The checkout page is dead.

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)

Come meet me, and your favorite Fintech leaders, operators, and founders in Miami at Fintech Nerdcon 👇

Hey Fintech Nerds 👋

Google just killed the checkout.

Well actually. Everyone doing agentic commerce is slowly killing it. Commerce is moving upstream to the chatbot. Merchants and their processors will revolt. It upends the payments ecosystem, and while volumes are low today, AI has an annoying habit of being gradually then sudden. Your Rant this week. 📣 The Checkout Page is Dead.

On the Tokenized Podcast, talking about Squads (one of our 💸 4 Fintech companies this week) it struck me. We talk a lot about AI-first fintech, but the reality is the new era is stablecoin-first fintech. And these two trends are converging. I’ll be in New York next week for Stablecon, and you should grab one of the last few remaining tickets if you want to get your head around this properly.

Oh, and Bitcoin hit another all-time high while you weren’t looking.

If you’re a Gmail user, you could be missing 50% of Brainfood. Click Read Online to see the whole thing:

Here's this week's Brainfood in summary

📣 Rant: The checkout page is dead.

💸 4 Fintech Companies:

1. Squads - Build fintech programs on stablecoin rails

2. PayOS - The wallet for Agents

3. Caten Labs - The AI-first Financial Institution

4. Zinc - A Retail API for Agents

👀 Things to Know:

1. Jamie Dimon / Chase to let customers buy Bitcoin

2. Airwallex raises $300m Series F and $6.2bn valuation

📚 Good Read: It takes a GENIUS to cause a financial crisis

Weekly Rant 📣

The checkout page is dead.

And your “pay now” button might not be far behind.

This week, Google announced its  "shop with AI Mode," letting users complete purchases directly within an AI experience. Saying “Google Search’s AI Mode will soon be able to shop for products on behalf of users”. The agent can even "pay autonomously," removing humans entirely from the payment flow.

If the agent buys before the page even loads, everything we did at checkout (fraud prevention, identity, conversion optimization) all has to shift upstream to the ad, chatbot, or “point of intent.” Merchants will hate it, acquirer processors might block it, and conversion will plummet unless we think this through.

Every technology eventually disappears. Cars used to require hand-cranking. Phones had operators. Computers had command lines.

The checkout page is commerce's command line - a necessary interface that exposed the underlying complexity. Like we moved from DOS to GUI to touch to voice, commerce is moving from pages to conversations to... nothing.

The whole commerce loop is being upended

The checkout page is dead. But here's what nobody realizes: if the agent buys before the page even loads, every single thing we built our business on has to move upstream."

Fraud detection. Identity verification. Conversion optimization. Upselling. Customer service. All of it. Gone. Upstream. 

• If the agent decides what to buy before your page loads, what exactly are you optimizing?

• If the agent is making the decision your funnel analytics are broken.

• Your personalization needs to be for agents not humans

• How do cart abandonment emails work when there is no cart?

Every metric you track, every optimization you run, every dollar you spend on conversion rate optimization - it's all becoming irrelevant faster than you think.

If you compare today’s e-commerce world vs tomorrow, it really stands out.

In e-commerce or mobile, it looks something like this. 

(Yes, I know this is a dramatic over-simplification and missing countless actors, competitors, and others from the ecosystem, chill PR person. It’s conceptual.)

1. Awareness: See an ad on social or Google

2. Consideration: Search for the product, read reviews

3. Intention: Go to the merchant page select items, checkout 

4. Payment: Click “pay” and use cards, or an express checkout

5. Retention: Receive offers, loyalty, rewards etc.

This flow changes dramatically in the Agentic commerce world. 

(There will be other AI companies, and incumbents won’t disappear, but the primary customer interface, the thing they see, changes)

1. Awareness: Become aware of a product through your AI assistant.

2. Consideration: Your chatbot makes a recommendation based on your context

3. Intention: Your AI Agent selects items, checks out or communicates directly with the merchant, and sends an authorization to the payment processor.

4. Payment: The AI Agent handles the credentials to complete payment. how is still up for debate.

5. Retention: Can we optimize the product for AI agents somehow?

The critical shift: Where humans sit in this flow is much further upstream. So everything we did around the user, like step-up authentication (3DS), fraud checks, upsell, conversion optimization, is now also upstream. 

