Things to know 👀
Jamie Dimon said on Tuesday's Q3 earnings call that the block covers any third party logging into a Chase account with the customer's credentials to pay bills or move money. Read-only data through Plaid still works, and there's no timeline for letting agents act. Dimon called them "rogue agents," warned customers of a "liability transfer" when they hand an agent their logins, and told regulators to prepare for what agents could do to deposits. JPMorgan is building its own in the meantime. Smart Cash is in pilot and moves money between your Chase accounts for yield. Citi has since told the Financial Times it also bars third-party AI agents from logging into customer accounts or transacting directly on its digital banking platforms.
🧠 The big firms block first and ask questions later - because they can. Amazon, Citi, and JPMorgan will not go out of business for blocking bots and screen scraping, but they will potentially have a bunch of new problems. Today’s customer isn’t going anywhere because of this block.
🧠 JP Morgan is growing just fine and can block from a position of strength. The bank made $15.3bn in Q3 on $50.9bn of managed revenue, up 8% year on year, and the Commercial & Investment Bank grew net income 20%. A $15bn quarter buys a lot of patience on agent access.
🧠 There’s a credible risk of fraud or consumer harm here. What if the agent fails and you lose all your money? What if your agent is compromised and your money is stolen? (Which I wrote about in Agent Wars Pt 2). When banks say agents are a security risk, they’re right, but it’s also the perfect excuse to do nothing, which has downsides.
🧠 Consumer agents are inevitable. You can't block everything forever. The risk for banks is you keep the account and lose the customer. Your salary still lands at Chase while your agent shops for your next card, savings rate and insurance policy somewhere else.
🧠 This week Mercury made payment requests available in their MCP server for consumers and businesses. This means your agent can ask, “Hey, an invoice just landed; looks right? Should I pay it?” It prepares the payment, and you review and approve it inside Mercury. The contrast with JPMorgan and Citi is noticeable. (Meow and Payman also do this).
🧠 Your app and your own agent aren’t a defense. I heard a lovely anecdote about a bank CEO saying, “Consumers don’t need agents; look, you can block your card in our app in 3 clicks.” Only to be asked, “ok, but wouldn’t they prefer to tell their agent, ' Hey, I’ve lost my wallet; can you block all my cards’ instead of logging into every banking app?”
Cue the CEO staying awkwardly silent for fifteen seconds.
🧠 We’re shifting from a world where consumers come to your app (or agent when you build one). To one where they express their intent to agents, and you have to be able to meet that. Just as banks had to build APIs to make mobile possible. They have to build MCP servers and KYA stacks (using things like personal agent protocol and x401) to make that secure and permissioned.
🧠 Expect the open finance playbook, faster. We eventually got to API and security standards. We’ll get there with agents too, but it could take 5 years or more. The difference this time is the speed at which AI can switch. We haven’t seen mass consumer adoption yet (if anything, far from it, but if it does, it could happen fast).
🧠 The marginal new customer is picking the innovative fintech companies. The banks already see fintech companies as competition. What remains to be seen is how much being AI agent-friendly accelerates that competitive threat.
🧠 Banks need to figure out how to become available to agents in a secure way. Give it a permissioned way in, read-only first, with a step-up in your own app before any money moves. In Thursday’s essay, The Mod Era of Finance, I broke down how I think x401 will let banks grant permissioned agent access to accounts within their CIP framework.
Ramp raised $1.85bn at a $60bn valuation on October 15, in a primary round led by Dragoneer, with Thrive Capital, Founders Fund, Sands Capital and Avenir Growth as co-leads. The company says it has passed $2bn in annualized revenue with positive cash flow, and serves more than 85,000 businesses. Annualized purchase volume is over $275bn, up 175% year on year. The valuation was $44bn when Ramp raised $750m in June, and $32bn last November.
🧠 Ramp is getting bigger AND growing faster. It says revenue growth is 25% faster than at the same time last year. Growth usually gets harder as the business gets larger, because you need to add so much more revenue just to maintain the same rate. Because they nailed the AI op model, they’re launching accounting, token routing, and an AI Harness and a ton of new products.
🧠 They’re valued more than PayPal or Block. PayPal, once the darling of Fintech, sits at $48.5bn, while Block is at $46.8bn, both on a much higher base revenue, with Block showing an impressive 25% YoY spike in gross profit.
