Writing
Notes on payments, AI operations, and where the two disciplines borrow from each other.
Mastercard's Chargeback Rules Tighten From April 2027
Mastercard replaces its chargeback monitoring with GMAP on 1 April 2027 and cuts its ECM line to 0.9% by 2031. What that means for high-risk merchants.
Visa's Agent Pilot Pays Only What a Human Approved
Visa, BILL and Billtrust have a live pilot in which an AI agent pays supplier invoices by card. The agent decides nothing, which is why it fits on cards.
The Agent Payments Tracker
x402, MPP, ACP, AP2, UCP, Visa Intelligent Commerce and Mastercard Agent Pay compared: who backs each, who holds the money, what's live. Updated weekly.
Visa's VAMP Ratio, Read From the Merchant Side
Visa's 'excessive' merchant VAMP line fell to 1.5% on 1 April 2026. For high-risk merchants, the acquirer's own 0.5% line is the one I'd watch.
What 'Should Have Known' Means for a Payment Processor
The FTC's Nuvei and Humboldt settlements read like a spec for merchant underwriting: records the merchant didn't write, checked before approval and every month after.
When the Card Networks Co-Own the Stablecoin
Visa, Mastercard, Stripe, Shopify and Coinbase now co-own a stablecoin. It strengthens the case that card networks absorb stablecoins, and raises a new question for merchants.
Cloudflare Is Building the Seller's Side of Agent Payments
Cloudflare's new tool lets a business charge an AI agent per request in USDC, with no card or subscription. It's the missing half of my stablecoin argument, and it's still a closed beta.
Someone Checked Whether AI Agents Are Paying Yet
TRM Labs screened $52.7M of x402 payments on-chain and found only 0.6% to 7.5% of the likely commerce looked agentic. The agent-payments story had been running on self-report.
SoFi and Mastercard Just Complicated My Stablecoin Argument
SoFi now settles its whole card program in its own stablecoin, on Mastercard's network. It's evidence for stablecoin rails, just not for the version where they replace the card networks.
Paymob's $35 Million Round Is a Bet on Fragmentation
Paymob raised $35M to give MENA merchants 60+ payment methods through one integration. I read it as a bet that regional fragmentation isn't going away, which is what orchestration is built on.
The $200 Billion Interchange Settlement Isn't Settled
Walmart and about 1,000 merchants want the Visa/Mastercard deal rejected. For anyone building a payments cost model, the fight says more than the settlement does.
Georgia's Payments Charter Is a Bigger Deal Than It Looks
A niche Georgia bank charter lets processors reach Visa and Mastercard without renting access from a sponsor bank. Fiserv, Checkout.com and Stripe now use it.
Most Agent Payments Won't Run on Cards
People assume AI agents will pay by card. I think most of that money will move over stablecoin rails instead, and a public chain changes what it takes to know what your agents spent.
Agentic Checkout Is Being Built Faster Than Anyone Trusts It
Mastercard and Visa are racing to build the infrastructure for AI agents that pay. Only 7% of shoppers say they'd let one. That gap is a verification problem, and payments knows those.
An Agent's Self-Report Is Not Evidence
Agents are mostly honest about the limits of what they searched. The expensive gap is between a number that sounds settled and a number anyone has checked.
Code Is Getting Cheap. Attention Isn't.
My expectation for the next few years of venture-backed software: the cost of building keeps falling, and the competition for a fixed amount of human attention gets much harder.
Four of Six Headlines
An ordinary morning, an ordinary news feed, and a count that says more about where business attention has gone than any forecast does.
Owned Media Matters Again
An old advertising distinction I hadn't thought about in years turns out to be the useful frame for what AI is doing to how people find things.
The People Who Need It Most Have the Least Time to Learn It
A business owner with a real automation problem doesn't have a free evening a week for the next two months. That isn't a motivation problem. It's arithmetic.
There's No Such Thing as a Bad Payment Provider
Almost every time a business tells me its provider is bad, the real problem is that it only has one. Those are different problems with different fixes.
The Model Will Forget What You Told It
Agents don't fail loudly. They drop the instruction you gave three steps ago and keep going, which makes verification something you build in rather than something you assume.
What Changed When I Went From Chat to Cowork to Claude Code
Most people evaluating AI stopped at the level where it answers questions and concluded that was the whole thing. There are two more levels, and they don't feel like the same tool.
What I Notice When Watching Startup Pitches
The best pitches make the connection between problem, solution, distribution and economics feel inevitable.
The Complexity Tax
A business can look profitable and still be exhausting to operate. Spreadsheets capture revenue, not complexity.
I Like Business Ideas. I Trust Experiments More.
Good ideas are cheap. The hard part is finding out which one survives contact with customers and reality.
AI Didn't Make People More Productive. Yet.
Handing someone a faster tool rarely changes the work. The opportunity nobody is spending on is redesigning the workflow around it.
I Am More Interested in Leverage Than Productivity
Productivity asks how to get more done in an hour. Leverage asks how to make that same hour worth more after it ends.
Why Payment Approval Rate Can Be a Dangerous KPI
A declined transaction isn't just a decline — it's a lost customer, a chargeback, a second attempt. Context matters.
Payment Orchestration Is Not a Fancy Payment Gateway
The value was never in the routing. It's in optionality — being able to move, compare, optimize, and survive.
The Best AI Use Case Is Usually Boring
The most valuable automation doesn't generate an impressive demo. It works through the folder of documents nobody wants to touch.
Most people use AI to write. I use it to work.
I built an AI system that runs my own sales pipeline — drafting, chasing, documenting. It never sends or decides. What it fixed, and what changed my mind about which asset is worth anything.