Amazon SageMaker AI introduces agent-guided workflows for model customization using nine modular skills across the full lifecycle.

Amazon SageMaker AI now offers an agentic experience where developers describe their use case in natural language, and an AI coding agent (such as Kiro, Claude Code, or Cursor) streamlines the entire customization journey from use case definition and data preparation through technique selection, evaluation, and deployment.

The solution includes nine modular skills—Use Case Specification, Planning, Fine-tuning Setup, Dataset Evaluation, Dataset Transformation, Fine-tuning, Model Evaluation, and Model Deployment—that encode AWS and data science expertise. Skills are fully customizable and can be modified to match team workflows, governance standards, and tooling preferences. Three fine-tuning techniques are supported: SFT (Supervised Fine-Tuning) for task-specific behavior, DPO (Direct Preference Optimization) for aligning tone and style, and RLVR (Reinforcement Learning with Verifiable Rewards) for programmatically verifiable tasks.

All generated code is fully editable and produces reusable artifacts that integrate seamlessly into existing workflows. Agent skills for model customization decrease token usage and boost productivity by automating rigorous evaluation, hyperparameter configuration, and deployment pathway selection to Amazon Bedrock or SageMaker AI endpoints.