6 min read · 15 sources

JPMorgan Chase has drawn a hard boundary across the open banking landscape: outside AI agents cannot touch retail balances. If you build autonomous agents designed to negotiate rates, optimize yield, or execute bill pay on behalf of bank customers, your software is restricted to read-only pipes.

The move lands as consumer tech platforms face steep revaluations. Bumble is in late-stage talks to go private with Ithaca Holdings in a deal pegged near $1 billion. That is a modest premium over its cratered $389 million public valuation, driven downward by declining numbers of paying users and weak engagement across younger demographics. But while consumer apps struggle with customer churn, enterprise and financial engineering teams are battling a different problem: managing the perimeter and infrastructure costs of autonomous software.

Ramp burned $1.5 million on model tokens in a single week before building an internal context layer to rein in runaway AI costs.

JPMorgan cuts off external AI agents at the API gateway

Source: cnbc.com ↗

JPMorgan announced that it is actively blocking autonomous third-party agents from signing into retail accounts, scheduling transfers, or executing payments. External software, including agents like Muse and Instinct, will remain confined to read-only access through licensed open-banking aggregators like Plaid. Direct transactional API access is off the table.

CEO Jamie Dimon cited operational and security liabilities, flagging the prospect of “thousands” of rogue automated agents running amok - a concern heightened by the Hugging Face breach involving OpenAI credentials. Dimon warned of the legal nightmare surrounding unclear liability transfers: if a user grants an autonomous bot authorization to sweep cash between accounts, and that bot hallucinates an instruction or dumps funds into an insecure protocol, the operating bank gets stuck holding the customer service bag.

Autonomous deposit migration also threatens the traditional retail banking moat. If third-party software continuously shifts customer deposits to whichever institution pays five basis points higher, sticky zero-interest checking account balances evaporate. JPMorgan is protecting that margin by building its own walled-garden agent: an internal tool dubbed Smart Cash designed to forecast liquidity and rebalance cash between internal accounts. The pattern is familiar: banks will gladly automate operations internally, but will reject outside write-actions until settlement frameworks explicitly assign legal liability for automated mistakes.

Agent payments are reinventing rails cards already built

Source: finextra.com ↗

The friction at JPMorgan highlights a fundamental shift in payment system architecture: the migration from attended human interactions to headless, automated agent settlement. Payment networks were designed around a physical human clicking checkout, tapping an EMV chip, or completing a 3D-Secure challenge. When an automated procurement bot orders warehouse supplies or an assistant rebooks travel mid-flight, those human-in-the-loop auth primitives break down completely.

As former Meta executive David Marcus pointed out, moving money without human interfaces requires connecting programmatic stablecoins, embedded fiat wallets, and dynamic card tokens behind machine-readable APIs. The current wave of “agentic commerce” isn’t inventing new financial rails from scratch; it is rediscovering the strict authorization rules, spend limits, tokenized scopes, and fraud chargeback mechanisms that the credit card networks spent half a century codifying. The immediate work for fintech infra teams is building machine-delegated authorization policies that pass compliance muster without demanding a human tap “approve” on a mobile device for every microtransaction.

Ramp tackles token sprawl with Glass

Source: x.com ↗

Spend management platform Ramp released details on Glass, an internal operating layer built to preserve enterprise prompts, sessions, and workflows across LLM-powered tools. The system was born out of an infrastructure nightmare: Ramp’s engineering team previously burned through $1.5 million in tokens during a single seven-day span due to runaway queries, redundant prompt contexts, and lack of cross-team coordination.

Glass acts as a centralized AI workspace and semantic cache connected to Linear, Slack, Notion, and Gong. Instead of hundreds of employees independently constructing unoptimized prompts and injecting massive raw system contexts into closed-box chats, Glass lets teams publish and reuse structured skills and sessions. Over several weeks of dogfooding, 323 internal employees generated 6,500 edits. Automated feature-request ingest jumped from 300 to 20,000 per month, while generating over 5,000 automated customer outbound messages. Critically, centralizing context routing cut Ramp’s blended AI infrastructure bill by 30%. If your engineering org runs headless agents or internal assistants, deploying shared context caches and project-level token quota systems has stopped being an optimization exercise - it is required financial discipline.

