Intro

We keep running into the same issue with AI coding agents when building and deploying on AWS. Agents can handle easy tasks well, like spinning up an Amazon S3 bucket or launching an EC2 instance, but the moment a task starts to get complex, say wiring a serverless API with least-privilege permissions or standing up a data pipeline end to end, they do weird things like choosing a service that isn't the right fit for the job, misconfiguring a resource or looping on retries against something they've clearly never seen.

We know this happens because the model is working from a training snapshot that's weeks to months old, and since then AWS has kept shipping new services and features the model has never seen. So if a task depends on a recently released service or feature, the agent doesn't accurately know how to achieve it.

Today, we've solved that problem with the Agent Toolkit for AWS . It gives your agent a live line to current AWS documentation, tested procedures and a secure way to build and deploy on AWS. It's free to use and you pay only for the AWS resources your agent spins up. Best of all, it supports the coding agents you already use, like Kiro, Claude Code, Cursor and Codex.