Duco migrated its AI stack from traditional machine‑learning models to Amazon Bedrock and Bedrock AgentCore, swapping a home‑grown LangChain/LangGraph agent layer for a fully managed runtime while preserving strict per‑tenant data isolation. Engineers care because the shift shortens time‑to‑value, reduces the burden of operating custom infrastructure, and aligns with the compliance posture demanded by tier‑one banks.
Why Bedrock Replaced Legacy Models
Classic models required large historical datasets, struggled with novel data formats, and needed extensive per‑use‑case tuning. Bedrock supplies a catalog of foundation models from multiple providers through a single API, letting Duco select the most suitable model per task without rebuilding pipelines. Security and compliance are baked in: customer data remains inside the AWS environment, is not used to train the models, and can be accessed via private VPC endpoints, a critical factor for financial data handling.
AgentCore as a Managed Runtime
Duco’s existing agent code, built on LangChain and LangGraph, runs unchanged on AgentCore because the service is compatible with those primitives. By offloading the runtime to AgentCore, the engineering team eliminates the need to provision, patch, and scale custom compute for agents. The service also bundles authentication, access controls, tracing, and evaluation, allowing teams to focus on agent logic rather than backend plumbing.
Multi‑tenant Isolation by Design
Duco delivers its platform as a SaaS offering to over 200 financial clients, each requiring strict isolation. The architecture enforces tenant separation at the service level, with private endpoints ensuring that data never traverses public networks. This design satisfies the governance and compliance frameworks expected by regulated institutions and reduces the risk of cross‑tenant data leakage.
“The classical prediction models worked to an extent, but they weren’t agile or flexible enough. Traditional machine learning models can only amplify what they’ve already learned — they can’t help with the next step or find new problems.”
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
Teams should evaluate whether a managed agent runtime like AgentCore can replace custom orchestration layers, especially when existing code aligns with supported frameworks. Verify that private VPC connectivity meets isolation requirements for regulated data. Leverage Bedrock’s model catalog to reduce the engineering effort of model selection and training. Finally, monitor the operational metrics provided by AgentCore to ensure that the managed service meets performance and compliance SLAs.


