AWS Opens 23 Ready-Made GenAI Components: DoorDash and Ricoh Cut Development Hours by 90%

Any team that has built an enterprise AI assistant knows the story. First a week on a RAG pipeline, then PDF parsing, then the scaffolding for the agent, then quality evaluation. Dozens of other teams have already written all of this, and the actual customer problem only gets attention a couple of months later. AWS proposes to put an end to this: the Generative AI Innovation Center, now part of AWS Frontier AI Engineering & Services, has assembled a catalog of 23 open components.

Why Teams Keep Rewriting the Same Thing

AWS names three reasons. First, there is no single place where everything lives: code is scattered across repositories, wikis and drives, and finding ready-made pieces is hard. Second, there is no quality signal: it is unclear whether a component has passed a security review and whether it is maintained, so it is easier to write your own. Third, there is no culture of reuse: what one team built stays with that team.

AWS also points out that even AI-generated code has to be brought to production standards: checked for security, load-tested and maintained. The ready-made components have already been through this work.

What's Inside and Who Uses It

The catalog is split into four groups: document processing, contact centers and voice, agents and chatbots, and RAG and knowledge bases. There are also some interesting standalone items: fine-tuning of the BGE reranker, a persistent semantic cache on Amazon MemoryDB, an agentic workflow for migrating code from C to Java, and a framework for migrating between LLMs.

The main case studies, according to AWS itself:

• DoorDash built a voice self-service contact center on Contact Center GenAI Agent. It claims more than 100,000 calls per day, a response latency of 2.5 seconds, a 50x increase in test capacity and a 50% reduction in development effort using Amazon Bedrock. It reached live testing in two months.
• Lyft used the same component for agentic AI for drivers and riders.
• Ricoh built multi-tenant document processing for health insurers on top of the GenAI IDP Accelerator. Throughput grew from 10,000 to 70,000 documents per month, deployment effort fell from 80 hours to under 5, and customer onboarding shrank from 4–6 weeks to 2–3 days.

Two New Templates for Agents

The Fullstack AgentCore Solution Template (FAST) deploys a secure React frontend connected to Amazon Bedrock AgentCore in under 30 minutes via infrastructure as code. It includes integrations with Memory, Gateway, Code Interpreter and Observability, with authentication through Amazon Cognito. Components are swappable: the identity provider, frontend framework, hosting or agent framework.

The Agentic Chatbot Accelerator is built on Strands Agents and AgentCore and deploys via AWS CDK, with ready-made patterns for authentication, hosting and scaling.

What Deployment Looks Like

Using Contact Center GenAI Agent as an example, the process is as follows: clone the repository, deploy two CloudFormation stacks (the knowledge base and the RAG solution), connect the ready-made Amazon Connect flows and verify everything through Jupyter notebooks. AWS estimates this takes under an hour. The package includes a Quick Sight dashboard for conversation analytics and Bedrock Guardrails for content filtering and hallucination protection. Access to Claude Sonnet and Haiku in Bedrock is required. After experimenting, the stacks should be deleted in reverse order so you don't pay for idle resources.

Limitations and Practical Takeaways

It is worth looking at the numbers soberly: this is an AWS marketing post, and the results come from the company's own case studies. Almost all components are tailored to the AWS ecosystem: Bedrock, Connect, CloudFormation, CDK. If you are not on AWS, portability will be limited, although the patterns and architectures are still useful as a reference.

What to do in practice. If you are building a contact center, document processing or an agentic application on AWS, first check the catalog on GitHub and in AWS Labs rather than opening an empty file. Evaluate a component by repository activity, documentation and security. Even if you don't take the code wholesale, the DoorDash and Ricoh architectures offer a good benchmark of how much can realistically be saved on infrastructure routine and where it is better to direct your effort: data, prompts and business logic.