For merchants, this changes everything. For decades, merchants optimized checkout pages for conversion:

• A/B tests on button colors

• Reducing form fields

• One-click purchasing

• Cart abandonment emails

If commerce moves upstream, this means your customer is the agent.

What happens when your customer is the agent?

If your customer is an AI agent then… 

When the checkout disappears commerce becomes ambient.

Ambient commerce isn't simply a little bit more convenient. It's the final step in making buying decisions invisible and moving from conscious choice to unconscious execution. If commerce becomes invisible, whoever controls the recommendation doesn't just win the sale; they shape human desire itself.

This may be capitalism’s final form. Where the technology that drives ads, drives purchases, and we let go of the wheel.

If this seems far off, or sci-fi, remember that Agentic payments are already happening. Google announced it this week. Perplexity is live. But nobody is quite sure how it will all work.

If the human is out of the loop, how do we manage security, risk, or even control?

It changes everything we knew about conversion, fraud and risk.

There’s a commerce battle coming. What’s coming next is agents getting their own tokens and identities (discussed more later). But who owns the successor to the checkout?

The battle for commerce is moving to the point of intent.

Google knows it has to adapt.

Google handles about 8.5 billion searches per day; its cash cow is the ads business around search. But its share price is eroding as the market fears competition from ChatGPT, which is now siphoning traffic from Google.

Chat GPT is the 7th most viewed web page on the internet, with ~4.5bn monthly users spending an average of 18 minutes per day on the platform (Similar Web). The point where I figure out what to buy, and where I get an intention to buy, are merging. Here’s a recent example.

I’m refurbishing my office to be better for podcasts and media. I worked with o3 to find equipment, local suppliers, and generate a good summary for each of them of my requirements. With Gmail app connectivity, it could even send emails to me and store replies and quotes in Google Drive.

Compare that to scrolling through countless links in Google, calling around suppliers, pulling quotes together, trying to figure out your requirement, and then getting a quote, and then paying. 

Exhausting. 

Two things are merging that make this experience 1000x better:

• The point of discovery (where I figure out what to buy) is moving from the search box to a chat-like experience.

• The point of intent (where I select what to buy, how many, and agree to pay) is moving from a checkout to some sort of AI Agent or chatbot experience.

This is a giant threat to Google.

And they know it.

Google is transforming search “into a big AI chatbot" with:

• Shopping features in search

• Virtual try-ons with your own photos

• Price tracking with notifications

• Agentic checkout in "a few months"

• Integration with your Gmail, Docs, and Calendar

This is brave. It’s putting their cash cow, ad revenue, at risk. And they’re not alone. OpenAI hired Instacart CEO Fidji Simo for consumer operations. Perplexity embedded Stripe and PayPal within its chatbot, and the card networks announced their Agentic tokens in April.

But there are still a ton of unanswered questions:

• What happens if a customer never receives their goods?

• How do they handle disputes?

• Who’s liable if an agent makes a purchase a human disputes? 

• What happens if the agent is compromised by a 3rd party?

The authentication problem

How can we tell Agent Smith from your friendly AI Assistant?

There are several approaches emerging to identify and authenticate AI Agents. I think they’re complimentary and solving for different parts of the ecosystem.

1. Agent Tokenization

2. Agent Identities 

3. Moving trust upstream (Know Your Agent) 

1. How Agent Tokenization will work. 

At their recent Visa Payments forum, noting that “Agentic Payments” are now the firms #1 priority. Here’s how I understood it:

1. Agent Token is generated: The AI Agent at the point of intent will ask customer to tap card to a device. A token will be given to agent that is "device" and "agent" specific. 

2. Agent registers with Visa: The AI Agent then becomes “trusted” in the way a mobile wallet does with tokenization today. 

3. The customer initiates a conversation with their Agent that involves shopping: e.g. holiday, new office equipment, a used Pedro Pascal shirt from the set of The Last of Us.

4. Auth instructions are finalized: eg Agent can buy the Pedro Pascal shirt at eBay for a maximum of $100.

5. Auth instructions define if human-step-up is required: If it’s over $100 for example.

6. Auth instructions are matched at the merchant: If the tokens match, the payment is authorized

7. Liability remains with the merchant: Expect merchants to revolt and decline a lot of transactions of this remains true. (This may be different in Europe where SCA makes issuers more likely to bear liability).

