How can developers more easily optimize and deploy AI models for production? New agent skills and coding tools are emerging to automate complex tasks, from benchmarking endpoints to ensuring compliance. The "aws-ai-ml" skill for coding agents, available through the Agent Toolkit for AWS, and Anthropic's Claude Code, powered by models on Amazon Bedrock, are designed to streamline the process of moving AI models from development to scalable, performant, and compliant production environments.
SageMaker AI's New Inference Optimization Skill
The "aws-ai-ml" skill for Amazon SageMaker AI's optimized generative AI inference introduces deep expertise in inference optimization and benchmarking for coding agents. This skill can be integrated with agents like Kiro, Claude Code, and Codex that support the Model Context Protocol (MCP). It enables agents to benchmark endpoints, recommend deployment configurations, compare performance runs, and generate executable SageMaker Python SDK v3 code. This aims to help engineers overcome the challenge of selecting appropriate instance families or serving containers for their AI models by allowing them to express their performance or cost goals in natural language.
Setting up the "aws-ai-ml" skill involves installing the Agent Toolkit for AWS and then adding the skill. Developers can use it with any MCP-compatible coding agent on their local machine or within an Amazon SageMaker Studio JupyterLab space. Once installed, the agent gains the ability to understand and act on natural language prompts related to SageMaker AI inference optimization.
Practical Applications of the Skill
One key capability is benchmarking existing endpoints. If a model is already deployed on a SageMaker AI endpoint, developers can ask the agent to benchmark it. The agent generates a Python notebook to run a load test, providing quantitative performance reports on throughput (requests per second, output tokens per second), latency (p50, p99, time-to-first-token, inter-token latency), and concurrency. These are measured values from real load, not estimates. The agent also suggests improvement mechanisms like prefill decoding to boost performance.
Another crucial application is identifying the correct instance type for a model. Developers can provide their model's S3 URI and their optimization goal, and the agent assists in finding the most suitable instance type for deployment on SageMaker AI. This eliminates the need for engineers to manually navigate the broad array of serverful hosting options, including real-time, batch, and asynchronous modes, on-demand or reserved capacity, and heterogeneous instances offered by SageMaker AI.
Claude Code for Regulated Workloads on Amazon Bedrock
For workloads with stringent regulatory or compliance requirements, such as those under International Traffic in Arms Regulations (ITAR), Anthropic's Claude Code now supports Amazon Bedrock in AWS GovCloud (US) Regions. This enables AI-assisted development workflows that maintain compliance while speeding up everyday tasks. Claude Opus 5.5 and Claude Sonnet 5.5 models are available on Amazon Bedrock in these regions, with certifications like FedRAMP Class D and DoD Impact Level 4 and 5 (IL4/IL5) for Claude Sonnet 5, helping government agencies meet their compliance needs.
Amazon Bedrock in AWS GovCloud (US) offers security-focused features, including built-in data protection that ensures customer content is not stored, logged, or used for training AWS models or shared with third parties. It also integrates with existing security controls and compliance frameworks. The platform supports two endpoint surfaces, bedrock-runtime and bedrock-mantle, both powered by the underlying Mantle inference engine with Zero Operator Access (ZOA) architecture.
Claude Code acts as an agentic coding tool that can read and edit codebases, run commands, and integrate with development tools directly within a developer's terminal or IDE. It can perform tasks like writing and fixing multi-file code, answering architectural questions, executing and fixing tests, searching Git history, and resolving merge conflicts. Furthermore, Claude Code can connect to external tools and data sources via the Model Context Protocol (MCP) and allows for customization through memory files, skills, and hooks.
Setting Up Claude Code with Amazon Bedrock
To use Claude Code with Amazon Bedrock in AWS GovCloud (US), developers need an AWS GovCloud (US) account with Bedrock access, appropriate AWS Identity and Access Management (IAM) permissions for Bedrock APIs (such as bedrock:InvokeModel for bedrock-runtime or bedrock-mantle:CreateInference for bedrock-mantle), and model access enabled for Claude Opus 5.5, Claude Sonnet 5.5, and Claude Sonnet 5. The AWS CLI must also be configured with valid credentials. Installation of Claude Code can be done via various methods, including curl commands for macOS/Linux/WSL, PowerShell for Windows, or Homebrew.
After installation, developers can configure Claude Code either through an interactive setup wizard, which guides them through selecting a 3rd-party platform (Amazon Bedrock), authentication method, region (us-gov-west-1), and model pinning, or by manually setting environment variables like CLAUDE_CODE_USE_BEDROCK, AWS_REGION, and ANTHROPIC_MODEL. This flexibility allows for both interactive and scripted deployments, catering to different enterprise needs.
Driving AI Innovation with Agentic Tools
These advancements highlight a broader trend towards agentic AI companions and tools designed to enhance developer productivity and ensure compliance. Whether it's optimizing inference on SageMaker AI or enabling secure AI-assisted coding for regulated environments with Claude Code on Amazon Bedrock, the goal is to make AI model deployment more efficient, auditable, and accessible. This shift allows developers to focus on their core intent rather than the underlying infrastructure complexities, making AI more practical for a wider range of enterprise applications.