to manual review and can kill any agents access in 500ms (especially important for 3rd party agents).
🧠 If you’re buying a 3rd party agent to do things for your company, you want to know exactly what it did. This assumes a world where companies will have multiple agent staff from multiple providers, and that’s not a crazy idea. The most traction I’ve seen in agentic payments is in B2B.
4. Conto Finance - Another Control Center for Agentic Payments
Conto is building a platform that helps build control around agents' spending, enforce policies, and provides real-time visibility into their activity. It also maps their relationships with vendors, suppliers, and any 3rd party. It also builds a “trust collective” of known agents who have interacted in their network previously.
🧠 If your agents are sourcing suppliers and paying vendors, you want to control that. I nearly called this “Ramp for agents,” but I’m pretty sure Ramp will try to be that. And this is something else. It’s agent orchestration for payments. Have them do the right thing in the right order. It places policy in the “payment path, not the prompt.” Right idea, but wow, does this look like Claude Code created a Vercel page for them.
Things to know 👀
🧠 Monzo has 15m UK customers (half the adult population). They’re the dominant UK bank for new accounts, and now, child accounts. If they could be that in Europe, which is arguably an easier expansion point. They’d be much more effective. Meanwhile Revolut and Nubank get to duke it out for, can a non-US Neobank ever crack the USA?
2. DeFi platform drift was hacked for $286m, most likely by the DRPK.
🧠 I remain the disappointed dad of crypto. It’s still too easy to hack, fund adversarial regimes, and commit crimes. The existing financial system’s AML approach isn’t the answer, but DeFi is now getting closer to mainstream financial markets (with CLARITY Act), and consumers via embedded yield. There’s still a long way to go.
3. My favorite product launch this week was Ramp’s CLI
🧠 The CLI and Claude code helped me set all of my expenses on autopilot across 5 separate Gmail accounts (I’m an edge case user).
4. Ramp also launched stablecoin financial accounts. Users can now Hold stables. Earn yield. Pay vendors in USDC. Pay off the Ramp card with stables, without any complex integrations or having to handle stablecoins directly.
🧠 Stablecoins are becoming regular dollars. Ramp threaded USDC through the existing spend management stack. The CFO sees the same controls whether a payment is USD or USDC.
5. Chime Prime, a new feature unlocked free at $3K+/mo direct deposit. 5% cash back on 1 MCC (groceries, gas, dining, bills, rideshare, or travel; up to $1,500/mo spend), 3.75% APY on savings, Priority Pass + luxury hotel perks + concierge, fee-free overdraft, early pay up to $500, fastest loan access, 24/7 support, onyx metal card.
🧠 Helluva perk for switching your paycheck! Most companies bundle the metal card behind a subscription product. Chime Prime is free. Their mission to be ultra-affordable, lean, and competitive really does create intense price competition in the market.
Good Reads 📚
“Most people's understanding of Tether is three to five years out of date. The crypto press still treats it as a stablecoin issuer with trust issues. The mainstream press still treats it as a possible fraud.
What I found is a company that generated over $10 billion in profit last year with just 300 employees (with plans to add 150 more), holds more US Treasuries than Germany, and is quietly building a technology conglomerate funded entirely by the interest on other people's dollars.”
🧠 This is one of the most comprehensive, unbiased, and 360 takes on Tether I’ve seen. Not FUD, but not a bullpost either. Tether is too big to ignore, and if you’re curious about stablecoins, you should read this piece to understand it better.
🧠 There’s a future world where geopolitical uncertainty and more wars lead to more failed states. Where stablecoins and decentralized infrastructure become much more relevant. In that world, Tether’s other investments start to make more sense.
Tweet of the Week 🕊
That's all, folks. 👋
Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)
Want more? I also run the Tokenized podcast and newsletter.
(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.
(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.
(3) Any companies mentioned are top of mind and used for illustrative purposes only.
(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held
(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out
