According to PYMNTS.com, agentic AI systems, which can plan and execute multi-step tasks, have rapidly moved from prototypes into enterprise roadmaps in 2024. Their October CAIO Report shows a dramatic shift: while 85% of CFOs had no implementation plans in May, follow-up data by July showed agentic AI moving into early, tightly-guarded test runs at a growing number of firms. These agents are first being deployed in “money-adjacent” workflows like invoice matching and dispute triage. PYMNTS CEO Karen Webster warns the shift is from automation to autonomy, creating a central trust problem, as data shows 97% of CFOs understand AI can act alone but only about 15% are considering using it. Key developments include new protocols like OpenAI and Stripe’s Agentic Commerce Protocol and security initiatives like Worldpay’s partnership with Trulioo for agent identity validation.
The Trust Paradox
Here’s the thing about that data: it reveals a massive, almost comical, gap between understanding and action. Basically, every finance exec gets that this tech can run on its own. But almost none of them are willing to let it off the leash. And you can’t really blame them. We’re talking about money. An error in a collections support bot is one thing. An autonomous agent making a bad call on a large transaction? That’s a regulatory and reputational nightmare. So the early playbook makes total sense: start where mistakes are visible and reversible. It’s a slow, cautious crawl toward autonomy. But that caution itself is a business risk. The firms that figure out the governance first will move faster later.
Protocols Are The Real Prize
Now, this is where it gets really interesting. Karen Webster hits the nail on the head by calling protocols the leverage point. Think about it. The big question isn’t just *if* an AI agent can buy something for you. It’s *how* it chooses what to buy and *where* it chooses to pay. Does it default to Amazon? Does it search for the best price across ten merchants? Does it use Apple Pay, a Visa token, or some new wallet? The protocol that guides that decision owns the customer relationship and, ultimately, the economics. OpenAI and Stripe’s move is clever—it tries to slot agents into existing payment rails. But other approaches with “signed intents” imagine a world where you delegate authority to an agent to act on your behalf. That’s a whole new level of complexity. Who’s liable if it goes wrong?
The New Security Battlefield
And speaking of things going wrong, the security implications are terrifying in a new way. The mention of the Anthropic Claude attack is chilling. We’re not just talking about data leaks anymore. We’re talking about AI systems being manipulated to autonomously run most of a cyber-espionage campaign. That’s a force multiplier for threats. So the industry’s response—things like Trulioo’s Digital Agent Passport and Visa’s tokenization—isn’t just nice-to-have. It’s an attempt to build an identity layer for non-human actors. We need to know who built an agent, who it works for, and what it’s allowed to do. Otherwise, the entire system becomes a playground for fraud. The need for real-time monitoring that keeps up with AI speed isn’t a future problem. It’s a right-now problem.
Who Really Ends Up In Control?
So, what’s the endgame? The article hints at a huge power shift. If autonomy increasingly lives inside third-party platforms (think Microsoft’s “agent-operated” vision or an AI platform’s commerce protocol), then banks and payment providers risk becoming dumb pipes. Their role shifts from customer-facing service providers to backend risk managers and compliance overseers. The “frontier” becomes whoever owns the agent interface. That’s why the governance models and vendor risk management are suddenly so critical. It’s a fight for relevance. Banks have the trust and regulatory know-how. Tech platforms have the AI and the user interface. The next few years will be about which side can adopt the other’s strengths fastest. The agents are just the tools. The real battle is for control.
