The Trust Gap Holding Back the Agentic Economy

There’s a lot of excitement right now around autonomous AI agents. You hear it everywhere, agents booking travel, managing portfolios, negotiating contracts, even paying for services without you lifting a finger.

Companies like OpenAI and Anthropic are pushing capabilities forward at a pretty wild pace. However, we’re now at the stage where capability isn’t the bottleneck anymore, trust is.

The big issue is whether banks, asset managers and treasury teams actually have the setup to trust these systems with real financial tasks, without things going off the rails in terms of control, accountability or compliance.

This trust chasm could be why a Deloitte survey of over 3,300 finance and accounting professionals found that only 13.5% said their companies are actually using agentic AI currently for finance and accounting tasks, with trust being cited as the biggest hurdle to adoption. 

The Illusion of Readiness

A lot of the conversation today is still stuck on whether agents are smart enough. That’s missing the point. The real issue is whether the systems they interact with, banks, payment networks, compliance layers, are ready to deal with non-human actors. Right now, they’re not.

Traditional finance is built around identity frameworks like KYC, where a person is verified, monitored, and held accountable, but agents don’t fit neatly into that model. They need their own identity, their own permissions, their own constraints. You can’t just bolt that onto human systems and hope it works. And institutions know this. 

Loading tweet...

View Tweet

There’s also growing unease about how to safely onboard autonomous actors into financial systems.

Research from Keyfactor found that 86% of security professionals believe AI agents require entirely new forms of dynamic digital identity. That’s basically the industry admitting the current setup isn’t fit for purpose.

So while the tech demos look impressive, there’s this quiet hesitation underneath. Banks don’t want to let something transact unless they know who, or what, is actually responsible when things go wrong. And right now, that accountability chain is pretty blurry.

When Speed Outpaces Control

Another thing people underestimate is just how fast this all moves. Agents don’t operate on human timelines, they don’t wait for approvals or batch processing windows. They make decisions in milliseconds, potentially thousands of them, all day, every day. That changes the entire risk model.

Compliance today is mostly periodic. Reviews happen after the fact. Limits are static. Humans step in when something looks off. That doesn’t work in an agent-driven system. You need continuous oversight with real-time identity checks, mandate validation, fraud detection, and automated escalation. Compliance has to become always-on, just like the agents themselves. Otherwise, small errors cascade.

And then there’s the question of responsibility. If an agent makes a bad decision, who’s on the hook? The user? The developer? The model provider? The platform hosting it? Regulators haven’t fully answered that yet, and you can feel the uncertainty. Without clear lines of accountability, institutions are going to stay cautious. 

Trust Is More Than Settlement

If there’s one lesson from traditional payments that still holds, it’s that moving money isn’t the hard part, trust is. Payments infrastructure evolved to include identity checks, fraud controls, dispute resolution, chargebacks, essentially all the messy human stuff that happens when things go wrong. That’s what made digital commerce work at scale.

Crypto brought something new to the table, programmability, transparency, instant settlement. That’s useful, especially for machine-to-machine transactions. But there’s been this assumption floating around that finality equals trust, which it doesn’t.

Loading tweet...

View Tweet

In an agentic world, mistakes are inevitable. Agents can misunderstand instructions, act on incomplete data, or get manipulated through things like prompt injection. So the system needs ways to reverse, dispute, and compensate, not just execute perfectly.

The risks are also evolving. It’s not just fraud in the traditional sense. You’re looking at impersonation, mandate abuse, collusion between agents, fake counterparties, compromised wallets. This means the defenses have to evolve too. Dynamic identity, behavioral monitoring, spend limits, audit trails, human-agent binding, all of that needs to be built in from the start, not patched in later.

Closing the Gap Before It Widens

The tricky part is that user behavior is already moving faster than the infrastructure.

People are starting to trust AI agents in their daily lives, sometimes more than they probably should. Meanwhile, institutions and regulators are still trying to figure out the basics like identity, compliance, accountability, and that gap creates risk.

If agents gain the ability to manage money, trade assets, or control treasury functions before the trust layer is fully in place, it only takes one major failure to set things back. We’ve seen that pattern before in tech, early hype, followed by a high-profile blow-up, then a wave of skepticism.

Loading tweet...

View Tweet

The more realistic path is a bit slower, maybe less flashy. Start with specialized layers, platforms that handle identity, risk, and compliance for agents. Over time, those capabilities get baked into standard frameworks, so developers aren’t rebuilding trust infrastructure from scratch every time.

At the end of the day, the agentic economy isn’t just about smarter AI. It’s about building systems that can trust those agents to act responsibly, and make things right when they don’t. At the moment, that trust is still a work in progress.

About the Author

Chandler Fang is the co-founder of t54. Prior to t54, Chandler was the Lead Product Manager of Payments at Ripple. Before Ripple, as VP of Product Management, he was in charge of JP Morgan’s Cash Flow Forecasting AI product. He also served as a Venture Partner at FoundersX Ventures, investing in DeepTech and FinTech for close to a decade. Chandler holds an MS in Financial Engineering from UC Berkeley Haas.

Linkedin: https://www.linkedin.com/in/znfang/