Claude Managed Agents Q&A: Evals, CI/CD and Multi-Tenancy (Part 4 of 4) — Anthropic
Anthropic's managed agent service runs 10,000 evaluation sessions as new models ship to update tool assumptions.
“one of the things we want to ensure as a managed service is to update our assumptions about tools as early as possible to remove what is no longer needed”
This Q&A covers operational details of Anthropic's Claude managed agents platform, including a Terraform-style YAML CLI (NCLI) for CI/CD, message queuing semantics during tool execution (400 errors block out-of-order sends), and internal evaluation practices. The most noteworthy signal is that Anthropic runs large-scale evals (10,000 sessions) against new models before public release to proactively update managed service tooling. Useful for teams building production agent infrastructure but lacks a major new announcement.