Market news

News

Big banks lock out AI agents while an analyst puts four to work on earnings season

JPMorgan and Citi lock out third-party AI agents, Coinbase works on banks from both ends, and Marc Rubinstein shows what four AI agents do with earnings season.

22 October 20262 min readUpdated 23 October 2026

Agents were the theme this week, from the banks blocking them and from an analyst who put them to work. Plus a smart read on how Coinbase is getting stablecoins into the banking system.

JPMorgan and Citi say no to outside AI agents

Simon Taylor's Fintech Brainfood has the details. JPMorgan won't let any third party use a customer's Chase login to pay bills or move money. Read-only data through aggregators like Plaid still works. Citi told the Financial Times it does the same. JPMorgan is building its own tools instead, starting with a cash-sweep pilot. Taylor's view is that the fraud worry is real, but it's also a convenient excuse not to build anything. Customers will get agents either way. The banks that win will offer a permissioned way in: read-only first, then payments with an approval step inside the bank's own app. Mercury already lets agents prepare a payment for the customer to approve, which makes the big banks look slow.

Sources: Fintech Brainfood

Coinbase works the small banks and the big ones at once

Tearsheet's read on Coinbase is worth your time. The exchange wants revenue beyond trading, and it's routing its stablecoin payments stack through banks to get it. Moov offers a channel to more than 1,000 community banks and credit unions that could never build this themselves, though nobody has promised they'll all turn it on. At the other end, Citi is connecting Coinbase to its acquiring and institutional banking platforms. It's one strategy, run from both ends of the market at the same time.

Sources: Tearsheet

An analyst and his four agents

Marc Rubinstein used to dread earnings season as a sell-side analyst. In Net Interest he describes how he now gets through it: the six largest US banks put out more than 250 pages and over eight hours of calls in one week, and four AI agents do the first pass. One compares each release with what the market expected and writes a short brief. Another tracks his long-running themes, such as agentic disruption and private credit. A third predicts the toughest question a well-known bank analyst will ask on the call. It's a refreshingly practical look at where agents save expert time today.

Sources: Net Interest

Sources: Fintech Brainfood · Tearsheet · Net Interest

Sources

Find providers with appetite for your model

Publish one structured RFP and let matching providers respond to it. You stay anonymous until you accept one, and it's free for buyers.

Related reading