AWS Ships Six Open-Source Skills to Let Coding Agents Deploy Hugging Face Models on SageMaker

AWS ML Blog · rss · 2026-09-18

AWS ML Blog shows how six open-source Hugging Face Skills (Python + AWS CLI only) let coding agents like Kiro and Claude Code deploy HF models to production SageMaker endpoints with autoscaling, CloudWatch alarms, and correct Deep Learning Containers, across real-time, scale-to-zero, serverless, async, and batch modes.

Unguided-agent tests failed in two ways: deploying Qwen3-0.6B, agents picked a stale TGI container from training-data habits, failing health checks and burning GPU money before pivoting to vLLM; with a newly released diffusion MoE model, agents confirmed it existed but still wrote TGI scripts, failing silently.

The root cause is missing current deployment knowledge, not weak reasoning. The six skills — planner orchestration, AWS context discovery, Python env setup, IAM preflight, serving image selection, and production defaults — turn this fast-changing knowledge into editable files driving a six-phase deployment workflow.

Original post →

More from coding & agent

coding & agent channel →