How Anthropic ships the #1 LLM in the World safely: Separating the Brain from the Hands

Katelyn Lesse and Angela Jiang join Latent Space to explain the engineering behind the Claude Platform: launching Opus 5.5, separating the model’s “brain” from the agent’s “hands,” scaling APIs, and building reliable managed agents. A new introduction covers Anthropic’s S-1, the improvement and investment loops, and why the platform matters. Watch the full episode on YouTube Show notes Claude Opus 5.5 — Anthropic’s launch notes on model behavior, evaluation and pricing. Scaling Managed Agents — Decoupling the brain, harness and sandbox. How Anthropic contains Claude — Isolation, credentials and network controls. Effective harnesses for long-running agents — Maintaining progress across context windows. Our Anthropic S-1 breakdown — Context for the newly recorded introduction. Timestamps 0:00 Anthropic’s S-1 and this interview1:06 Revenue per employee and efficiency3:13 The improvement and investment loops4:04 Why Claude Platform matters6:25 Training incidents and production safeguards7:30 Opus 5.5 launch and pricing9:25 Distillation and preserved thinking14:35 Product and engineering at Anthropic18:05 Scaling the Claude Platform19:25 Model access and safety decisions26:25 Trust grants and product roadmaps30:25 How Anthropic launches models35:55 Benchmarks and model personalization42:10 Claude Managed Agents44:25 Memory, dreaming, and agent harnesses48:25 Evaluating agents53:25 CPU constraints and scaling56:05 Rebuilding the Messages API in Rust1:03:25 The brain and the hands1:05:25 Protecting credentials and sandboxes1:08:25 Training incidents versus production1:10:55 Pacing and launching frontier models1:12:25 Safety classifiers and model routing1:19:25 Emerging agent workloads1:20:55 Fighting AI slop1:27:25 Management, writing, and working styles1:33:25 Building Anthropic’s DevRel team1:40:25 Developer relations and product voices1:43:25 What to try with Opus 5.5