AWS AI Updates: May 5, 2026

1. SageMaker AI Adds an Agent Experience for Model Customization

AWS. SageMaker AI now ships a set of agentic skills that wrap the model-customization pipeline — use-case definition, fine-tuning, data transformation, LLM-as-a-judge evaluation, and deployment — so developers can drive the workflow from natural language inside an IDE. The skills run inside Visual Studio, Cursor, SageMaker Studio Notebooks, and against coding agents including Kiro, Claude Code, and Copilot, and target Amazon Nova, Llama, Qwen, and GPT-OSS model families. They cover supervised fine-tuning, Direct Preference Optimization, and reinforcement-learning recipes, and emit reusable code artifacts that can be checked into AIOps pipelines and deployed to either Bedrock or SageMaker endpoints. Source

2. Amazon Quick Adds Dataset Q&A With a Text-to-SQL Agent

AWS. Dataset Q&A lets analysts pose natural-language questions against governed enterprise data and have Quick’s text-to-SQL agent identify the right tables, generate SQL, and return answers across Redshift, Athena, Aurora PostgreSQL, and S3 table buckets. An Explain capability surfaces the generated query before execution so users (and reviewers) can validate the logic, and existing row- and column-level governance still applies, so Q&A respects whatever access controls the dataset already enforces. Source

3. Amazon Quick Generates Whole Dashboards from a Prompt

AWS. Generate Analysis turns a natural-language description plus up to three datasets into an editable plan and then a fully laid-out sheet — visuals, filter controls, and calculated fields included — collapsing what AWS positions as hours of dashboard assembly into minutes. The feature is gated to Enterprise subscriptions and Author Pro users, and the plan-then-generate flow lets authors review and adjust the proposed structure before Quick commits to layout. Source