Patrick Collison put it plainly at Stripe’s Sessions conference in April: “AI is the biggest platform shift for the economy since the internet, and in the not-too-distant future agents will account for most transactions online,” according to Stripe’s newsroom release. Forbes contributor Dara-Abasi Ita built an entire argument around the implications of that statement, and it deserves attention because it names a structural problem most coverage of agentic AI avoids: the financial plumbing was designed to prevent exactly what agents need to do.
The Anti-Bot Stack
Bill.com, Brex, Ramp, Coupa. These became multi-billion-dollar businesses by making humans faster at moving money. Every one of them was architected around a core assumption: a person initiates each transaction. Accountants approve invoices. Controllers sign off on wires. CFOs click the button. The software made the click faster and more accurate, but the human remained the bottleneck.
That bottleneck was also the security model. As Leor Ceder, CEO of agent-native payouts startup Payouts.com, told Forbes: “Legacy platforms were built for humans alone. From databases to security protocols, the assumption is always that a human is in the loop.” These platforms were explicitly engineered to stop bots and flag non-human activity as potential fraud. Authentication, behavioral monitoring, anomaly detection: every layer is designed to detect and block the kind of autonomous actor that agents represent.
This creates an architectural contradiction. The same fraud detection systems that make legacy fintech safe are the systems that prevent agent-based workflows from functioning.
The Data on Demand
The demand side is not speculative. A May 2025 Wolters Kluwer survey of 392 finance leaders found that 6% were already using agentic AI, with another 38% planning adoption within 12 months. That puts projected 2026 usage at 44%. Separately, Deloitte’s January 2026 CFO Signals survey found that 54% of finance chiefs name AI-agent integration a digital transformation priority for 2026, as reported by CFO Dive.
HPE CFO Marie Myers told CFO Dive that her team is now deploying “far more agentic AI” inside finance after piloting an internal agent co-developed with Deloitte. The tool, CFO Insights, is cutting HPE’s financial reporting cycle by about 40%.
Nearly half of finance functions want agents. Their existing software was built to stop them.
The Retrofit Problem
Ita’s argument against the “bolt-on” approach is structural, not tribal. FINRA’s 2026 oversight report treated autonomous AI agents as a distinct supervisory category from chat-style generative AI for the first time, as reported by WealthManagement.com. Regulators are already distinguishing between a copilot that suggests and an agent that acts. The compliance controls are different. The liability frameworks are different. The authentication models are different.
An agent-native stack exposes every function through machine-readable protocols, issues cryptographic identities to agents, signs mandates with keys instead of clicks, and runs compliance checks as background services that humans approve by exception. The Model Context Protocol, originated by Anthropic and now governed by the Linux Foundation, has become the closest thing to a common wire standard for agent-to-tool communication.
Era, a personal finance startup, became the first personal finance connector in Anthropic’s Claude directory in early May, built on MCP. CEO Alex Norcliffe framed the shift in the FFNews launch announcement: “AI has been able to talk about your money for two years. The interesting thing is when it can start fixing it.” Era supervises over $250 million in user assets and operates an SEC-registered RIA subsidiary as the adviser of record for paid users.
Where the Pressure Lands
The incumbents are not passive. FIS partnered with Anthropic to bring a Financial Crimes AI Agent to its banking platform. Microsoft’s Wave 1 2026 release for Dynamics 365 Finance added an MCP gateway and an autonomous Payflow Agent. Stripe and Visa have invested in real-time payments and tokenization infrastructure.
But the structural question remains: can systems designed to detect and block autonomous actors be retrofitted to welcome them? The authentication models, fraud detection heuristics, and approval workflows are not cosmetic features. They are load-bearing architecture. Replacing them requires rebuilding the foundation while the building is occupied.
Today’s $84.5 million in vertical agent funding (Samaya AI, Autonomize AI, Scamnetic) lands in exactly this context. Investors are not just funding agents that operate within financial systems. They are funding agents that replace the human checkpoints those systems were built to protect. Whether the legacy stack bends or breaks under that pressure will determine the next decade of fintech market share.