AI agents are increasingly capable of analyzing markets, developing trading strategies and executing orders. But for financial institutions, deploying these systems at scale presents a different challenge: ensuring that autonomous activity remains within clearly defined operational and risk limits.
That was the central point raised by Diana Pires, co-founder of agentic finance automation company BYTEX, following her participation in discussions at Ripple Swell 2026, held October 27-29 at The Shed in Manhattan.
Pires has experience across both traditional and cryptocurrency markets. Before establishing BYTEX, she held senior positions at OKX, B2C2 and Crypto.com, following earlier roles at ICE, S&P Global, Fitch and Numerix.
The event brought together companies working across AI, tokenization, stablecoins, payments and financial infrastructure.
One of its sessions, "What Happens When AI Moves Money," featured representatives from Mastercard and Amazon Web Services, examining agentic finance in the context of crypto infrastructure and payments.
For Pires, the discussion reflected a broader shift toward financial services that operate with greater automation and fewer interruptions.
"Every institution we speak to wants to offer AI-driven trading to its clients," Pires said. "Very few can do it on the stack they run today. The models aren't the bottleneck. Permissions, limits and controls are."
Infrastructure gap in automated trading
BYTEX is designed to help institutions introduce automated execution into their existing products without having to develop an entire infrastructure stack internally.
Its open-source system connects with platforms already used by financial companies, while leaving control over funds, permissions, proprietary algorithms and trading strategies with the operating institution.
The company does not have access to those assets or systems. Instead, institutions define the parameters under which agents can operate, including exposure, position sizes and execution limits.
The infrastructure also supports multiple autonomous strategies running within the same environment.
The business model separates access to the open-source code from commercial deployment. Institutions can use the code and pay when moving into production, where BYTEX offers enterprise support, service-level agreements and premium control modules governing agent activity.
The approach reflects a distinction between developing an AI model and deploying it in an institutional trading environment. Even when an agent can identify an opportunity and execute an order, the institution still needs mechanisms to determine which actions it can take and under what conditions.
Continuous markets increase demand for embedded controls
Pires also pointed to the implications of financial markets operating continuously. As tokenization and stablecoin infrastructure expand, financial activity could increasingly take place outside traditional market hours.
In such an environment, processes built around sequential settlement and manual intervention may become harder to maintain. If transactions can be initiated around the clock, institutions will need execution systems capable of applying checks automatically rather than relying on a person to approve every step.
"When there's no close, you can't put a person between every step," Pires said. "The checks have to live inside the execution layer."
She expects the number of autonomous agents carrying out transactions to grow faster than the number of people supervising them. This would make scalable governance a central requirement for institutions adopting agentic finance.
Rather than managing each agent individually, companies will need systems that apply consistent permissions and restrictions across large numbers of automated strategies.
"Agents will keep getting smarter on their own," Pires said. "What institutions need to decide now is the rules they run under."
BYTEX is positioning its infrastructure around that requirement. As AI systems take on more market research and execution tasks, institutional adoption will depend not only on what agents can accomplish, but also on whether financial companies can define, enforce and monitor the boundaries within which they operate.
