
Easily create AWS Serverless applications using AI Agent steering Files
Create and deploy a secure and scalable architecture for application using steering files with AWS serverless
AI coding assistants can generate entire applications in minutes, but how do you ensure the output follows a secure, scalable, well-architected pattern? Steering files let you define your preferred tech stack once and reuse it everywhere. They follow the DRY principle: tell your AI assistant how you build once, and every project starts from a consistent, vetted foundation.
AI coding assistants like Kiro, Claude Code, and others generally support steering files, agent instructions or rules. Example references: Kiro Steering Files , Claude code (CLAUDE.md file).
A steering file is simply a text markdown document that instructs an AI Agent on project-specific preferences, infrastructure-as-code frameworks, build tools, and deployment patterns. When you ask your AI Agent to generate code or infrastructure, it will reference the steering file to maintain consistency with the defined tech stack.
Recommended sections: Use the following key sections in your steering file when deploying to AWS.
- AWS Configuration: Account # or CLI profile
- Frontend
- Backend
- IaC
- Build Automation
- Guardrails
Steering file contents
You can use AI to generate a good steering file. For example, you can provide the following prompt in your favorite AI Agent. Feel free to customize this to your preferences:
You are an experienced AWS Solutions Architect. Generate a concise tech stack steering file that I can use to build and deploy an application on the AWS serverless stack that follows well architected principle. Include the following key sections: AWS Configuration: Account # or CLI profile, Frontend, Backend, IaC, Build Automation, Guardrails.As an example, in 2026, this AWS serverless is my favorite tech stack for building applications ( see aws-serverless-steering-file.md ).
Project Tech Stack Configuration
## AWS Configuration- Region: us-west-2- AWS Account #: 1234576890- CLI Profile: dev (use a named profile instead of default for better security)
## Infrastructure as Code- Framework: AWS CDK (Python)- Alternative: Terraform can be used instead of CDK- Python virtual environment at repo root- Generate a mermaid architecture diagram
## Frontend- Framework: React with CloudScape Design System- Alternative: Vue, Angular, or other frameworks based on team preference- Hosting: S3 + CloudFront with Origin Access Control (OAC)- Bucket policy configured for CloudFront access only
## Backend Services- API: Amazon API Gateway with Cognito Authorizer- Authentication: Amazon Cognito User Pool- AI/ML: Amazon Bedrock for inference- Compute: AWS Lambda (Python runtime)- Alternative runtimes: Node.js, Go, Java, or .NET based on team expertise- Data: Prefer Amazon DynamoDB for backend data storage, verify the proposed schema during application design
## Build AutomationCreate Makefile with these targets:- `make cdk`: Deploy CDK infrastructure using named profile- `make npm`: Build frontend, push to S3, invalidate CloudFront cache- `make backup`: Create .tar.gz archive and upload to backups S3 bucket
## GuardrailsAlways apply the following guardrails:- Any S3 buckets must be private- API Endpoints must be authenticated using JWT- Use least privellege Execution Role for Lambda- Always run checks for IaC: Checkov and cfn-lintWhether you follow a vibe coding approach or try Kiro's Spec mode, the above steering file will guide your AI Agent to implement your application with the following serverless architecture:

This stack eliminates operational overhead, and provides a solid foundation for building well-architected, secure, and scalable applications. The front-end is a modern React UI hosted on CloudFront, the backend is fully serverless with API Gateway and Lambda, and IaC is handled with AWS CDK. Deployment is automated through simple Makefile targets.
Steering File Hierarchy and Inclusion modes
Some Agents such as Kiro supports a two-level hierarchy with project-level and global steering files, letting you keep a consistent baseline while tailoring instructions per project. Similarly, Claude code supports CLAUDE.md scopes at a User and Project location. You might keep separate project-level steering files for a frontend-only project, an IaC module, or a microservice with its own deployment pipeline.
Closing thoughts
LLM outputs are non-deterministic. A steering file improves consistency, but it does not guarantee the AI will follow every instruction 100% of the time. Always review the generated artifacts, and CLI commands the AI Agent is going to execute. To get started, create a steering file in your next project and iterate on it as you build.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article