Artificial intelligence is opening new routes between banks’ legacy cores and the applications built on top of them, expanding what banks can build around systems they already have.

The change reaches beyond chatbots and employee copilots. AI agents retrieve information from banking systems, invoke software functions, move information between applications and complete defined portions of workflows. Coding agents help engineers understand and modify the software connecting old and new systems.

A PYMNTS examination of OpenAI’s GPT-6 Astra looked at computer use, which allows AI to operate existing applications through their interfaces. OpenAI also identifies legacy-system modernization as a financial services use case, including migrating COBOL and other legacy code. At the same time, banking technology providers are making application programming interfaces (APIs) and core functions accessible to AI agents, creating additional ways to connect newer applications with established systems.

A modern banking application typically reaches a core through APIs, which retrieve an account balance, open an account or initiate another defined function. Middleware between systems handles jobs such as authentication, routing and data translation. An AI agent can call those same APIs.

Model Context Protocol, or MCP, adds a way for the agent’s discovery of the functions are available and how to use them. Instead of developers constructing a separate AI integration for each function, an MCP server can present approved capabilities to an AI application as tools.

Mambu’s Core MCP offers an example of how that works in banking. It exposes hundreds of core operations to AI clients. Functions that retrieve information can be made available separately from functions that alter information or execute actions, giving institutions control over what an agent can do.

Source link