Things to know 👀
Open Standard announced its stablecoin, OUSD, in June with 140+ partners and switched it on yesterday on Tempo, Ethereum, Solana and Base. Businesses can mint and burn at par ($1 = $1) at no cost through Coinbase, Stripe, Visa and Mastercard. Bridge issues the token, and the reserves sit at BlackRock, BNY and Lead Bank with monthly attestations. Coinbase, Mastercard, Shopify, Stripe and Visa each took an equal founding equity stake in Open Standard, and CEO Zach Abrams says the overwhelming majority of the company's equity will go to partners over the next four to five years, in proportion to the OUSD supply and activity each one drives.
🧠 This is a different business model, one that banks and the market tell me they prefer. Tether keeps the yield on USDT reserves. Circle keeps most of USDC's yield and pays Coinbase and other distributors a share. Open Standard pays partners in rewards and equity for the supply they bring, and takes a small transaction fee for itself.
🧠 That makes Open Standard a member-owned utility. Mastercard and Visa were bank-owned associations until their IPOs in 2006 and 2008, and SWIFT is still a cooperative owned by its member banks. The two card networks are founding a stablecoin on the structure they started with.
🧠 A bank distributing USDC gets whatever share of the yield it can negotiate with Circle. A bank distributing OUSD earns rewards, and a slice of Open Standard, in proportion to the supply it brings. BNY, Standard Chartered, DBS and SoFi are already on the partner list.
🧠 Circle is still the behemoth onshore and Tether offshore. They have massive liquidity, chain access, and integrations. Tether is facing increasing scrutiny now over AML, and Circle needs to find business models outside the stablecoin itself (hence the launch of its chain Arc)..
🧠 I’ve been at SIBOS this week, where the model is getting a warm reception from financial markets folks. One large global bank told me, “we cannot work with a private issuer.” Think about why. Anything that’s competing to offer yield is competing for your deposits. OUSD, something you back and share in the yield of, is less competitive.
Parafin lets software platforms like DoorDash, Gusto, and Amazon offer loans to the small businesses that run on them, underwritten on the sales and payroll data the platform already sees. Founded in 2020, it has since lent $3B+ to 60,000+ small businesses through working capital, B2B pay-over-time, and a business card. It was last valued at $750M in Dec 2024, and the price was not disclosed. Stripe already runs Stripe Capital, so a lot of my timeline asked why buy another lender.
🧠 Parafin differentiated on underwriting. Their ability to use varied data sets and a custom transformer model to underwrite that risk gave them a consistent lending edge. Stripe now has that.
🧠 Parafin gets distribution. Stripe gets more to sell. DoorDash, Amazon, Gusto, Worldpay and SpotOn already use Parafin’s financing infrastructure. Those relationships took years to build, and some are with payment processors that compete with Stripe. Stripe gets an established set of financing programs and the team that built them.
🧠 Parafin had the funding to keep going alone. In 2026, it added $150M of debt, a $300M forward-flow into a rated vehicle, and warehouse lines from Goldman Sachs, which is a securitization path. Stripe is buying a working credit business with its capital markets partners attached.
🧠 The M&A ambition here is well ahead of some of its peers. Stripe has bought Bridge ($1B), Privy ($500m), Metronome ($1bn) and OpenRouter ($7B+) in under two years. They even swung $53B (w/ Advent) at PayPal before walking away. Each deal adds a product it can sell to customers it already has. (Although Adyen did acquire Orb, the billing for AI companies platform similar to Metronome).
🧠 When incumbent payment companies made acquisitions, they were usually for geographic expansion or business line expansion. Companies like Nexi and Worldline are about 10 different smaller companies in a trench coat. Those companies were never fully integrated. Stripe’s acquisitions usually take them into new territory. Metronome is a new kind of billing, Bridge brought in a new type of stablecoin orchestration, and OpenRouter is a real swing.
Robinhood Agents can execute trades out to Robinhood's 29 million users. It works a bit like OpenAI’s dots. You name an agent in the app, open a dedicated agentic account, pick a model (GPT-Luna is free until year-end) and tell it what you believe in plain English, and Loops (coming soon) run that on repeat.
