SYSTEM Cited by 1 source
AWS SAM (Serverless Application Model)¶
What it is¶
AWS Serverless Application Model (SAM) is AWS's open-source
framework for defining + deploying serverless applications. A SAM
template is a superset of CloudFormation
— same YAML/JSON dialect, plus the AWS::Serverless::* transform that
expands shorthand resource types (AWS::Serverless::Function → Lambda
function + execution role + event source wiring) into CloudFormation
primitives. SAM is to serverless applications what systems/aws-cdk
is to general AWS infrastructure — both synthesise to CloudFormation,
with SAM specialised for the Lambda-centric shape.
Role in this wiki — local emulation for agentic development¶
SAM's load-bearing feature for this wiki is the SAM CLI's local emulation surface, named by the AWS Architecture Blog as the canonical AI-agent feedback-loop entry point for serverless apps:
"Serverless applications built with AWS Lambda and Amazon API Gateway can be emulated locally using the AWS Serverless Application Model (AWS SAM). With the
sam local start-apicommand, an AI agent can invoke Lambda functions through a locally emulated API Gateway, observe responses immediately, and iterate in seconds rather than minutes." (Source: sources/2026-03-26-aws-architecting-for-agentic-ai-development-on-aws)
This places SAM in the patterns/local-emulation-first tier — the cheapest tier that can falsify an AI-generated change to a Lambda handler. Iteration time collapses from "minutes to hours per change" (deploy to real Lambda + wait for pipeline + wait for logs) to "seconds per change" (local invocation against a Docker-based Lambda runtime + local API Gateway mock).
SAM CLI local-emulation surfaces¶
The AWS post names sam local start-api specifically; the broader
CLI also offers:
sam local invoke <Function>— single-function invocation with event payload (tests a handler in isolation).sam local start-api— locally-emulated API Gateway over Lambda functions; agent or browser hitshttp://localhost:3000/...and gets the real Lambda code path.sam local start-lambda— local Lambda service endpoint the AWS SDK can target for testing service-to-service invocation paths.sam build— build the deployment artefact (Python deps installed, Java jars assembled, container images built).sam deploy— deploy via CloudFormation.sam sync— incremental code updates against a deployed stack, faster than full redeploy (useful for the tier above local emulation).
Relation to concepts/agentic-development¶
Agentic development depends on feedback-loop speed, and feedback speed for serverless apps is gated by how long it takes to validate a Lambda change. SAM's local emulation compresses that validation from the deploy-and-wait tier to the subprocess tier. The AWS blog's escalating tier ladder for agentic feedback:
| Tier | Tool | Latency |
|---|---|---|
| 1. Local emulation (Lambda + API Gateway) | SAM sam local start-api |
seconds |
| 1. Local container run | docker run same image | seconds |
| 1. Local DynamoDB | systems/dynamodb-local | seconds |
| 1. Local Glue | Glue Docker images | seconds |
| 2. Hybrid cloud (no emulator) | patterns/hybrid-cloud-testing via CloudFormation/CDK small stacks | minutes |
| 3. Preview environment (full app) | patterns/ephemeral-preview-environments IaC stacks | minutes |
SAM sits at tier 1 — the default path the agent should try first.
Caveats¶
- Emulator drift.
sam localruns Lambda handlers in a Docker-based Lambda Runtime Interface Emulator; API Gateway is re-implemented client-side. Behaviour is close but not identical to production — throttling, timeout, cold-start, IAM evaluation, authoriser composition all differ. The AWS post doesn't discuss this drift; real projects eventually need hybrid tests + smoke tests against deployed envs to catch what the emulator silently passes. - Cold-start profile is not reproduced. Local invocation hot-loads the handler and skips the init-phase costs that dominate production tail latency.
- IAM evaluation is not simulated. Missing
iam:PassRoleor cross-service permissions only surface in deployed envs, which is why the AWS post mandates smoke tests "to surface configuration or permission issues that only appear at runtime."
Seen in¶
- sources/2026-03-26-aws-architecting-for-agentic-ai-development-on-aws
— AWS Architecture Blog names
sam local start-apias the canonical AI-agent feedback entry point for serverless apps; positions SAM as tier-1 of the agentic-development feedback-loop ladder; implicitly pair with DynamoDB Local + Glue Docker for the fully-local loop and with CloudFormation/CDK for the hybrid-and-preview tiers.