Duco processes more than 20 billion transactions every month for over 200 financial institutions, including seven of the world’s top 20 banks. We see this financial data automation shifting from “predicting labels” to “understanding business rules and executing workflows,” with AI agents—AI systems capable of calling tools and completing tasks step by step—entering core business operations.

What this is

Duco originally used traditional machine learning models to automate more than 20,000 data workflows for banks, including information extraction and result publishing. But these models depended heavily on large volumes of historical data and required repeated customization and retraining whenever data formats or business rules changed.

Duco later adopted Amazon Bedrock to access foundation models, or large language models, and began using Amazon Bedrock AgentCore to build, connect and run AI agents. AgentCore handles identity authentication, access control, operational tracing and performance evaluation, enabling agents to connect to banks’ internal tools and data while maintaining customer data isolation.

Industry view

The core of this development is not larger model parameter counts, but AI agents beginning to assume backend workflows previously handled through coordination among multiple software systems and human workers. Duco says the new architecture has shortened deployment times and reduced onboarding costs for new clients. For banks, this means AI projects may no longer require a separately trained model for every workflow.

The risks are equally clear: financial operations demand extremely high accuracy, explainability and audit accountability. If an agent misinterprets a business rule, the error can compound across payment, settlement or reporting chains. AgentCore provides operational and governance infrastructure; it does not automatically eliminate model hallucinations—plausible-looking but incorrect generated information—or workflow design risks. AWS is both a cloud service provider and a model platform, so Duco’s adoption of its stack further ties critical business capabilities to a single vendor ecosystem.

Impact on regular people

For enterprise IT: The purchasing priority for enterprise AI is shifting from chat tools toward auditable backend execution. Whether a system can integrate with existing data, permissions and workflows will determine project success more than the quality of a demo.

For individual careers: Repetitive work such as reconciling data, preparing reports and applying rules is more likely to be taken over by agents. Employees’ value will increasingly center on defining workflows, handling exceptions and reviewing results rather than manually completing every step.

For consumer markets: Banks may process data faster and more consistently, while agentic automation will expand the capacity to scale and personalize financial services. Consumers may not directly perceive changes in the underlying models, but they could experience faster or more standardized experiences when opening accounts, making payments and handling disputes.