Money20/20 Vegas 2026: Coder's James Frey on AI Governance and Infrastructure

Quick Summary

The journey toward full AI integration is proving to be a longer and more complex process than many in the financial sector initially anticipated. James Frey of Coder explains at Money20/20 Vegas 2026 that as institutions move beyond the initial hype, the focus is shifting toward governance, observability, and the underlying infrastructure required to support AI agents safely and effectively.

What are the primary hurdles for banks adopting AI agents?

According to James Frey, there are two critical pillars that financial institutions must address to make AI agents work correctly within a regulated environment. The first is governance, which involves the ability to impose specific policies that ensure the agent operates within the parameters set by the bank. The second is observability or audit. This provides the capacity to look throughout the entire lifecycle of how an agent builds code, allowing the institution to understand what the agent has done, why it made those decisions, and what its next steps will be.

How is the AI transformation journey evolving?

Many organisations are realising that their AI transformation will take longer than first thought. Consequently, their priorities are shifting toward achieving better choice, control, and consistency. James Frey notes that institutions are looking for ways to de-leverage and de-risk their partnerships. By doing so, they ensure they have the flexibility to make future pivots and achieve cost optimisations as the technology matures. This involves a multi-layered approach to security, including:

• Basic access controls to ensure tools have the right data context.
• Policy enforcement based on the specific type of user interacting with the agent.
• Setting strict boundaries on what an agent is allowed to do on a user's behalf.

Why is traditional infrastructure like VDI insufficient for AI?

A common misconception is that a Virtual Desktop Infrastructure (VDI) can house these new tools, but James Frey argues this is not the case. AI agents require more than just a remote desktop: they need a full, secure execution environment. This environment must be packaged with all the necessary tools, dependencies, configurations, and context required to perform work accurately. A standard VDI simply does not support this level of integrated functionality.

What is the recommended path forward for financial developers?

The advice for institutions, ranging from private equity and asset management to large global banks, is to apply the lessons learned from cloud-native computing over the last two decades. James Frey suggests that the most effective move is to transition all inner loop development environments away from local laptops. Instead, these should be moved into cloud-hosted development environments, ideally situated inside the institution's own private infrastructure. This shift allows for better management of tools and a clearer viewpoint on how AI impacts operating processes and people.

FF NEWS TAKE:

The transition from experimental AI to industrial-grade AI agents requires a fundamental rethink of developer infrastructure. As James Frey highlights, the reliance on local hardware or basic virtual desktops creates significant gaps in governance and security. By moving to cloud-hosted environments, financial institutions can gain the observability required to satisfy regulators while providing the high-context environments that AI agents need to be productive. This move toward centralised, policy-driven development is likely to become the standard for any bank serious about scaling its AI capabilities without compromising on risk management.