This all implies tokenization is embedded in the chat interface somehow, or that there’s a link between the chat interface and a digital wallet. 

But what happens if you want to use pay by bank, or stablecoins? Visa and Mastercard would argue, you can use their tokens on those rails too. But the practical reality is, there will never be one token network to rule them all. 

2. How Caten Labs gives agents identities.

Caten is one of the 4 Fintech companies this week, has released their Agentic Commerce Kit (ACK) which does two things.

A. ACK-ID: Creates a verifiable agent identity (using w3c’s standard). 

• A cryptographically verifiable and auditable link between legal owners and agents

• Secure authentication allowing agents to prove who they are and that they are authorized to act

• Privacy preserving, disclosing only needed attributes

B. ACK-Pay: Creates a rail agnostic payment communication channel (on chain or off chain)

• Standard patterns for payment initiation 

• Execution routing, pick your favorite payment rail

• Cryptographic “reciepts” or proofs of purchase completion

• Integration points for human oversight

These are great frameworks, but they leave unanswered questions.

Often, the risk around an Agent is things like

• Has the agent been compromised?

• Has the users wallet been compromised? 

• Has a users intent been compromised before it got to a trusted agent?

I’m biased but I’ve been working on something for Sardine* this week that’s an attempt to answer that.

3. How to move trust upstream.

One framework emerging is Know Your Agent (KYA) - a concept I helped explore at *Sardine with PayOS. Here’s the framework whitepaper on a Trusted Agent Directory (TAD).

Full disclosure: I helped develop the KYA concept and Trusted Agent Directory at Sardine, but here's why it matters beyond my day job. We need something like it, even if that framework isn’t it.

Here's how it works:

• Agent creator/supplier checks: Links to verified developer credentials

• Consent orchestration: Giving users the ability to grant permission to agents by amount, context, etc

• Fraud data sharing: User consents to share timestamp, device ID, and origin for merchants' fraud systems

• Transaction routing: Manage checkouts, processors, and fraud engines downstream, optimize and share data needed to improve conversion/fraud detection

• Step up decisions: Pushing biometric or OTP authentication when high risks are detected to the point of intent (user)

• A user dashboard: See all your agents across all of your apps, and manage their permissions

All of these tasks are performed by merchants and their processors today.

This space moves so fast that tomorrow, OpenAI could just launch something that does this and washes the dishes at the same time. So I’m not saying this is the final answer. a) That would be wildly self-serving because it’s something I helped write for my day job. I also really like what Caten Labs is building, and there’s a project to build an Agent Name Service I need to dig into).

I am saying it's a framework that adds to the conversation. 

And that’s worth your attention.

Takeaways

Visa is pretty good at marketing. But they tend not to make whole company, #1 priority level bets lightly. 

Things will move pretty quickly.

Here are some ways you could react.

• Founders: Build at the first click, not the landing page. If the journey starts in chat, your token, risk management, and authentication must be waiting there.

• Regulators and Risk Managers: Audit trails now start at intent, not at tap-to-pay. Fin crime checks need to move upstream (and you thought getting tech to the table was hard, imagine getting AI to the table).

• Bank CEOs: Own real-time identity or become plumbing. The moat lives with whoever can read and link human intent to identity in the millisecond. (This is also why you should pay more attention to Worldcoin).

• Merchants: Work closely with your PSPs, your tokenization partner and your fraud partners to ensure you don’t kill conversion, but you can see upstream.

• PSPs: The goal now is moving from being seen to do something with Agentic commerce, to securing agentic commerce end to end and upstream.

The worst thing you can do is to do nothing.

The excuse is easy, volumes are low, we could enter when it gets bigger, but for now its a distraction, and we don’t even know if it’s just hype or the next IOT (remember that?).

Engaging with this and positioning yourself accordingly when there’s so much noise and you have a business to run. That’s hard.

But that’s what I’m here for.

And that’s why I’m building Fintech Nerdcon. 

So we can work on this stuff together as an industry. 

So, before we veer into a nightmare of AI Agents trying to blackmail humans or defraud us, let's put in place the secure infrastructure today.

We might not have long to get this right.

Prioritize accordingly.

ST.

