Amazon SageMaker AI now features an agentic experience that transforms model customization from a months-long process into a workflow completed in days or hours. Customers building an AI solution need to carefully frame their use case goals and success criteria, prepare data, choose the right models, configure, run, and analyze multiple experiments with various models and fine tuning techniques. Once a suitable model candidate that meets the success criteria is identified, they need to figure out the most cost performant way to deploy the model. Throughout this workflow customers need to manage the undifferentiated heavy lifting of setting up the infrastructure to train and deploy the models. The new capability now enables developers to use natural language interactions with coding agents to streamline the entire journey from use case definition to production deployment of a high quality model.

The agentic experience, based on SageMaker AI model customization agent skills, delivers expertise on fine-tuning applied to a builder’s specific use case, transformation to the required data formats, comprehensive quality evaluation using LLM-as-a-judge metrics, and flexible deployment options to Amazon Bedrock or SageMaker AI endpoints. Customers can install these skills in any IDE of their choice, such as Visual Studio and Cursor. Developers can work with multiple coding agents including Kiro, Claude Code, and CoPilot, in order to optimize popular model families like Amazon Nova, Llama, Qwen, and GPT-OSS. The experience generates reusable, editable code artifacts for transparency, reproducibility, and automation through integration into AIOps pipelines

Install SageMaker AI skills in your favorite IDE using the sagemaker-ai agent plugin. SageMaker AI model customization skills are also available and pre-installed in SageMaker Studio Notebooks, along with the Kiro coding agent. All you need to do is just sign up for Kiro subscription, open the chat window in Studio Notebooks and start chatting with the agent to build the workflow. The experience supports advanced customization techniques including supervised fine-tuning for instruction tuning, direct Preference Optimization for adjusting tone and preference selections, and Reinforcement Learning for use cases with verifiable correctness.

To learn more about model customization with the AI agent experience in Amazon SageMaker AI, visit the SageMaker model customization documentation.

US East (N. Virginia) — us-east-1

Europe (Ireland) — eu-west-1

US West (Oregon) — us-west-2

Asia Pacific (Tokyo) — ap-northeast-1