Claude Code on Amazon Bedrock
Built the Anthropic on AWS specialisation from scratch; 28 fixes to the Claude Code on Bedrock guidance unblocked a 3,000-developer rollout.
What I Learned
Getting a coding agent to work is only the start of a company-wide rollout. People need to sign in with their existing accounts, and the organisation needs to know who is spending what and where to set limits. Those operational details determine whether the tool can reach thousands of developers.
The Specialisation
I built the Anthropic on AWS specialisation from scratch. It’s now one of the most in-demand technical depths in EMEA, covering everything from running Claude models on Amazon Bedrock to rolling out Claude Code across engineering organisations.
The Guidance
Getting Claude Code into a large company isn’t just CLAUDE_CODE_USE_BEDROCK=1. Enterprises need single sign-on, per-user cost attribution, usage monitoring and quotas before they hand an AI coding agent to thousands of engineers.
The open-source Guidance for Claude Code with Amazon Bedrock covers exactly that:
- Authentication: developers sign in with their existing identity provider (Okta, Entra ID, Auth0, Google, Cognito and others), and the OIDC token is exchanged with AWS STS for short-lived, Bedrock-scoped credentials
- Monitoring: OpenTelemetry telemetry with per-user identity, landing in CloudWatch dashboards
- Quotas: per-user limits enforced before credentials are issued
- Analytics and distribution: an S3 + Athena data lake, and packaged binaries for Windows, macOS and Linux
My Contribution
I contributed 28 fixes to the guidance. Together they unblocked a 3,000-developer Claude Code rollout for a customer.