AWS has released two reference implementations for teams trying to turn an AI-driven software-development process into working systems. Both use Amazon Bedrock AgentCore with coding agents, while keeping human review around consequential architecture and security decisions.
The first blueprint reads SQL database schemas and generates Mermaid entity-relationship diagrams. It runs the agent through AgentCore Runtime and demonstrates how a coding assistant can convert an existing technical artifact into maintainable documentation. The second analyzes code security with several specialized agents. It combines AgentCore Gateway and memory with external tools so separate components can investigate findings and preserve context.
AWS frames the examples as the construction phase of an AI-Driven Development Lifecycle, or AI-DLC. In that model, AI handles routine execution and proposes code or plans, while people oversee requirements, trade-offs, and approval. The repositories include deployment instructions, architecture choices, and code patterns rather than only a conceptual description.
The blueprints do not establish that agent-generated code is safe by default. A multi-agent security analyzer can itself miss vulnerabilities, and generated diagrams can reproduce errors in the source schema. Their practical value is as inspectable starting points for teams already considering AgentCore, Kiro, or Claude Code. Organizations still need their own evaluations, access controls, testing, and review process before applying either pattern to production repositories.