AWS Makes Its AI/ML Skill a Core Part of the Agent Toolkit, Taking Coding Agents From Inference Tuning to Full SageMaker Workflows

Amazon Web Services said on 13 October 2026 that its AI/ML skill is now a core skill in the Agent Toolkit for AWS [1]. According to AWS, a coding agent with the skill can select, train, optimise and deploy models on Amazon SageMaker AI, its machine learning service, “in a single natural-language workflow” while also working with the rest of a customer’s AWS setup [1].

What an agent “skill” is

Coding agents such as Kiro, Claude Code and Codex are AI assistants that write and run code when you describe a task in plain language. They are general-purpose, so they do not automatically know the details of every cloud service.

A skill fills that gap. AWS’s product page describes skills as “curated packages of instructions, code scripts, and reference materials” that help agents complete specific tasks on AWS [2]. Think of it as a specialist manual plus ready-made tools that the agent can pick up when needed. AWS says skills are loaded on demand, so an agent only pulls in the ones relevant to the job in hand [2].

Not new, but bigger

The skill itself is not brand new. AWS’s Machine Learning Blog introduced it, under the name aws-ai-ml, on 5 October 2026 [3]. That first version focused on one job: helping agents benchmark SageMaker endpoints and pick cheaper, faster ways to run AI models (known as inference optimisation) [3].

The 13 October announcement changes three things, according to AWS [1]:

• Core status. The skill now sits alongside the toolkit’s other AWS skills, such as those for EC2 and Bedrock.
• More of the machine learning process. On top of inference optimisation, AWS lists guided fine-tuning “from use case definition through training, evaluation and deployment”; training for both traditional machine learning and generative AI; MLflow for tracking experiments and managing models; deploying ready-made models from SageMaker JumpStart, with recommendations on which model suits a task; and processing jobs for preparing data.
• One supported route. AWS says the Agent Toolkit is now “the single, supported integration point”, replacing the earlier AWS Labs skills and plugins with one way to install and update.

How to get it, and what it costs

AWS says the skill is pre-installed in SageMaker Studio JupyterLab, for use with Kiro or another coding agent, and can also be installed from the Agent Toolkit for AWS on GitHub [1]. The product page says the wider toolkit sets up with one command, “aws configure agent-toolkit”, which needs version 2.35 or later of the AWS command-line tool [2].

There is no extra charge for the skill. AWS says customers pay only for the AWS resources their agents use [1], which for training and hosting models can still add up.

What this means

• Agents are moving up the stack. Instead of only writing code, agents are being given the know-how to run whole model pipelines on a cloud platform.
• AWS is tidying its offering. Folding scattered AWS Labs add-ons into one toolkit gives developers a single, officially supported place to get these skills.
• The bill still matters. The skill is free, but the training jobs and endpoints an agent starts are not, so teams will want spending limits in place.

Sources

1. AWS What’s New, “Bring deep SageMaker AI expertise to your coding agent with the AWS AI/ML skill in the Agent Toolkit for AWS”, 13 October 2026. https://aws.amazon.com/about-aws/whats-new/2026/10/bring-sagemaker-ai-expertise-to-your-coding-agent/
2. AWS, “Agent Toolkit for AWS” product page. https://aws.amazon.com/products/developer-tools/agent-toolkit-for-aws/
3. AWS Machine Learning Blog, “New agent skill: Amazon SageMaker optimized generative AI inference for your coding agent”, 5 October 2026. https://aws.amazon.com/blogs/machine-learning/new-agent-skill-amazon-sagemaker-optimized-generative-ai-inference-for-your-coding-agent/

Related stories

• Do AI Agents Need Elaborate Harnesses? A Minimal-Agent Study and Two New Toolkits
• Fences for AI Agents: Microsoft’s Execution Containers and AWS’s Strands Box
• MCP Explained: The USB-C for AI Tools (Model Context Protocol Guide)
• Google’s Agent Development Kit Now Covers Five Languages: Python, Go, TypeScript, Java and Kotlin