🧠 That valuation is nearly 12x what Brex sold for. That’s a testament to three things: 1) continued and accelerating growth. 2) A credible AI story with products showing revenue and traction. 3) Possibly the best B2B marketing in the world. With a Broadway show, Tony Hawk and flooding the timeline with X articles by employees on their ways of working.
🧠 Investors are buying Ramp’s ability to expand into new products and new products are driving revenue growth. A majority of customers use two or more products, and a quarter use three or more. The number of customers generating over $100,000 in revenue for Ramp grew 113% year on year.
🧠 The valuation still asks a lot. Using $2bn as the revenue run rate puts it at about 30x revenue. Ramp’s definition includes interchange, subscriptions, interest and other revenue, less customer incentives, so you have to look at the economics of each part. I’d love to see how much the software contributes as customers add more products.
🧠 Alex Johnson’s piece this week was staggeringly good. He argues all of this capital allows them to stay private, which in turn helps them continue this pace of growth (something he calls perpeptually private machines). Unlike Alex, I don’t see the machine stopping due to talent exodus or an eventual breakup. Like Stripe, they’ll keep producing founders and attracting former founders.
🧠 Speaking of AI products, it’s worth taking a much closer look at one to understand it better…
Ramp built Glass so every employee gets a working AI setup on day one. You sign in once, and it is already connected to Slack, Notion, and the company's internal tools, with the right permissions. It ships with a marketplace of 350+ skills (workflows one person built that anyone can reuse), memory that carries your context between sessions, and scheduled automations that post results back to Slack. Ramp calls it the AI harness its staff work in; 99.5% of them use it, and CEO Eric Glyman says it is one of the reasons the company moves so fast. Early access is now open to other companies.
🧠 Name another major fintech company launching AI products? This is why I say “fintech is dead.” The companies leading the next generation have to compete to be the best, period, and with Glass, Ramp's rivals are AI labs and workplace software giants. The products the best finance companies launch are drifting beyond finance itself.
🧠 Glass is multiplayer by default. Most company AI rollouts are one person in one chat window, and whatever that person learns stays with them. In Glass, a skill one analyst builds on Monday is in everyone's toolbox by Tuesday, and the output lands in the Slack channel where the team already works.
🧠 Memory is way harder than it sounds and magic when done right. Glass builds one per person from their own logins and refreshes it every 24 hours, so the agent knows your projects, people, and documents before you type. One employee's memory is hard enough; a whole company's, where every memory must respect who can see what, is insanely hard.
🧠 Ramp is three AI products deep in three months. Token Spend Management (16 July) counts the tokens, Router.com (19 Aug) picks which model gets them, and Glass is where the tokens get made.
🧠 There’s an AI productivity flywheel running. The more your op model becomes AI-native, the more products you can externalize and ship. Which attracts revenue, talent, and investors internally. Start with how you work internally.
Mastercard and BVNK launched Global Money Platform on October 15, combining fiat and stablecoin accounts, conversion, stablecoin-linked cards and local bank payouts through a single API. Banks, fintechs and enterprises can embed the accounts into their own products, under their own brand, or use the platform for their own treasury and payments. It’s the first major integration since Mastercard completed its acquisition of BVNK in August. The platform is available now through BVNK, with Nuvei and Marqeta among the first companies set to integrate it.
🧠 Moving between stablecoins and bank accounts is still too much work. A business might collect money onchain and need to pay a supplier into a local bank account. Someone has to manage the conversion, payout and liquidity between the two. That adds costs and delays and destroys the “24/7” promise.
🧠 Mastercard is giving BVNK a much bigger route to market. MasterCard Send can payout to 95% of the banked population according to its own stats. That’s a baked in off-ramp for stablecoins in a single package.
🧠 There are two main flows emerging. Visa’s Cuy Sheffield told me on Tokenized “We’re seeing either stablecoins in fiat out, or fiat in and stablecoins out.” A lot of banks are on one side of the transaction and either don’t want to, or can’t technically hold stablecoins. But they recognize the need for it.
🧠 Visa has moved into production too. Inyo Global is the first provider to bring Visa Direct’s stablecoin prefunding into commercial use. Eligible payment companies can fund cross-border payouts with USDC, around the clock, through the same Inyo APIs they already use. Stablecoins become another funding option inside an existing payments service, which is a much easier adoption decision than rebuilding the whole operation.