Since agentic trading opened to power users in May, 150,000 customers have opened agentic accounts, and their agents hit Robinhood's tools almost 30 million times a day. It's all "coming soon" to eligible US customers.
🧠 This is a good example of how to do consumer agents well. Launch an MCP server for the sophisticated user, let them access a very tightly scoped delegated account, watch what they do, and then build that for the mass market.
🧠 Robinhood Agents have limits you set and manual trade approval on by default. The agent can only touch the dedicated agentic account, and every trade needs a tap unless you switch that off.
🧠 The functionality is where chat as an interface starts to make sense. You state a belief in English, and it gets traded against premium data. That starts to look like a wealth advisor.
🧠 You can see a business model emerging here for agents for consumers. Usage is free until year-end. From January, I'd expect a Gold upsell or a per-agent fee, plus a cut of every $10 data app.
🧠Credit where it’s due Public.com did this first
Airwallex has rebuilt its Business Account around one view of every entity in the group, with sub-accounts for operating cash, yield and credit. It has also added agent integration so Claude, Cursor, or any MCP client can read balances, prep FX, and issue cards inside the permissions you set. CEO Jack Zhang's essay calls it the start of the "Agentic Banking era."
🧠 Multi-entity single view is insanely powerful for companies operating cross borders. If you've got entities in the UK, US, Singapore, and Australia, each with its own bank login, you probably run group cash in a spreadsheet someone updates on a Friday. That on a single screen before money moves is high-key amazing for finance folks used to dealing with bank portals.
🧠 Airwallex customers tend to have multiple entities. Think of an e-commerce brand selling into four markets, a travel operator paying hotels in a dozen currencies, or a SaaS company that opened a US entity to sell to US customers. Mid-sized, in several countries, and typically not big enough to have a large treasury team.
🧠 Bring your own agent (BYOA) makes sense on top of that. Jack Zhang's example is an agent that sees EUR 100k arrive from a German customer, knows EUR suppliers are due next week, and keeps the euros rather than converting them. Most of these companies make that call late, badly, or never.
🧠 Meow, Mercury, and Ramp have all shipped agent access to business accounts since April but are all domestic US-focused mostly. Meow's agent opens the account, Mercury's runs day-to-day ops, and Ramp's agents do AP and procurement. Airwallex is the only one of the four pitching the agent across borders.
🧠 Incumbents are miles away from doing a bring-your-own-agent interface. As of May, Chase and Bank of America had no first-party MCP endpoint for a business account. It’s becoming a trend, and one that will likely prove durable.
Most decisions in finance come from a table. Rows are customers or transactions, columns are what you know about them, and one column is the thing you want to predict (will this customer default, fraud, churn). For twenty years, the tool for that job has been gradient-boosted trees, best known as XGBoost, and some version of it sits under nearly every credit score and fraud model you've met.
Kumo Tabular, from the team NVIDIA bought in June, takes a table with some answers filled in and predicts the rest in one pass.
🧠 XGBoost runs credit, fraud, collections and pricing at nearly every lender. Matching it with zero training is the story.
🧠 LLMs are bad at tabular data. They read text, not tables. Earlier tabular foundation models only worked on small tables.
🧠 Tabular Foundation Models (TFMs) will do everything XGBoost is doing today. It's licensed for commercial use and ranks first on the four main tabular benchmarks against tuned gradient-boosted trees.
🧠 An XGBoost model build takes months. If a pretrained model gets there in one pass, a risk team can ask ten times as many questions of the same data. It means your data science teams go from managing ML pipelines to doing data science.
🧠 This is an immediate upgrade for anyone who’s been using machine learning at scale with good data hygiene. That is not the case inside many banks today (outside of the really big ones). But I expect fintech companies to adapt this rapidly.
Soon your AI agent will turn up at a bank and ask to open an account for you. Today the bank has no way to know which agent it is, and if that agent has any delegated authority from you. AFSP is an open spec (v0.1) designed for that use case. The agent's operator registers, you approve the session with Face ID on your own device, and the bank gets one signed package to check. KYC and the account decision stay with the bank.
