Agentic AI Meets Its Missing Layer: Market Access

Two years of research into agentic AI led to the development of systems capable of planning, calling tools, and completing multi-step tasks without supervision. However, financial markets have been automated for decades, with orders routed, payments cleared, and settlements finalized without human intervention. The issue lay not in the technology, but in the assumption that a person was present to authorize each decision.

Recently, a market access layer has been created for agentic AI. This infrastructure allows the software to send decisions down the financial rails that were previously reserved for human authorization. The key question now is where authority should sit within the stack and how far a piece of software should be allowed to reach.

Various layers of financial infrastructure assume the presence of a human operator, from account opening to transaction authorization and dispute resolution. These assumptions break down when the actor is software. Legal requirements for payment orders to be traceable to authorized instructions from account holders or recognized agents pose a challenge for agentic systems.

Gartner projects that by 2028, 33% of enterprise software applications will incorporate agentic AI, up from under 1% in 2024. The industry is working to determine where inside the stack authority should reside and how much software should be permitted to accomplish. A three-layer separation has emerged as the clearest articulation of this split.