🧠 You can fund the weekend during the weekend. If your funding rail closes on Friday, you have to load up for demand you can’t yet see. Stablecoin top-ups let you respond as payments come in, with less money parked at the settlement provider. In an earlier card-program example, Visa says USDC settlement helped Crypto.com cut prefunding from eight days to four.
🧠 Stablecoins are becoming part of the normal payments stack. They’re no longer controversial. Usually the question now is more how than why.
On October 8 the OCC fined American Express National Bank $350m and issued a cease-and-desist order (an order to fix the program, with deadlines), and the Fed issued its own order against American Express Company and its Travel Related Services arm. The OCC found that from June 2014 to May 2025 the bank processed about $13bn of suspected trade-based money laundering through card charges and the repayments of those charges, "in certain instances through accounts associated with bank insiders." Trade-based money laundering means dressing dirty money up as payment for goods, e.g. an inflated invoice settled on a corporate card and repaid from a clean account. Amex consented without admitting the findings.
🧠Eleven years! The fintech press in particular likes to do a full post mortem everytime a non bank has an AML failing, but is often silent when the incumbents have these issues. Amex is an incredible company, doing amazing things with tech, and I hope this means they’ll apply that innovative nature to their compliance stack. This is something I think fintech companies don’t get enough credit for having pioneered.
🧠 Amex aimed its risk assessment at the wrong product. The OCC says the bank focused on its small deposit business and under-weighted the card business, which is most of the company. The $13bn went through cards.
🧠 This is surprising given Amex sees both ends of a payment on its own cards. It issues the card and signs up the merchant, so the cardholder data and the merchant data sit inside the same company. An inflated invoice between the two is the pattern that data exists to catch.
🧠 The regulator says its a people issue - I think it’s people AND tech. The bank had too few people, people without the right expertise, weak independent testing, and training gaps for staff and directors. And the regulator always says it’s a staffing issue. To me that’s a proxy for staff and tech. You don’t always need more people, you need to spot the right things and raise those. AI can help here.
🧠 The Fed order covers the whole group. The New York Fed found enterprise-wide weaknesses in transaction monitoring, fraud referrals and third-party risk, and the board has 90 days to show how it will fund and oversee the fix. Travel Related Services, the operating company that sits above the bank, must separately fix its sanctions screening.
🧠 Five days later Bank of America paid roughly 3.5x as much for the same category of failure, and admitted it.
On October 13 Treasury announced $1.21bn in penalties against Bank of America. FinCEN assessed $1.2bn, with the OCC's $750m (the largest penalty in that agency's history) credited against it. OFAC settled 475 apparent sanctions breaches for $11.6m. BofA admitted it willfully violated the Bank Secrecy Act. It failed to run an adequate AML program, failed to check foreign correspondent banks, and failed to file suspicious activity reports.
The bank failed to report more than $5bn of suspicious transactions on time, including more than $1bn tied to Treasury's priority risks: Chinese money laundering networks, fraud, and trade-based laundering linked to the US southwest border.
🧠 Treasury's sharpest finding is that BofA "effectively outsourced" detection to third parties. Law enforcement subpoenas and other banks' requests surfaced the suspicious customers, while BofA's own system failed to properly monitor trillions of dollars of transactions.
🧠 The alerts that did fire sat uninvestigated. The OCC says the bank knew for years it was failing to pass potentially suspicious activity to investigators and left it uncorrected. The late SARs that followed covered billions of dollars, many with signs of money laundering, and Treasury calls them "of limited utility to law enforcement."
🧠 The compliance team did its job. Staff escalated concerns about trade-based laundering, fraud, and customers taking payments from unknown third parties with possible narcotics links, and the bank let them sit. Treasury's words are that BofA "prioritized its revenues and minimized the impact on its required compliance resources."
🧠 Compliance is always the last to get budget, but it doesn’t have to be expensive. Help the compliance team. Give them agents like those from Sardine* or Bretton that can work those alerts and create a SAR (or dismiss the 95% of alerts that are false positives, staff are literally drowning in them). .
🧠 AML is hard for everyone, and staying good at it means treating it as a product you keep investing in. Amex's gap ran 11 years and BofA's ran at least three after the OCC first flagged it. The next CFO who trims a compliance budget now has two 2026 precedents to explain to the board.
