Back to News
InnovationAI Understanding briefing

Duco leverages Amazon Bedrock AgentCore to accelerate financial data automation

Duco rebuilt its agentic AI stack on Amazon Bedrock and Bedrock AgentCore, cutting context overhead by 90% and delivering up to 10× faster reconciliation deployments for tier‑one banks.

4 min readRead the primary source
Source-provided image accompanying Duco leverages Amazon Bedrock AgentCore to accelerate financial data automation
Primary-source documentSource recorded
Publisher
aws.amazon.com
Source type
Primary document — an official announcement, paper, filing, or first-party page we read directly.
ContextUnderstand this in 60 seconds

Key terms

MCP (Model Context Protocol)
An open protocol that lets AI applications connect to external tools, data sources, and context providers in a standard way.
Memory (Agent Memory)
Stored context an AI agent uses across steps or sessions to improve continuity.
Context Window
The maximum amount of input tokens a language model can process at once.

What happened

Duco, a financial‑services operations platform serving over 200 institutions, migrated its AI workload from traditional machine‑learning models and a LangChain‑based stack to Amazon Bedrock and the newly announced Amazon Bedrock AgentCore. The move introduced a multi‑tenant, isolated architecture where each bank’s data and reasoning remain separate. Duco also introduced a skills architecture that loads tool descriptions on demand, reducing context size by roughly 90%. The upgraded stack powers two high‑impact workflows: onboarding new reconciliation processes and handling exception‑driven quality gates. One global bank reported a 70% reduction in onboarding costs and a 20%+ increase in straight‑through processing, while overall reconciliation deployments became ten times faster.

Duco’s platform processes roughly 20 billion transactions each month for more than 200 clients, including several top‑tier banks and asset managers. To meet growing data‑processing demands, the company transitioned from custom machine‑learning pipelines to Amazon Bedrock’s managed foundation models and the AgentCore runtime, which provides built‑in authentication, isolation, and monitoring.

The new architecture enforces strict per‑tenant isolation at every layer, ensuring that each bank’s data, context, and reasoning remain separate. This is achieved through identity‑based authentication and a gateway that only exposes tool names, with full tool details fetched on demand via the MCP (Model‑Control‑Plane) discovery mechanism.

Duco introduced a skills architecture that moves large tool‑description payloads out of the static prompt context. Instead of loading ~200,000 tokens of tool metadata at startup—causing context overload—the system registers tools at boot, then uses find_tool and invoke_tool calls to retrieve details only when needed. This reduces context size by about 90% and improves response quality.

Two critical workflows were rebuilt on the new stack: (1) onboarding of new reconciliation processes, which previously required weeks of manual effort, now sees a 70% cost reduction and a 10× speed increase; and (2) handling of daily exception‑driven quality gates, where agents can reason over complex, variable data paths, improving straight‑through processing rates by over 20% for a global bank.

Source details: aws.amazon.com

Why it matters

The announcement shows that large‑scale, regulated financial institutions can adopt foundation‑model‑driven agents without compromising strict data isolation or compliance requirements. By using Amazon Bedrock’s managed foundation models and AgentCore’s agent runtime, Duco avoided the overhead of building and maintaining its own model infrastructure, cutting deployment time and operational costs. The dynamic tool‑discovery approach solves a common production challenge— overload—by loading only needed tool metadata, a technique that could be replicated across other enterprise AI deployments. The reported business impact—significant cost savings, faster onboarding, and higher straight‑through processing—demonstrates tangible ROI for AI‑driven automation in finance, a sector traditionally cautious about AI adoption. Moreover, the architecture’s ability to swap models per task offers a path to future‑proofing as model capabilities evolve, reducing vendor lock‑in and enabling cost‑effective scaling.

Regulated financial institutions have long been hesitant to adopt generative AI due to concerns over data privacy, isolation, and auditability. Duco’s implementation demonstrates that managed services like Bedrock can meet these stringent requirements while delivering operational benefits.

The dynamic tool‑loading pattern addresses a well‑known limitation of large language models—finite context windows—by ensuring agents only carry the minimal necessary information, a practice that can be generalized to other enterprise AI use cases.

By decoupling model selection from the agent logic, Duco can swap foundation models per task, optimizing for cost and performance without disrupting service. This flexibility is crucial as model capabilities evolve rapidly.

The reported business outcomes—substantial cost savings, faster onboarding, and higher processing efficiency—provide concrete evidence that AI‑driven automation can move beyond pilot projects to production‑scale impact in finance.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Interactive Concept Check+10 Points
What is AI? Quiz

A route planner searches possible journeys using explicit rules. What does this illustrate about AI?

What to watch next

Future updates from Duco on broader rollout across its client base, any regulatory audits or compliance certifications tied to the Bedrock deployment, and Amazon’s roadmap for AgentCore features such as enhanced memory or cross‑tenant collaboration. Observers should also monitor whether other financial‑services platforms adopt similar multi‑tenant agentic architectures and how the industry addresses auditability and risk‑management standards for foundation‑model‑driven agents.

Duco’s future announcements on expanding the agentic stack to additional financial workflows or to more of its client base.

Regulatory feedback or certifications that validate the compliance posture of Bedrock‑based agents in highly regulated environments.

Amazon’s roadmap for AgentCore, especially features that enhance memory, cross‑agent communication, or tighter integration with AWS security services.

Adoption signals from other fintech or enterprise SaaS providers that may replicate Duco’s multi‑tenant, on‑demand tool architecture.

Related guides & quizzes

Found this useful?