Post Script

_{Sometimes, I riff with Claude to help find patterns or diagrams in Rants. Sonnet 4 saw something fascinating I wanted to share verbatim because it blew my mind. This right here makes me want to start a whole new newsletter about AI…}

4 Fintech Companies 💸

1. Squads - Build fintech programs on stablecoin rails

Squads builds simple open-finance-like APIs for companies who want to build Fintech programs on stablecoin rails. The Grid APIs help customers create stablecoin accounts, move money, program money (e.g. subscriptions or bill payments), and get transaction history for fraud or credit underwriting use cases. The accounts are global and no FBO structure is required.

🧠 A layer on top of a layer, or a long-term play for “bank-direct” stablecoin accounts? Squads sits on top of BVNK (for the stablecoin and payments infrastructure), as well as companies like Galileo and Lead bank to enable cards. Immediately, my fear was “another middleware?” But the prevailing thesis in the stablecoin world is that the GENIUS Act will be a paradigm shift, bringing banks directly into the stablecoin market. In that world, Squads would cut out a lot of TradFi middlemen and integrate directly with the bank. Making stablecoins a genuine alternative to the FBO structure. It’s noteworthy that Modern Treasury launched a similar product last week. Stablecoins are the new FBO (or suspense account for my British readers). 

2. PayOS - The wallet for Agents

PayOS helps manage agentic flows for checkout, sending, receiving, and bill payments. The payment flows invoke 3D secure or step-up authentication if a high risk of fraud is detected. The idea is that the human is pulled back into the loop whenoversight is needed or risk is spotted. The wallet works for every payment flow, every PSP and merchant.

🧠 This team helped build and launch card tokenization at Visa. With agent tokenization coming from the card networks, they’re super-well placed to lean into where the card networks are likely to head. Generally, the pattern is that the card networks release a framework, which then goes to EMVCo to become a standard, and then the world can adopt it. This is a solid bet for agents that act on behalf of consumers and businesses. There’s still the open question of B2B payments (e.g. Payman) or agent-to-agent payments (e.g. Skyfire or Caten Labs)

3. Caten Labs - The AI-first Financial Institution

Caten Labs is building an AI Native financial institution. It “allows an AI Agent to securely identify itself and transact securely with other agents or human collaborators.” It’s built on stablecoins by the co-founder of the USDC stablecoin project. The Agentic Commerce Kit gives more clues like using verifiable credentials for agents (using W3C standards), identifying which payment rail to use, and a defined mechanism for human oversight. 

🧠 Great pitch, but it’s hard to tell why this isn’t another AI Agents on stablecoins play. There’s a handful trying this model (x402 from Coinbase, Skyfire etc). So why this? The big thing for me is their open-source Agent Commerce Kit (ACK) SDK. They’ve elegantly put together three things, the identity, rail selection and human oversight. They also support x402 and other non HTTP patterns (as you’d see in legacy payment rails).

PS. How lovely is ACK-command as a shorthand for agents doing things. 

4. Zinc - A Retail API for Agents

Zinc gives agents real time ordering information and retail data to allow businesses to buy online and checkout with a single API call. It supports some of the worlds largest retailers like Amazon, Aliexpress, Walmart, Costco and Home Depot. It can handle cancellation, errors and returns.

🧠I was surprised they didn’t have an MCP server. It feels like every AI Agent needs a simple way to traverse retail websites. This API is well built out for that, so why wouldn’t you want to make this available to coding tools like Cursor or Windsurf? Interestingly the company is ~10 years old, but reframing what they do around AI. This is becoming a pattern, where AI is a use case for many infrastructure businesses that were grinding it out before.

Things to know 👀

1. Jamie Dimon / Chase to let customers buy Bitcoin

Customers of JP Morgan Chase will now be allowed to buy Bitcoin again. Dimon has been a long time critic of Bitcoin, though the bank won’t custody it. He has previously labelled Bitcoin a fraud, and in this news. I loved the line “I don’t think you should smoke, but I will defend your right to do it.”

🧠Jamie is simultaneously able adaptable while remaining internally consistent. One of the worlds largest banks should move a bit slower than some smaller banks and non banks. But when the oil tanker turns, it usually turns the right way. 

🧠The lack of custody is in contrast to others who are now keen to get into custody. Until recently regulatory guidance was that banks could not custody cryptoassets.