🧠I’m sure both organizations take this very seriously indeed. But I am literally screaming into my keyboard here. You can’t hire your way out of an alert visibility and volume problem. If 95% of what your alerts show is a false positive, you need to 10x your staff to stand still. If alerts aren’t going to investors, you need a new transaction monitoring system.
4 Companies 💸
1. HIFI Stablecoin payments and settlement for financial platforms
HIFI lets fintechs, marketplaces, and financial institutions collect money, convert between stablecoins and conventional currencies, and pay recipients into bank accounts, wallets, or cards through one integration. It also supports institutional transactions involving tokenized assets, and DTCC independently lists HIFI among participants in its July production trades using tokenized securities.
🧠 One API for stablecoins and tokens is a compelling promise; it's also one with a crowded market. I can imagine the combination of stablecoins and digital assets works really well for anyone who’s doing treasury management in import/export. Its pitch is a dedicated engineering and compliance partner, which could appeal to fintechs whose customers and payment flows need more attention than a standard integration. Vs a Bridge or BVNK, those customers have a little more handholding required.
2. Mika Bookkeeping and tax preparation for German small businesses
Mika handles bookkeeping, VAT filings, annual accounts, and tax returns for small German limited companies, using AI with in-house bookkeepers checking the work. Owners can also upload a confusing tax-office letter and get an explanation of what it means and what to do next. Mika exposes financial data through MCP, so customers can use their preferred AI assistant.
🧠 Tax admin is white hot as a category. I’m seeing more companies like this pop up daily for the US and elsewhere. Turbotax and QuickBooks are on notice for the US. But Germany has a very complex tax code, is notoriously strict, and centralized (without market solutions like TurboTax). Small and medium-sized enterprises (SMEs) in Germany spend an average of 7% of their total working time purely on managing bureaucratic requirements, with tax cited by 70% as the most burdensome. This is ripe for AI support.
3. Snag Furnished sublets for students and mobile renters
Snag helps people find furnished rooms and apartments for a few months, and helps existing tenants find someone to take their place while they’re away. Renters search by location and dates, contact the person listing the property, and arrange a sublet through the service. Its inventory includes individual tenants’ homes, while June Homes also identifies Snag as a distribution partner.
🧠 Internships have a housing problem. You need somewhere for 3 months only. Snag is organizing behavior that already happens through friends and group chats, while partnerships with housing operators could help keep apartments available when those people arrive. The fintech opportunity is making it safe to hand a stranger your rent and deposit, which could also give both sides a reason to complete the booking on Snag. Is this a big niche? Probably. Also seems like a feature Airbnb could ship?
4. Atum A payments network connecting stablecoin providers
Atum lets wallets, payment companies, and fintechs send payments across different stablecoins and blockchains without connecting separately to every provider. Settlement providers compete to complete each payment, with the sender specifying what they’ll pay and the recipient getting the asset they requested. Atum says it also supports authorization, identity, and reversible payments without holding customer funds.
🧠 Again, another crowded category of orchestrate the orchestrators. Companies like Mesh are growing rapidly and connect the connectors. The competing for business is neat. As is Authorization and reversibility. Sending money is nice; the protections make it reliable for bigger workflows and companies. But are they late to the party? Hard to say. This space is still early and growing. Payments often have many winners too.
Good Reads 📚
Geoff argues the employment market has already moved on. If you’re in software engineering, especially, you cannot hold out. You must be AI-native. This piece argues every employee needs to learn the magic tricks under the hood of Claude Code and other agents, and if you’re a people manager not doing that, you’re failing at your job.
🧠 The headline hit me like a bullet. But it’s right. If you’re not helping your team reskill rapidly, that is negligence, especially in software engineering roles. But I’d argue broader than that too.
Former PayPal CEO and Meta payments head, David Marcus, argues that AI agents will replace apps and checkout screens, forcing payments to work without a human clicking the button. His “headless money” stack combines stablecoins, wallets, bank connections and cards, with spending limits and clear liability when an agent gets tricked. He expects the biggest rewards to go to whoever controls the agent deciding what we buy and from whom.
🧠 I buy the vision, but if the agents try to take too much of the payment economics, merchants will block it. I honestly think agents will have to find a new balance. How do you be genuinely user-centric without just feeding them what the buyer of advertising wanted to show you?
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
