Roadmap

Future features and enhancements considered during development but deferred for later versions.

Table of Contents

v1.0.0

  • 3 tiers (starter, professional, enterprise)

  • Generic use-case parameterization

  • CodeCommit + S3 source control options

  • 12 CLI actions

  • SSM Parameter Store outputs for downstream provisioners

v1.1.0

SageMaker Provisioner Integration

  • Direct integration with SageMaker Provisioner via SSM

  • SageMaker Studio domain configuration outputs

  • Lifecycle configuration references

  • AWS::SageMaker::Project provisioning — deferred from ML Provisioner because it requires either Service Catalog (rejected, see Decision 7 in APPLICATION_ARCHITECTURE) or an S3-hosted CloudFormation template. The SageMaker Provisioner has the S3 artifacts bucket available (via SSM from ML Provisioner) and is the natural home for Studio governance resources

Cross-Account Model Sharing (Enterprise Tier)

  • Add optional ModelPackageGroupPolicy property to AWS::SageMaker::ModelPackageGroup

  • Allows sharing models across AWS accounts via resource-based policy

  • Would require new config fields: cross_account_sharing.enabled, cross_account_sharing.account_ids

  • IAM JSON policy document constructed by CfnGenerator and attached to the Model Package Group

  • Relevant AWS doc: ModelPackageGroupPolicy on AWS::SageMaker::ModelPackageGroup

Shared CodeCommit Repositories

  • Add optional repository_override config field to reference existing CodeCommit repositories instead of creating new ones

  • Allows multiple environment stacks (dev, staging, prod) to share the same repositories with branching strategies

  • Similar pattern to s3_prefix which allows bringing an existing S3 bucket as source control

v1.1.1

Configuration Updates via Stack Update (apply-changes)

  • New apply-changes action for minor safe config changes on a deployed stack, avoiding delete-and-redeploy for low-risk updates

  • Supports safe changes only: tags, alerts_email, log_retention_days

  • Creates a CloudFormation changeset, inspects it for resource replacement (Replacement: true), and aborts if any destructive changes are detected

  • Immutable fields (ml_name, tier, region) blocked at validation before changeset is created

  • SSM parameters update automatically as CFN resolves new resource ARNs/IDs

  • Complements the existing show-changes preview action

  • Note: full lifecycle management of live stacks remains outside the provisioner’s design scope

Prerequisites before implementation:

  • DeletionPolicy: Retain on S3 artifacts bucket, SageMaker Model Registry, and KMS key to prevent data loss on resource replacement

  • ml_name immutability enforcement — detect and block config changes that would alter ml_name on an existing stack

  • Replacement detection — parse ChangeSet and warn/block on any Replacement: True resource

  • KMS key change detection — block changes that would replace the KMS key

  • SSM path migration — if ml_name change is ever allowed, old SSM params must be cleaned up and consuming services notified

v1.2.0

Enhanced CloudWatch Dashboard (Enterprise Tier)

  • Replace placeholder text widget with pre-built ML-specific metric widgets

  • Widgets to include: SageMaker model approval rate, CodePipeline execution status, CodeBuild build success/failure rate, CloudWatch Alarm state

  • Widget metric dimensions reference physical resource names from the same stack

  • Enterprise tier only — differentiator justifying premium pricing

Cost Estimation

  • Add cost-traffic, cost-estimate, cost-refresh-prices actions following the same pattern as vpc-provisioner and s3-provisioner

  • Fixed cost resources: CodePipeline pipelines, KMS keys, CloudWatch Dashboards, CloudWatch Alarms, VPC Interface endpoints

  • Usage-based resources: CodeBuild build minutes, S3 storage/requests, CloudWatch Logs ingestion/storage, SNS notifications

  • Traffic assumptions file pattern reused from existing provisioners

v2.0.0 — Phase 3 ML Solutions

  • Use-case-specific templates (churn, fraud, demand, sentiment)

  • Solution-specific notebook templates

  • Pre-built feature engineering pipelines

  • Model monitoring configurations per use case