How would you pass data between Lambda functions in Step Functions?

Why payload design matters: Step Functions passes output of each state as input to the next via JSON. The 256 KB payload limit forces architectural choices—large payloads break. Mechanisms: ResultPath merges Lambda output into the input; ResultSelector reshapes output; Parameters inject values (e.g., {"userId.$": "$.userId"} passes through). For large data: write to S3, pass S3 URI in state; next Lambda reads from S3. Scalability: At high throughput, 256 KB means you cannot pass 10K records—reference storage. Cost: Step Functions charges by state transitions; keeping payloads small reduces memory and execution time in Lambda. Canonical envelope: Use traceId, executionId for observability; standardize error shape for retries. Avoid binary data; avoid passing full datasets—reference by path. In production, we use a thin envelope: {"traceId": "...", "dataRef": "s3://bucket/key"}.