The CFPB finalized US open banking under Section 1033 in October 2024, a federal court enjoined enforcement a year later, and the rule is now back in reconsideration. The data layer that survives that whiplash is the deliverable, not the API. Build for reversals, not launch day.
Since October 2025, euro credit transfers run a Verification of Payee check before they settle, and a name-IBAN mismatch triggers an alert. It hits US treasuries paying EUR vendors and payroll as hard as it hits EU banks, and the same name checks are a fair bet to reach FedNow and RTP next.
FedNow matched RTP at a $10M network limit in November 2025 and added new risk rails - a receiver-account pre-check pilot, ScamClassifier reporting, an Exception Resolution Service, and 1,500+ banks live. Instant is treasury cash now, and batch-era controls just got a 24/7 bypass.
AI agents are arriving faster than most banking career ladders can adapt, and the market is already rewarding people who work with them. The career moat was never wiring up another API call. It is the judgment calls agents cannot own - which rail to trust, how much latency to spend, who the regulator will be.
The API worked. The books did not. A clean REST interface with pretty JSON still leaves accounting stitching exports for three days to answer one question - what happened to this payment across every rail and every failure path. ERP-grade APIs treat the contract as a shared language with finance.
Stripe, OpenAI, and Google shipped protocols that let AI agents complete purchases inside a conversation. The rails are already fast enough. The problem is that fraud detection was architected around a human clicking buy on a trusted surface, and agents break that assumption.