🧠 It’s step-up authentication for agents. AFSP is the first I've seen aimed at the bank (or fintech) to consumer to agent relationship that’s neutral. There are things like Visa’s TAP, and the KYA work Visa, Mastercard, and Alipay+ are working on, but that’s payment method-focused.
🧠 A neutral standard makes a ton of sense. AFSP covers the step where an agent asks to start the relationship, with one certification and one integration instead of a bilateral deal per bank.
🧠 v0.1 only opens a savings account, but the design works for any account or action, and the roadmap components are the full agentic banking channel.
🧠 Worth a look. The spec is still early, and focused on savings accounts for now, but dig in and let's make this happen.
4 Companies 💸
1. Piston - Fuel payments for fleets, without fuel cards
Piston lets US fleet operators control which driver fills which vehicle, without issuing fuel cards. The app checks the driver's location, pump, and mobile app to unlock the fuel. Drivers pay through an app that checks the driver, vehicle, and location before allowing the pump to dispense fuel. The company reports more than 2,000 participating stations across 48 states and 98% customer retention.
🧠 This only works at participating pumps and gas stations. That makes local delivery contractors and trade fleets a sensible starting point: they revisit the same areas, so you can recruit stations around their routes. I like the idea of building a payment network one fleet at a time if the economics are there, but I wonder about the long-term cap on scale?
2. Luminary - Help wealth advisors plan estates after a bereavement
Luminary helps wealth advisors show families who would inherit a family member's assets how changes to their estate plans affect the outcome. Advisors upload estate documents to build visual plans they can review with clients, attorneys, and tax advisors. Caprock’s COO estimates that Luminary increased advisor capacity by 25% and shortened new-client onboarding by more than a week.
🧠 This is an important way for advisors to win clients. That makes sense for advisors serving wealthy families whose affairs involve several trusts and professional firms. The lasting value is to keep those people working from the same plan as the family’s circumstances change.
3. ZeroRisk - Cybersecurity for merchants through their payment providers
ZeroRisk helps banks and payment providers identify cybersecurity problems across the merchants they serve, then guides those merchants through addressing them. It combines security monitoring with help completing compliance checks and fixing vulnerabilities. The company says it already works with Bank of America, Global Payments, Checkout.com, and Trust Payments.
🧠 I like the distribution here: a small merchant already has a relationship with the company that processes its payments. That gives ZeroRisk a way to reach businesses that are unlikely to buy standalone security software. But telling a busy shop owner that their website is vulnerable only helps if someone makes the fix easy. Could they expand to practical help too?
4. Triver - Get paid sooner on small-business invoices
Triver helps UK small businesses receive cash against unpaid customer invoices. Businesses connect their banking and accounting data, and Triver says approved invoices typically take less than five minutes to fund. It reports 2,400 clients and £400 million of invoices financed, with repeat usage accounting for more than 90% of monthly revenue.
🧠 This is factoring with more data, but brought to SMBs. Triver’s fees start at 1.8% for a 30-day invoice, or £180 on £10,000. That can make sense if it lets you take another profitable order or pay staff while a customer takes their time. The problem with factoring has always been that it becomes a drug, easy to get on, hard to get off. Still, for many who need to order inventory well ahead of sales, dealing with net 90-day, or even net-30 collections its a crucial cash flow tool.
Good Reads 📚
Private Equity firms are buying companies to transform them with AI, and they identified a pattern that I see in 95% of companies, even growth companies “Most portcos use AI for email drafts and meeting notes, and most of the partners couldn't describe what comes after that.”
🧠 If you add “writes the code” to that, you’re up to 98% of companies. Very few are changing how they work around the tools.
🧠 The best line in the whole piece “The companies that failed at this ran AI like a consulting program.” They ran a long diagnostic, drew a target architecture, projected ROI on assumptions, and spent 18 months on integration before anyone measured a dollar. The last step on that roadmap is a final presentation.
🧠 The companies that shipped ran it as a loop. They picked one workflow with real users, real data, and an outcome the business tracks, proved the smallest useful slice in two to four weeks against a measurement they built before the agent existed, and then made a decision.
🧠 So much alpha in that contrast.
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
