Anthropic staff describe Claude Opus 5.5 and managed-agent controls

In a Latent Space interview, Anthropic staff describe Opus 5.5 and managed-agent isolation as a platform discussion, not a new release.

Katelyn Lesse and Angela Jiang told Latent Space how Anthropic ships the Claude Platform, including the Opus 5.5 launch, pricing, and evaluation. They describe separating the model’s “brain” from the agent’s “hands,” with the session and harness kept apart from a sandbox that executes code so a failed sandbox does not end the session and service credentials need not sit with generated code. Opus 5.5 is described as cheaper to run than its predecessor, and thinking blocks are kept across turns to hinder distillation. Near-term attention has shifted toward safety; later platform ideas should be treated as plans rather than shipped features. The item is an interview, not a standalone release.