Monzo hunts for a $13B exit at a fraction of Revolut's multiple

Source: lex.substack.com ↗

UK digital bank Monzo is seeking an exit at a $13 billion valuation, engaging private equity giants Advent and CVC after earlier acquisition discussions with Brazil’s Nubank stalled. Monzo’s fundamentals are in line with a mature credit institution: FY2026 revenue climbed 39% year-over-year to £1.7 billion, with pre-tax profit expanding 44% to £87 million. The growth engine is classic spread banking: deposits rose 55% to £25.7 billion, powered by a 3.5% APY savings tier.

The disconnect is in the multiple. Monzo was valued at $4.5 billion in late 2021, and an internal share sale pushed that to $5.9 billion last year. A $13 billion exit values the company at roughly 6x trailing revenue. Contrast that with Revolut, which traded at a $115 billion valuation in a recent secondary round - nearly 19x revenue. The difference lies in the balance sheet. Monzo is valued like an agile balance-sheet retail bank tethered to the UK interest rate cycle. Revolut is priced like an asset-light, multi-currency transaction engine driven by cross-border interchange, foreign exchange spreads, and subscription SaaS fees.

Carta unlocks €1T Luxembourg fund ecosystem

Source: carta.com ↗

Cap table and private markets infrastructure provider Carta secured Specialised PSF authorization from Luxembourg’s financial watchdog, the Commission de Surveillance du Secteur Financier (CSSF). The license allows Carta to offer fully compliant fund administration, investor onboarding, and transfer agency operations within the Grand Duchy.

This is a targeted infrastructure play for Carta’s private-capital ERP. Luxembourg is Europe’s primary investment vehicle hub, holding approximately €1 trillion across 5,000 alternative funds. Carta’s software stack can now natively manage alternative legal entities such as the SCSp (Special Limited Partnership) and RAIF (Reserved Alternative Investment Fund), which are the standard structures used by cross-border private equity, venture, and private credit firms. Integrating investor KYC, capital calls, and transfer agency ledgers into a single core database reduces reliance on legacy, bespoke European service providers that still calculate drawdowns and distribute investor capital statements using manual spreadsheets.

Airtable sets post-acquisition roadmap under Bending Spoons

Source: airtable.com ↗

Following its takeover by Bending Spoons, Airtable laid out a roadmap to 2027 alongside an aggressive internal headcount reduction. Under the ownership of the software consolidator, the company is stripping back bloated product experiments to focus on enterprise database performance, large-base scale, and low-latency API syncs.

On the technical front, Airtable is pivoting toward becoming a collaborative system of record for both human operators and autonomous systems. That includes expanded enterprise audit logs, fine-grained access control for external agents, an upgraded Model Context Protocol (MCP) server integration, and tools to monitor AI credit burn. For engineers who rely on Airtable as a rapid prototyping datastore or internal CRM, the shift guarantees fewer decorative UI features, cheaper operational overhead, and a tighter emphasis on stable query latency.

Product shifts and integrations: Splitwise, Klarna, and rate dynamics

Source: finextra.com ↗

A quick run of notable deployments across consumer fintech and merchant payments:

• Receipt OCR parsing goes native: Splitwise launched Itemize, a free in-app mobile tool that uses image recognition to parse physical receipts. The feature handles item identification, automatically splits tax and tip proportionally among table attendees, and does not require every participant to maintain an active Splitwise account to claim lines.
• PSP rails expand BNPL coverage: Global payments processor Ecommpay has integrated Klarna directly into its merchant gateway. The single-API connection gives regional enterprise e-commerce merchants access to Klarna’s installment plans and consumer financing without having to maintain separate integrations or fragmented settlement reconciliation pipelines.