Anthropic Ships Terraform-Style Deployments for Claude

Anthropic has rolled out a new infrastructure-as-code workflow for its Managed Agents platform, introducing the ant apply command to the ant CLI. This update allows developers to define Claude agents, environments, skills, memory stores, and scheduled deployments as local Markdown, YAML, or JSON files. Instead of manually configuring resources in the developer console or writing custom scripts to chain API calls, teams can now manage their entire agent architecture as code.

The system relies on a claude-lock.json file to track resource IDs and cryptographic hashes, which prevents duplicate resources and enables precise drift detection. When developers reference other resources using relative file paths, the ant apply command automatically resolves the dependency graph and deploys them in the correct order. For external capabilities, skills can be pulled directly from GitHub URLs and pinned to specific commits, with an optional --upgrade flag available to fetch the latest versions.

This declarative approach integrates directly into continuous integration workflows. Developers can run ant apply with the --dry-run flag on pull requests to preview changes, and execute the command with the --yes flag to automatically deploy updates after a merge. Security is handled via Workload Identity Federation rather than static API keys. If external changes cause drift, the CLI will halt unless overridden with the --force flag, while the --prune flag can be used to clean up and archive removed resources.

For existing setups, the tool cannot directly adopt resources built via the console or the older ant beta:agents create command. However, developers can use the console's export option to generate the necessary configuration and lockfile. By treating the codebase as the single source of truth, this update eliminates the need for custom orchestration glue code, bringing standard software engineering practices like pull requests and git-based rollbacks to AI agent development.