🧠The ban was probably warranted. There is still, sadly, a massive amount of scam activity related to crypto. Its sensible risk management to limit that damage. But I think whats changed is behind the scenes, banks are collaborating much more with good actors in Crypto through thinks like Sonar, and have better tools to detect and prevent this kind of thing.

🧠But don’t assume JP Morgan is sitting on its hands. It’s Kinexsys payments product is moving trillions of dollars in its closed loop blockchain. They’ve dipped their toe into the waters of “permissionless blockchains” with Avalanche and Chainlink over the past couple of years. Signalling, I think, that as digital assets evolve, JP Morgan payments will be a part of that story. Just in their own, very idoyosyncratic way. 

🧠This is in the same week that the GENIUS Act passed the Senate by a 64 - 32 majority. I’m pretty sure JP Morgan’s bankers see the stablecoin money train leaving the station and will position themselves accordingly in time.

2. Airwallex raises $300m Series F and $6.2bn valuation

Airwallex announced its $300 million in Series F funding, lifting its valuation to $6.2 billion from $5.6B in 2022. T The company says it on track to surpass $1B in annualized revenue this year with a 90% YoY revenue growth.

🧠 Every payments business wants a banking licence. This news follows news that the company plans to obtain banking licenses in the UK and US, signaling a strategic shift from payments to full-service banking.

🧠 Airwallex wants to get into lending. Payments ✔ FX ✔ Cards ✔ Spend Management ✔. What’s left? Working-capital credit surfaced right in the API. Think Stripe Capital but in 160 countries instead of 40. Once Airwallex holds deposits, it can underwrite from live cash-flow data instead of P&Ls.

🧠 Airwallex is an interesting mix of Stripe and Wise. Like Wise they sell borderless accounts and FX, and they then sell that as infrastructure. Like Stripe they offer a checkout, payouts, card issuing and global treasury management. What makes them different is their Geographic spread. Direct licences in more than 60 countries mean they have a cost advantage that often gets missed

🧠 Airwallex is ditching the “APAC darling” label. This dry powder lets them expand meaningfully into Europe and North America. Those with the most to fear are incumbent PSPs. Stripe and Adyen also have a worthy competitor reaching a similar scale. The funding drives them into LATAM, Japan, Korea too.

🧠 The silence on the stablecoin topic is deafening. Airwallex has stuck the landing on north atlantic expansion, now its going elsewhere. Stripe’s model is a little different. The stablecoin pivot gives them a huge new way to access customers and markets.

Good Reads 📚

It takes a GENIUS to cause a financial crisis

Brian Shearer, who was until recently at the CFPB, argues that the GENIUS Bill, and allowing stablecoins to be pegged to anything other than fiat, will lead to big tech oligopolies in banking that become too big to fail, and possibly too big to bail out. PS. This was written before the GENIUS act passed the senate by 64 - 32.

🧠 This is the coherent, policy-based push back against stablecoins. Although I disagree with almost every point made, they’re well made. 

🧠The argument centers on private money being what creates financial crises. To which the solution is to ensure only tightly constrained banks can create money.

🧠The piece suggests that once big tech has a banking monopoly, it can crank fees ever higher. I have some sympathy for this perspective. This is written at a time when Apple has lost its monopoly on the App Store 30% take rate. Without question government intervention helped here, but so did massive efforts by companies like Epic.

🧠The argument for a solution for high transaction fees? Simple, cap fees! Except.. That doesn’t work. Fee capping is a pin cushion. If you cap one fee, another one pops up somewhere else. Yes, banks are wildly profitable animals, but government intervention has done nothing to prevent that, the solution is competition not government intervention.

🧠 Brian argues the core difference with stablecoins and banks is that they “use a different kind of spreadsheet,” and while from an accounting perspective, that’s accurate, it also misses the point. Stablecoins are instant, 24/7, global, and programmable. Fiat, e-money and even FedNow are not. 

🧠We need to stop viewing finance through the prism of the financial crisis or corporate monopoly risk. Instead, view it as a system where encouraging safety and soundness is one policy stick, and encouraging competition is the carrot. 

🧠 The stablecoin genie is out of the bottle. The question is one of how to harness it. Even if we say you have to be a bank to issue or custody stablecoins, I imagine many existing stablecoin and crypto businesses will go for, and achieve charters (don’t forget, Anchorage has a federal charter). 

Tweets 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