HeadsUpAI

AWS Launches Agentic SageMaker Workflows to Automate End-to-End Model Customization

AWS launched agent-guided model customization in Amazon SageMaker AI, a natural language interface that uses AI agents to orchestrate the machine learning lifecycle. Users describe requirements in plain text, and the system triggers specialized agent skills (reusable, domain-specific capabilities) to handle task planning, data transformation, and model evaluation.
Tuning techniques
SFT, DPO, and RLVR
Agent skills
Planning, data transformation, and evaluation
Supported IDEs
Kiro, Claude Code, Cursor, and more
Interface
Natural language
Timeline reduction
Months to days

This update moves model adaptation from manual MLOps into the realm of agentic engineering. By automating complex techniques like DPO (Direct Preference Optimization) and RLVR, AWS is lowering the technical barrier for high-performance fine-tuning. It builds on AWS's managed agent harness to accelerate enterprise deployment cycles.

You can now use these agentic workflows to customize models within Kiro or external IDEs like Claude Code and Cursor. The system provides expert advice on model selection, mirroring AWS's AgentCore Optimization preview for refining agent reliability. These capabilities are available now within the SageMaker AI console.

View the full update on aws.amazon.com
AWS AI
AWS AI
@AWSAI
X

Agent-guided model customization is now available in Amazon SageMaker AI. Describe your use case in natural language, and agent skills guide you from data transformation to deployment. Model customization that used to take months now takes days.

4likes
View on X

Still wondering? A few quick answers below.

Agent-guided model customization is a new capability in Amazon SageMaker AI that uses AI agents to manage the end-to-end process of fine-tuning foundation models. Instead of manually configuring every step, users describe their use case in natural language. The system then uses specialized agent skills to build specifications, transform data, and handle model deployment automatically.

The workflow begins when a user describes their specific requirements in natural language. An AI agent then orchestrates the entire lifecycle, from initial planning and data transformation to fine-tuning and final evaluation. The agent provides expert advice on model selection and hyperparameter configuration, generating the necessary code to match an organization's specific standards and deployment needs.

The agent-guided system in Amazon SageMaker AI supports several advanced model customization techniques. These include Supervised Fine-Tuning, Direct Preference Optimization, and RLVR. By automating these complex methods, the system allows developers to apply sophisticated alignment and optimization strategies to their models without needing deep manual expertise in hyperparameter configuration or orchestration.

Users can access these agentic workflows through multiple environments. It is natively available via Kiro in SageMaker Studio JupyterLab. Additionally, the system is designed to be flexible, allowing developers to bring their own agents and integrated development environments, including popular tools such as Claude Code, Cursor, and VS Code, to manage the customization process.

The introduction of agent-guided workflows significantly accelerates the model development lifecycle. According to AWS, model customization tasks that previously required months of manual effort from specialized machine learning engineers can now be completed in just a few days. This speed is achieved by using AI agents to automate data preparation, tuning, and evaluation steps.

Share this update